The One-Page Decision Brief: What the Research Says About Writing Something an Executive Can Decide From.

5 hours ago

47 min read

The request arrives at ten past four on a Thursday. Can you put the supplier decision on one page for me, before the exec meeting tomorrow. You have four days of work behind you: three quotes, a risk register, two conversations with legal, a spreadsheet with a tab nobody else will ever open. What you have in front of you is one side of A4 and about ninety seconds of a very tired person's attention, and all four days of it have to survive the journey through that gap.

Most of us were taught to write the paper and hope. The research has rather more to say than that: on how people read a document they have to decide from, on what the shape and order of the options does to the answer, on whether a recommendation helps or hinders, on how many options a page can carry, and on what happens six months later when somebody asks why on earth we did that. It does not all point the same way, and the places where it disagrees are some of the most useful parts. This post is that evidence, turned into a page you can actually write.

One side of A4, and about ninety seconds of somebody's attention. Everything depends on what goes in the first four lines.

What you will get out of this

What 98.6% of readers do first with a document, from a 2024 survey of people who read for a living, and what that means for where your recommendation goes
The registered trial of 2,305 people that compared the same facts in a table and in prose, and the size of the difference
Why the same three options can produce two different answers depending on how they sit on the page, from the research on joint and separate evaluation
What happened when 402 doctors were given one recommended alternative or two, and why the number two matters more than any number after it
The meta-analysis of fifty experiments that found no choice overload effect, the meta-analysis of ninety-nine observations that found one, and how to read the disagreement
What a probability word actually means to the person reading it, the one formatting change that doubled agreement, and why the range you write is probably too narrow
The pre-mortem evidence, and what a paper looks like when the objection is written into it rather than waited for
The brief check: a tool that reads your decision and tells you what shape the page should take
The Decision Brief Pages: the one-page template, the option table and the decision record

What the first thirty seconds decide

Start with how people actually read something they have to act on, because almost everything else follows from it.

In 2024 a team surveyed science and health researchers about the order in which they read papers. Of the whole sample, 98.6% read the abstract first, and the reason they gave was that it lets them decide whether to read on. Seventy-five per cent found it the easiest section and 62.4% rated it as very important. Most of them did not read the paper in the order it was written in, and what predicted whether a section got read at all was how important and how easy they expected it to be. These are people whose profession is reading, behaving exactly like your executive at half past nine at night.

So the summary at the top of the paper turns out to be the paper itself, for most readers on most days, and everything below it stands as evidence for anyone who wants to go looking.

The next question is what that summary should look like. A registered, pre-planned trial put 2,305 UK residents in front of the same information about the benefits and harms of a medical option, as flowing text or as a small table, and measured comprehension straight away and recall six weeks later. The table version won on comprehension with an effect size of 0.39 and held a small edge on recall six weeks on. Readers found it more engaging, and there was no difference in how informed or how trusting they felt. The table simplified nothing; it was the same content, arranged so that the comparison was visible at a glance.

That result sits inside a wider pattern. A systematic review of thirty-one experiments on presenting performance information found the same thing from every angle: people understand more and choose better when the display is less complex, the language avoids technical terms, the number of choices is reduced and the results sit in a familiar context. Three studies on hospital quality reports reached the conclusion the researchers summed up as less is more, and the effect was strongest for the readers least comfortable with numbers. An honest counterweight belongs beside that: a systematic review of six randomised trials of evidence summaries written for policy-makers found them easier to understand than the full reviews, with little sign that they changed how much the evidence was actually used.

Then there is the writing itself, and here the evidence is bracing. In five experiments published in 2006, texts were made more complex by swapping in longer words, and readers judged the authors to be less intelligent. The effect held regardless of the quality of the original essay and of what readers expected going in, and it was carried by how easy the text felt to process, which the researcher demonstrated by putting the same words in a hard-to-read font and getting the same drop. A separate study redrafted a legal contract in stages and found that simplifying words and sentence structure reliably improved comprehension, with a footnote worth keeping: absolute comprehension stayed very low, because plain wording will only carry an idea so far when the idea itself is hard.

Headings earn their place too. Nine experiments found that they helped recall, search and retrieval from text, and that it made no difference whether they sat in the margin or in the body. The wider research on signalling, meaning any device that marks what matters in a document, has been meta-analysed more than once and supports the same habit.

One last finding, for anyone writing to be read on a phone. A meta-analysis of fifty-four studies and more than 171,000 participants found a small but consistent advantage for paper over screen, growing when readers were under time pressure and when the text was informational rather than a story. Your executive is reading an informational text under time pressure on a screen, which is the exact combination the research says is hardest.

A graphic titled How a one-page brief is actually read, with the eyebrow From surveys of professional readers and a registered trial of 2,305 people. Standfirst: Nobody reads a decision paper from top to bottom. They read the first block, decide whether to carry on, and then hunt for the part that answers their question. Here is the order the evidence describes, and what each stage asks of the person writing. Five numbered rows, each carrying a figure in a small chip on the right. One, The first block, and a decision about whether to read on, 98.6%: In a 2024 survey of people who read for a living, 98.6% read the summary first, and said they did it to decide whether the rest was worth their time. Put the decision, the recommendation, the reason and the deadline here. Two, A scan for the headings that look relevant, 9 experiments: Across nine experiments, headings helped readers recall, search and retrieve, and it made no difference whether they sat in the margin or in the text. Name each heading for what the reader gets from it. Three, The comparison, if there is one to look at, 2,305 people: The same facts in a small table beat the same facts in prose on comprehension, and still held an edge on recall six weeks later. One table, identical rows for every option. Four, A judgement about the writer, made unconsciously, 5 experiments: Longer words made authors look less intelligent in five experiments, and the effect ran through how easy the text felt to read. Short words carry rigour rather than lacking it. Five, Everything else, only if something does not add up, 54 studies: Paper beat screen by a small but consistent margin across 54 studies, and the gap widened under time pressure and for informational texts. Your reader is on a phone, in a hurry, reading the hardest kind of text. Foot: Shiely and colleagues (2024); Brick and colleagues (2020); Hartley and Trueman (1983); Oppenheimer (2006). — Nobody reads your paper in the order you wrote it, which is a little wounding and extremely useful to know.

The summary is not the introduction to the paper. For most readers, on most days, the summary is the paper.

Try this. Write the first four lines last, and write them as though they were the only four lines that will be read: the decision, the recommendation, the reason, and the date by which it must be made. Then read the rest of the page and ask what is in it that a reader could not reconstruct from those four lines. That is the part worth keeping.

The research behind this section

Shiely, Gallagher and Millar (2024) surveyed science and health academics and found that 98.6% read the abstract first because it provides an overview and allows a decision about whether to read on, that 75% perceived it as the easiest section and 62.4% as very important, that most respondents did not read papers in the order in which they are written, and that perceived difficulty and perceived importance influenced reading order. Brick, McDowell and Freeman (2020) ran a pre-registered longitudinal randomised trial with 2,305 census-matched UK residents and found that fact boxes, meaning simple tabular messages, led to more comprehension (d = 0.39) and slightly more knowledge recall after six weeks (d = 0.12) than the same information in text, consistently across topics and across all levels of numeracy and education, with fact boxes rated more engaging and no differences in decisions, feeling informed or trust. Kurtzman and Greene (2016) systematically reviewed thirty-one experimental articles on presenting health care performance information and found that consumers understood more and made more informed choices when the display was less complex, achieved by reducing the quantity of choices, displaying results in a positive direction, using non-technical language and evaluative elements, and situating results in common contexts. Peters and colleagues (2007) reported three studies supporting the idea that less is more when presenting comparative performance information, with the strongest results for people lower in numeracy, who had higher comprehension and made better choices when the format eased cognitive burden and highlighted the meaning of the important information. Oppenheimer (2006) ran five experiments manipulating the complexity of texts and found a negative relationship between complexity and judged intelligence, holding regardless of the quality of the original essay and of prior expectations, mediated by processing fluency, with texts in hard-to-read fonts judged to come from less intelligent authors. Masson and Waldron (1994) redrafted legal contracts in three stages and found comprehension reliably enhanced by simplified words and sentence structure, while absolute levels of comprehension remained very low. Hartley and Trueman (1983) summarised nine experiments and found that headings aided recall, search and retrieval, and that marginal and embedded positions made no difference. Lorch and Lorch (1996) examined the effects of headings on text recall and summarisation. Alpizar, Adesope and Wong (2020) and Richter, Scheiter and Eitel (2016) each meta-analysed the signalling principle in multimedia learning. Hartley (1999) rewrote three journal abstracts in structured form and found them longer, more informative and judged clearer by readers. Hartley (2000) compared thirty traditional abstracts with their structured versions and found the structured ones more informative and no less accurate. Hartley, Sydes and Blurton (1996) tested whether structured abstracts let readers obtain information more accurately and quickly. Petkovic and colleagues (2018) systematically reviewed six randomised controlled studies of evidence summaries for policy-makers and found that summaries were likely to be easier to understand than complete systematic reviews while their effect on the use of that evidence in policymaking remained unclear, with little to no difference in reported use, knowledge, understanding or perceived usefulness. Kerwer and colleagues (2021) found in a preregistered within-person study of 166 people that plain language summaries were more comprehensible than scientific abstracts, that laypeople understood the information more correctly from them, and that they were perceived as more credible with greater confidence in deciding from them. Delgado and colleagues (2018) meta-analysed fifty-four studies involving 171,055 participants and found an advantage of paper over digital reading (Hedge's g = -0.21) that increased under time-constrained reading and held for informational texts. Clinton (2019) systematically reviewed and meta-analysed thirty-three experiments on reading from paper compared with screens. Eppler and Mengis (2004) reviewed thirty years of research on information overload across organisation science, marketing, accounting and management information systems. El-Sayed and colleagues (2021) ran three experiments and found that more readable disclosures improved judgements of future performance, and that readers were prone to a recency effect from the ordering of information when narratives were complex, with no primacy effect when they were simple.

How you lay the options out is part of the decision

Here is the finding that changes how you think about a comparison table, and it is one of the most robust in the field.

People evaluate the same option differently depending on whether they see it on its own or beside its alternatives. The researchers call these separate evaluation and joint evaluation, and the review that named the pattern explains the divergence. Some attributes are easy to judge in isolation, such as price. Others are almost impossible to judge without a comparison, such as whether a forty-page implementation plan is thorough or bloated. In separate evaluation the easy ones dominate, because they are the only ones the reader can get a grip on. Put the options side by side and the hard ones come into focus, because now they can be compared.

Put plainly: if you send the supplier options as three sections one after another, the decision will turn on price, because price is the only thing a reader can evaluate without the comparison in front of them. Put the three in one table with the same rows, and implementation time, exit terms and support cover become decidable.

A set of seven experiments published in 2017 tested exactly this and found that people were more likely to choose the dominating option, meaning the one that was better on every dimension, when options were presented simultaneously rather than one at a time. That held whether or not the dominance was obvious, and the mechanism was depth of processing rather than memory load. A later study of decision quality across choice architectures found joint presentation winning again.

Then there is framing, a word that covers three different things, and a 1998 typology is still the cleanest way to keep them apart: risky choice framing, where the same gamble is described in terms of lives saved or lives lost; attribute framing, where a single feature is described positively or negatively; and goal framing, where a message emphasises what you gain by acting or what you lose by not acting. The effects are real and smaller than the famous examples suggest. A meta-analysis of 136 papers, nearly 30,000 participants and 230 effect sizes put the overall framing effect at small to moderate, with the size depending heavily on how the experiment was designed, and a more recent meta-analysis has revisited the question. As an honest counterweight, when researchers gave 159 people two classic measures of the framing effect, the two did not correlate, which raises a real question about whether framing is one stable thing at all. Small to moderate is still worth knowing when you are the one choosing the wording. Ninety-four per cent uptime and six per cent downtime are the same fact, and they do not land the same way.

The other lever on the page is the default. A meta-analysis of fifty-eight studies and 73,675 people found a considerable effect of pre-selected options, d = 0.68, with substantial variation: most studies found positive effects, several found none, and two found the effect reversed. Defaults worked most strongly when they were read as endorsement, meaning the reader took them as what the person who built the page thought they should do, or as endowment, meaning the reader took them as the status quo. A later pair of experiments found that people assume a default represents a recommendation, and that they assume it reflects the most popular option whether or not they are told anything of the sort. Two preregistered replications of the classic status quo experiments supported the bias in three of four scenarios.

The practical reading of all that is uncomfortable and useful. Whichever option you put first, bold, or in the row marked current approach is doing work you may not have intended. The honest response is not to pretend otherwise; it is to say which one you are recommending and why, so that the reader is weighing an argument rather than absorbing a layout.

One tempting trick deserves a warning. The attraction effect, where adding a clearly inferior third option makes the option it resembles look better, has been reproduced dozens of times. In 2014 a set of studies found that it mostly appears when every attribute is reduced to a number on a page, and largely vanishes when people experience the thing or when even one attribute is shown perceptually. Research on decoys inside recommendation systems found that in a personalised context a decoy reduced demand for the option it was meant to flatter and pushed people to walk away. A straw option in a brief mainly makes the reader wonder what else on the page has been arranged for them.

A graphic titled Same facts, different page, different answer, with the eyebrow What the research on framing, defaults and joint evaluation says about the shape of a page. Standfirst: Nothing here is about changing the evidence. Every line on the left is a perfectly honest way to lay out a brief, and each one quietly pushes the reader somewhere. The habit is on the left, and what the evidence says it does is on the right. Five rows, the way of laying the page out on the left in a dark panel and the evidence on the right in a pale one, an arrow between them. Three options, written up one after another, a section each: The decision turns on price. Read in isolation, only the easy-to-judge attributes can be weighed at all. Implementation time and exit terms need something to be compared against before they mean anything. The same three options in one table with identical rows: The hard-to-judge things become decidable. Across seven experiments, people chose the better option more often when the options sat side by side, because they processed them more deeply rather than because they remembered more. One option pre-ticked, or listed first as the current one: It reads as your recommendation whether you meant it or not. Across 58 studies and 73,675 people, pre-selected options moved decisions, and moved them most when readers took them as an endorsement. Ninety-four per cent uptime, or six per cent downtime: Same fact, different feeling. The framing effect is real and more modest than the famous examples suggest, and it is still your choice of words doing the pushing. A deliberately weak third option, added for contrast: Do not. The effect mostly lives in tidy tables of numbers and fades when people meet the real thing, and inside personalised recommendations a decoy made people walk away from the choice altogether. Foot: Hsee and colleagues (1999); Basu and Savani (2017); Jachimowicz and colleagues (2019); Frederick, Lee and Baskin (2014). — Five perfectly honest ways to lay out a brief, and what each of them quietly does to the answer.

Try this. Build the comparison as one table with identical rows for every option, and choose the rows before you fill them in. If a row is hard to fill for one option, that is information; leave it visible rather than quietly dropping the row.

The research behind this section

Hsee (1996) set out the evaluability hypothesis as an explanation for preference reversals between joint and separate evaluations of alternatives. Hsee and colleagues (1999) reviewed the evidence that people evaluate options differently and reverse their preferences between joint evaluation, in which multiple options are presented simultaneously and evaluated comparatively, and separate evaluation, in which options are presented in isolation, and proposed that attributes which are difficult to evaluate in isolation have a greater impact in joint evaluation. Basu and Savani (2017) ran seven experiments and found participants more likely to choose dominating options when options were presented simultaneously rather than sequentially, both when the dominance relationship was transparent and when it was not, with depth of cognitive processing rather than memory load as the mechanism. Basu and Savani (2019) reviewed the research on sequential and simultaneous presentation. Schneider and colleagues (2019) investigated decision quality across choice architectures with sixteen objectively ranked options and found joint presentation outperforming separate presentation. Levin, Schneider and Gaeth (1998) developed a typology distinguishing risky choice framing, attribute framing and goal framing, arguing that mixed findings in the literature followed from different operational definitions tapping different underlying processes. Kühberger (1998) meta-analysed 230 effect sizes from a pool of 136 empirical papers with nearly 30,000 participants and found the overall framing effect between conditions to be of small to moderate size, with profound differences between research designs. Ding, Li and Qiao (2025) published a further meta-analysis of the relationship between framing effects and risky decisions. Best and Charness (2015) meta-analysed eighteen studies with 3,232 subjects on age differences in framing. Zhen and Yu (2016) gave 159 young adults a vignette-based task and a gambling task and found no significant correlation between them, indicating low convergent validity between two classic measurements of the framing effect. Jachimowicz and colleagues (2019) meta-analysed fifty-eight default studies with a pooled n of 73,675 and found a considerable influence of defaults (d = 0.68, 95% confidence interval 0.53 to 0.83) alongside substantial variation, with defaults more effective when they operated through endorsement or endowment. Buchanan and colleagues (2023) found across two experiments that options were retained more when presented as the default, that the default effect was stronger when the context was framed as complex, and that participants rated defaults as more likely to represent the most popular option regardless of context or explanation. Xiao and colleagues (2021) ran two preregistered replications of the classic 1988 status quo bias scenarios with 311 and 316 participants and found strong support in three of the four scenarios. Johnson and colleagues (2012) set out the tools available to anyone presenting people with choices, divided into those that structure the choice task and those that describe the options. Frederick, Lee and Baskin (2014) found that the attraction effect is largely restricted to stylised representations in which every product dimension is a number, and does not typically occur when people experience the product or when even one attribute is represented perceptually. Mousavi, Adamopoulos and Bockstedt (2023) found that in a personalised recommendation context a decoy minimised demand for the target option and pushed consumers to opt out, deviating from the conventional decoy effect, while in a non-personalised context the conventional effect held.

A recommendation, a menu, or both

Now the question people in your seat ask most often. Do you tell them what you think, or do you lay it out neutrally and let them decide?

The largest body of evidence on this comes from a long programme of research into real organisational decisions. Across 358 decisions in medium and large organisations in the United States and Canada, roughly half failed, and the failures traced back to the same handful of moves: managers who imposed a solution, limited the search for alternatives, and used power rather than participation to implement. Tactics prone to failure were used in two of every three decisions studied. An earlier study of 168 decision cases looked specifically at how solution ideas were identified and found that the tactic of imposing a fully developed solution was widely used and seldom successful.

Read that carefully, because it is easy to take the wrong lesson from it. The finding concerns imposing a single fully formed answer with no visible search behind it, and says nothing about whether you should share your view.

The evidence on that second question is newer and pleasingly direct. In a randomised trial reported in 2025, 402 primary care doctors were shown two clinical scenarios and asked whether to stay with the existing plan or move to an alternative. Half saw one appropriate alternative; half saw two, three or four. Presenting two or more raised the proportion who moved off the existing plan from 44.0% to 62.1%, an adjusted odds ratio of 1.90. And here is the part I would write on the wall: increasing the number beyond two added nothing.

A separate line of work explains why. Across three online experiments simulating a choice of holiday package, showing people several recommendations rather than one significantly raised how likely they were to accept a recommendation, whether or not the system making it was accurate, and the researchers put it down to the need for autonomy. A single recommendation asks the reader to obey, while two ask them to decide, which is the thing they are actually there to do.

One option is a request for approval. Two options with a recommendation is a decision. The difference is whether the reader gets to exercise judgement, which is the only reason they are in the chair.

There is a real cost on the other side, and it deserves its own paragraph. An experiment published in 2025 isolated the effect of giving people extra information and the effect of giving them a recommendation. Recommendations made decisions more accurate, faster, and the system more usable. They also reduced people's ability to tell when the underlying information was wrong. People scrutinised the information more carefully when no recommendation came with it. That is the price of recommending: you make the decision easier and you make it less likely that anyone catches your mistake.

Which suggests the structure that gets the benefit without the whole of the cost. Recommend clearly, then make the working visible: what you compared, what would change your mind, and what you are least sure about. The recommendation carries the speed; the working preserves the scrutiny.

One more caution, for anyone tempted to sprinkle a paper with flags. Three experiments on recommendation signs in shops found that for people with well-developed preferences, the signs widened rather than narrowed what they considered and made the decision harder. If your executive knows the domain better than you do, a page covered in stars and highlights will slow them down.

Try this. Write the recommendation as one sentence in the present tense with an owner and a date, then immediately below it write the line beginning "I would change my mind if". The second sentence is what turns a recommendation into something a senior person can push against rather than simply accept or reject.

The research behind this section

Nutt (1999) reported that about half the decisions in organisations fail, drawing on studies of 358 decisions in medium to large organisations in the United States and Canada, and traced failures to managers who impose solutions, limit the search for alternatives and use power to implement their plans, noting that failure-prone tactics were used in two of every three decisions studied. Nutt (1993) studied 168 decision cases at the idea development stage and found that a synthesised template integrating practices from several sources had the most success but was seldom used, while the tactic of imposing a fully developed solution was widely used and seldom successful. Nutt (2004) set out three approaches for expanding the search for alternatives and avoiding what he called the limited-search trap, including identifying stakeholders and their claims, increasing the scope of the objective, and using multiple perspectives to uncover options. Nutt (1998) examined how alternatives are evaluated in making strategic choices. Keeney (1996) argued that alternatives matter only as means to achieve values and set out procedures for creating better alternatives by beginning with values rather than with the options in front of you. Altinger and colleagues (2025) randomised 402 United States primary care physicians across two clinical scenarios and found that those shown two or more appropriate treatment alternatives chose an alternative in 62.1% of decisions compared with 44.0% for those shown one, an adjusted odds ratio of 1.90 (95% confidence interval 1.09 to 3.30), and that increasing the number of alternatives beyond two did not increase the effect. Fink, Newman and Haran (2024) ran three online experiments and found that showing users multiple recommendations rather than a single recommendation significantly increased acceptance of the recommendation, both when the system was perfectly accurate and when it was not, which they attributed to the psychological need for autonomy. Gegoff and colleagues (2025) experimentally separated additional information from decision recommendations and found that when recommendations were provided participants made more accurate and faster decisions and rated the system as more usable, but that recommendation provision reduced their ability to discriminate the accuracy of the underlying information, with people scrutinising the information more when no recommendation was given. Goodman and colleagues (2013) found across three experiments that recommendation signage created preference conflict for consumers with more developed preferences, leading them to form larger consideration sets and experience more difficulty, with the effects mitigated for those with less developed preferences and for small assortments.

How many options belong on one page

The obvious follow-up is whether there is a ceiling, and this is where the literature has had its most public argument.

The famous demonstration is from 2000. Across three experiments in a shop and in a classroom, people were more likely to buy gourmet jams or chocolates, or to take on an optional essay, when offered six choices rather than twenty-four or thirty, and they reported more satisfaction and wrote better essays when the original set had been smaller. It is a wonderful finding and it launched a thousand slide decks.

Then in 2010 a meta-analysis of sixty-three conditions from fifty published and unpublished experiments, with 5,036 people in total, found a mean effect size of virtually zero, with considerable variance between studies and no sufficient conditions for the effect that the researchers could identify. Choice overload, on that reading, is not a law of nature.

The response came quickly, and it is the more useful of the two. A commentary argued that looking for a single main effect across all conditions was the wrong question, and a later meta-analysis of ninety-nine observations covering 7,202 people identified four factors that decide whether more options hurt: how complex the choice set is, how difficult the decision task is, how uncertain the reader is about their own preferences, and whether the reader's goal is to minimise effort. Where those conditions are present, the effect is real and significant. Where they are not, more options are fine.

Look at that list and then look at a Thursday evening executive reading a supplier brief on a phone: complex set, difficult task, preferences not yet formed, goal firmly set on minimising effort. Those are the exact conditions the meta-analysis says produce overload. Whether choice overload exists in general matters less to you than the fact that your particular reader is standing in the middle of all four of its preconditions.

Two further findings keep this from becoming a rule about magic numbers. Research on best-first lists found that anything encouraging people to consider more of the list makes their choices worse, including showing the top fifty rather than the top fifteen. A study of choice architectures found that a tournament structure, comparing a large set in small rounds, reduced overload without reducing the number of options. Working memory research gives the underlying constraint: the central capacity limit is about four chunks, not seven.

So the answer turns out to be a number of comparisons rather than a number of options, and the number of comparisons is about four.

A graphic titled How many options belong on the page, with the eyebrow From a 2025 trial of 402 doctors and two competing meta-analyses of choice overload. Standfirst: The bar shows roughly how well each shape of the page serves the person deciding, according to the evidence below it. The gain arrives at two options and stops there, and the losses start once the reader has more comparisons to hold than a working memory will carry. Five rows, each with the shape of the page on the left, a horizontal bar with a short label under it in the middle, and the evidence on the right. One option, bar at about two fifths, labelled A request for approval: When 402 doctors were shown a single appropriate alternative, 44.0% moved away from the existing plan. Organisational research finds the same: imposing one fully formed answer was widely used and seldom successful. Two options, with a recommendation, bar full, labelled Where the gain appears: Shown two or more alternatives, 62.1% of the same doctors moved. Showing several recommendations rather than one also raised acceptance in three experiments, through the reader's need to exercise their own judgement. Three or four options, bar full, labelled No further gain: In the same trial, adding a third and fourth alternative added nothing at all. A third live column costs the reader attention and buys nobody anything. A long list, ranked best first, bar at about a fifth, labelled Choices get worse: Showing the top fifty rather than the top fifteen recommendations made people choose worse, because they considered more of the list and were less selective about it. More comparisons than four, bar shortest, labelled Past the limit: The central capacity limit people actually work with is about four chunks, not the seven everybody quotes. The ceiling on a brief is a number of comparisons rather than a number of options. Foot: Altinger and colleagues (2025); Nutt (1993); Fink, Newman and Haran (2024); Diehl (2005); Cowan (2001). — Two options is where the gain shows up and where it stops. The third column is for you, not for them.

Try this. Bring two live options and one line naming what you ruled out and why. The ruled-out line does the work of showing the search was real, which the organisational research says is what separates a decision from an imposition, without asking the reader to hold a third column in their head.

The research behind this section

Iyengar and Lepper (2000) found across three experiments in field and laboratory settings that people were more likely to purchase gourmet jams or chocolates, or to undertake optional class essay assignments, when offered a limited array of six choices rather than a more extensive array of twenty-four or thirty, and reported greater satisfaction with their selections and wrote better essays when the original set had been limited. Scheibehenne, Greifeneder and Todd (2010) meta-analysed sixty-three conditions from fifty published and unpublished experiments (N = 5,036) and found a mean effect size of virtually zero with considerable variance between studies, identifying potentially important preconditions but no sufficient conditions for choice overload. Chernev, Böckenholt and Goodman (2010) replied that because choice overload has multiple antecedents, searching for a main effect across all conditions is not informative, and that the interesting question is when it occurs rather than whether. Chernev, Böckenholt and Goodman (2015) meta-analysed ninety-nine observations (N = 7,202) and identified four moderators, choice set complexity, decision task difficulty, preference uncertainty and an effort-minimising decision goal, finding each to have a reliable and significant impact and reporting that once moderators are taken into account the overall effect of assortment size is significant. Diehl (2005) found across three studies that in ordered environments, where options are sorted from predicted best to worst, factors that prompt consideration of more options degrade choice quality, including presenting the top fifty rather than the top fifteen recommendations. Besedeš and colleagues (2015) found that a tournament-style sequential architecture allowed the choice set to remain large while reducing choice overload, whereas an architecture comparing each subset to the most preferred option so far failed because of status quo bias. Cowan (2001) brought together evidence that the central capacity limit of short-term memory averages about four chunks rather than the seven often quoted.

Saying how sure you are

Every brief carries uncertainty, and the way it is written decides whether the reader adjusts for it or ignores it.

Begin with the single most quietly damaging habit in professional writing, which is the probability word on its own. In 1982 a researcher inside a professional forecasting organisation asked its own experts to put numbers on thirty verbal probability expressions taken largely from the organisation's published forecasts, and then to do the same for those expressions inside fifteen real paragraphs. There was high variability in how the same words were read, and the variability was higher in context rather than lower. The people involved were colleagues, writing in a shared house style, about their own subject.

Forty years of work has confirmed the pattern and found one formatting change that helps a great deal. In an online experiment with 924 participants, researchers compared four ways of presenting probability words against the numerical ranges in the United States intelligence community's own standard. Putting the numbers in brackets in the text, as in this is very unlikely (5 to 20%), lifted average correspondence with the intended meaning to 66%, against 32% when no guidance was given. Offering the same table behind a link or in a tooltip did almost nothing, and a later re-analysis using a fairer measure of agreement reached the same practical conclusion.

Two more findings sharpen the point. Three studies found that the same probability word is read as meaning a higher probability when the event it describes is severe, so that likely attached to a data breach carries a different weight from likely attached to a delayed delivery. And a 2025 study of 1,315 people found that when forecasts carried both a probability word and a separate confidence label, readers used the confidence label to work out the likelihood, and adding it did not improve their interpretation at all.

A graphic titled Saying how sure you are, with the eyebrow From forecasting organisations, intelligence standards and forty years on probability words. Standfirst: Every brief carries uncertainty, and the way it is written decides whether the reader adjusts for it or reads straight past it. Five habits, and what happens to the person on the other end of each one. Five rows, the habit named in a brown panel on the left with an example phrase under it, and the evidence in a pale panel on the right. The word on its own, Likely. Probable. A risk.: Forecasters asked to put numbers on their own organisation's probability words disagreed widely, and disagreed more when the words sat inside real paragraphs than when they stood alone. Your likely and their likely are different numbers. The word plus a number, Likely (55 to 80%): With 924 people, putting the numbers in brackets in the text lifted agreement with the intended meaning from 32% to 66%. Offering the same table behind a link did almost nothing. Four keystrokes, double the agreement. A single figure, The saving is 84,000 pounds: People give ranges that are systematically too narrow, sometimes only forty per cent as wide as they would need to be. Write the low and the high first, and then widen both of them. A confidence label, High confidence. Moderate confidence.: Given 1,315 readers, a separate confidence label was used to work out how likely the event was, and adding it did not improve how the forecast was understood. Say what you do not know and when you will know it instead. Red, amber and green, The risk matrix: Analysed properly, a typical matrix compares only a small fraction of pairs of risks correctly, gives identical ratings to very different risks, and in some cases produces worse than random decisions. It has the look of analysis and the behaviour of a colour scheme. Foot: Beyth-Marom (1982); Wintle and colleagues (2019); Soll and Klayman (2004); Duke (2025); Cox (2008). — Your likely and their likely are different numbers. Four keystrokes fix it.

So write the number down. That leaves the question of which number, and the research here is humbling in a way that is worth sitting with.

When people give a range they are X per cent sure contains the answer, the answer falls inside it far less than X per cent of the time. The main cause is that subjective intervals are systematically too narrow given the quality of the information behind them, sometimes only forty per cent as wide as they need to be. Being told to widen them does not help. Three experiments found that overprecision fell substantially when people were made to consider the whole range of possible outcomes divided into intervals and to assign a likelihood to each, with the effect carrying over into later, conventionally elicited ranges. A further set of experiments found that reasoning exercises after the first estimate, such as hypothetical bets, did little, while mechanically stretching the initial range and asking again worked better.

There is one common device I would retire on the evidence. The red, amber and green risk matrix, analysed properly, can correctly compare only a small fraction of randomly chosen pairs of hazards, assigns identical ratings to quantitatively very different risks, and where likelihood and severity are negatively correlated can produce worse than random decisions. It has the look of analysis about it, and it behaves like a colour scheme.

The honest counterweight is that uncertainty written well reads as competence rather than as hedging. An analysis of more than 1,500 strategic intelligence forecasts produced over roughly six years found a high level of discrimination and calibration, which is to say that professionals who write probabilities down, regularly and with a scoring rule waiting for them, get good at it. A meta-analysis of thirty-six studies and 1,485 clinicians found confidence better calibrated with accuracy than the field had assumed. Knowing how sure you are improves with the writing down, which is an argument for the practice rather than against it.

Try this. Replace every probability word in the brief with the word plus a bracketed range, and for every number you are estimating rather than quoting, write the low and the high before you write the middle. If the low and the high feel uncomfortably far apart, the research says they are probably still too close together.

The research behind this section

Beyth-Marom (1982) asked experts in a professional forecasting organisation to give numerical translations for thirty verbal probability expressions, most taken from the organisation's own published political forecasts, and then to translate the same expressions inside fifteen real paragraphs, and found high variability in interpretation which was higher still in context. Wintle and colleagues (2019) compared four formats for presenting verbal probabilities against the numerical guidelines in the United States Intelligence Community Directive 203 with 924 participants and found that correspondence with the standard improved substantially only when numerical guidelines were bracketed in the text, reaching an average of 66% against 32% in the control. Mandel and Irwin (2021) re-analysed those data using a measure of agreement that does not penalise within-range precision and reached the same conclusion about embedding numeric equivalents in text. Budescu, Por and Broomell (2012) tested ways of communicating uncertainty more effectively in the reports of the Intergovernmental Panel on Climate Change. Budescu and colleagues (2014) examined how the same probabilistic statements were interpreted in countries around the world. Harris and Corner (2011) found across three studies, using a manipulation of severity designed to avoid a confound with base rate, that verbal probability expressions referring to a severe event were interpreted as indicating a higher probability than the same expressions referring to a neutral event. Duke (2025) gave 1,315 participants intelligence forecasts expressed with verbal probability and verbal confidence and found that participants used the expressions of confidence to interpret the likelihood of events and that including them did not improve interpretation. Teigen (2023) reviewed what verbal probability terms and numeric ranges each convey beyond strength of expectation, including directionality and information about the speaker. Soll and Klayman (2004) found substantial overconfidence in interval estimates, with the correct answer falling inside intervals much less often than stated, and identified the main cause as intervals that are systematically too narrow given the accuracy of the judge's information, sometimes only forty per cent as large as needed for calibration. Haran, Moore and Morewedge (2010) reduced overprecision significantly across three experiments by presenting the entire range of possible outcomes divided into intervals and asking participants to estimate each interval's likelihood of containing the true answer, with carryover effects into subsequent conventional interval estimates. Ferretti, Montibeller and von Winterfeldt (2023) tested debiasing techniques in expert elicitation across two experiments and found hypothetical bets and counterfactual argumentation not very effective, while multipliers that increased the initial range coupled with re-elicitation gave more positive results. Cox (2008) analysed risk matrices and found poor resolution, with typical matrices correctly and unambiguously comparing only a small fraction of randomly selected pairs of hazards, errors in which higher qualitative ratings are assigned to quantitatively smaller risks, and cases with negatively correlated frequencies and severities in which matrices can be worse than useless and lead to worse than random decisions. Mandel and Barnes (2014) applied quantitative accuracy measures to more than 1,500 strategic intelligence forecasts spanning about six years of production and found a high level of discrimination and calibration skill. Miller, Spengler and Spengler (2015) meta-analysed thirty-six studies from 1970 to 2011 in which the confidence ratings of 1,485 clinicians were compared with the accuracy of their judgements and found a small but statistically significant effect (r = .15), indicating that confidence is better calibrated with accuracy than previously assumed.

Putting the objection in the paper

Every experienced person in your role has watched a decision get made in a meeting and unmade in a corridor. The research on dissent has a lot to say about why, and most of it can be built into a document rather than left to the room.

Start with what a group does to information when it already agrees. Three experiments, one run with managers, found that groups prefer information supporting the option they are leaning towards, and that the strength of the preference depends on how homogeneous the group's starting preferences are. The more members who had already chosen the same option before the discussion, the harder the group sought material supporting it, with higher commitment and confidence explaining the effect. A room that agrees is a room that has stopped checking.

The old remedy for this was structured disagreement. Two laboratory studies in the 1980s compared dialectical inquiry, where a plan is met with a counter-plan, devil's advocacy, where somebody is assigned to attack the plan, and plain consensus, once with students and once with fast-advancing middle managers. A meta-analysis in 1989 revisited the same body of studies and reached different conclusions from an earlier narrative review, a useful reminder of how contested this ground has been. A study of what people expect from these techniques found that expectations lined up with the feelings the earlier studies produced rather than with the objective outcomes, which is to say that people dread structured disagreement rather more than the results justify.

The modern evidence is more precise about when disagreement helps. A meta-analysis of 116 studies covering 8,880 groups found stable negative relationships for relationship conflict and for process conflict, and no strong negative association for task conflict. Task conflict looked best where it was least entangled with personal conflict, in top management teams rather than elsewhere, and where the outcome measured was financial performance or decision quality. An earlier study of top management teams described conflict's consistent and contradictory effects on decision quality, consensus and acceptance, which is the paradox the whole literature circles.

The practical translation is that you want the argument about the work rather than the argument about the people, and a document is an unusually good place to have the first without the second. A written objection has no tone of voice.

Which brings us to the pre-mortem, the single technique I would most want in a decision brief. The method is simple: before the decision is taken, assume it has already failed, and write down why. A study comparing it with brainstorming for risk identification found that teams using the pre-mortem identified better quality risks, proposed more quality changes to the plan, and surfaced more of the hard-to-predict, potentially catastrophic risks the researchers were looking for. A software and human factors team used it at the early design stage of a complex system and reported finding risks they had not anticipated, with better communication as a side effect. In a 2025 study, twenty-three participants were told a community health programme had failed and asked why, and the dominant answer concerned how well it fitted the setting rather than anything about the programme itself.

A pre-mortem has nothing to do with pessimism. It is the one reliable way to hear the objection while there is still time to do something about it.

The counterweight belongs here rather than in a footnote. Structured analytic techniques of this family have not been tested nearly as thoroughly as their popularity suggests. A 2018 analysis argued that they treat two-sided biases as one-sided, so nobody knows when a technique aimed at overconfidence has tipped a team into underconfidence, and that the assumption behind breaking a problem into a grid, that decomposition reduces noise, may add noise instead. An evidence-based review of twelve core techniques reached a similarly cautious verdict. The pre-mortem has better direct evidence behind it than most of its cousins.

Try this. Put two lines at the foot of the brief under the heading "What would have to be true for this to be wrong". Write them yourself, in the voice of the person in the room most likely to object, and name them: the finance director will say the payback assumes no price rise in year two. Naming it turns an ambush into an agenda item.

The research behind this section

Schulz-Hardt and colleagues (2000) found across three experiments, one replicated with managers, that groups as well as individuals prefer information supporting their existing preference, that the bias was stronger the more group members had chosen the same alternative before discussion, and that higher commitment and confidence in homogeneous groups mediated the effect. Schweiger, Sandberg and Ragan (1986) compared dialectical inquiry, devil's advocacy and consensus approaches to strategic decision making in a laboratory study of groups. Schweiger, Sandberg and Rechner (1989) compared the same three approaches longitudinally with fast-advancing middle managers involved in strategic planning. Schwenk (1989) used meta-analysis to revisit this body of studies and reached different conclusions from those drawn in an earlier narrative review. Priem and Price (1991) examined expectations of cognitive conflict, social conflict, decision confidence and post-decision group affect for the three techniques and found that expectations showed some congruence with the affective outcomes of prior studies but not with the objective ones. De Wit, Greer and Jehn (2012) meta-analysed 116 studies of intragroup conflict covering 8,880 groups and found stable negative relationships for relationship and process conflict, no strong negative association between task conflict and group performance, and task conflict more positively related to performance where it was weakly associated with relationship conflict, in top management teams, and where performance was measured as financial performance or decision quality. De Wit, Jehn and Scheepers (2013) examined task conflict, information processing and decision making, and the damaging effect of relationship conflict. Amason (1996) provided evidence from two samples of conflict's consistent yet contradictory effects on decision quality, consensus and affective acceptance in top management teams. Gallop, Willy and Bischoff (2016) compared brainstorming and the pre-mortem technique for risk identification and concluded that teams using the pre-mortem identified better quality risks, more quality changes to the plan, and more hard-to-predict catastrophic risks. Bettin and colleagues (2022) used the pre-mortem method at an early stage of designing a novel sociotechnical system and found it valuable in recognising and mitigating previously unanticipated risks and in enriching team communication. Wippold and colleagues (2025) ran implementation pre-mortem activities with twenty-three participants who were told a community health effort had failed and asked to identify sources of failure, and found that the most frequently identified priority concerned the fit of the programme with its setting. Wippold, Garcia, Domlyn and Wong (2025) set out the theory behind the implementation pre-mortem and how to analyse the data it produces. Veinott and Lehman (2024) compared human and large language model responses in pre-mortem planning. Chang and colleagues (2018) argued that structured analytic techniques treat bipolar biases as unipolar, leaving no way of knowing when a technique aimed at one bias triggers its opposite, and that the assumption that problem decomposition reduces noise has never been tested and may degrade the reliability of analytic judgement. Coulthart (2017) conducted an evidence-based evaluation of twelve core structured analytic techniques.

What a structure actually buys you

It is fair to ask whether any of this makes decisions better or only makes them faster, and there are two literatures with an answer.

The first is about organisational process. A longitudinal field study followed fifty-two strategic decisions in twenty-four companies to see whether how a decision was made related to how it turned out. The same research group had earlier studied fifty-seven decisions in those companies to ask why some procedures are more rational than others, and found environmental threat, perceived external control and the uncertainty of the issue all playing a part, some in the opposite direction to what had been predicted. A later study in Egypt found decision, environmental and firm characteristics all shaping rationality, with no single perspective modelling it adequately. A 1984 paper on comprehensiveness had already warned that in an unstable industry, being thorough correlated with worse performance, which is the standing caution against turning a good habit into a ritual.

The second literature is more encouraging and much larger, and it comes from medicine. The Cochrane review of patient decision aids, updated in 2024, covers 209 randomised trials and 107,698 participants across seventy-one different decisions. A decision aid, at minimum, makes the decision explicit, sets out the options with their benefits and harms, and helps the person work out what they value. Compared with usual care, decision aids improved knowledge by about twelve points on a hundred-point scale across 107 studies, nearly doubled the proportion of people with accurate risk perceptions, reduced the sense of being uninformed and unclear about one's own values, and reduced the proportion who left the decision entirely to the professional, with no increase in regret. Those are high-certainty findings, and a decision brief that does those four things is doing what a decision aid does.

Checklists tell a similar story with a sharper edge. A meta-analysis of nineteen studies found surgical safety checklists improved teamwork and communication with a large effect, and smaller but significant effects on compliance with the safety measures themselves. A 2018 analysis argued that in general surgery the checklist delivers far less than the early enthusiasm promised, a reminder that the effect depends on what a format makes people do rather than on the format itself.

Two smaller findings keep the humility going. A meta-analysis of seventy-five clinical judgement studies covering 4,607 clinicians found experience improving accuracy by about thirteen per cent, which is less than most experienced people expect. And a 1975 paper showed that treating your criteria as equally important, instead of agonising over weights, performs surprisingly well. If you have been putting off a brief because you cannot decide whether cost should count for thirty or forty per cent, that paper is your permission to stop.

Try this. Check your brief against the four things a decision aid does: it names the decision explicitly, it sets out the options with what each one costs and gains, it says what the organisation actually values here, and it tells the reader what is still unknown. If a section does not do one of those four jobs, it belongs in the appendix.

The research behind this section

Dean and Sharfman (1996) used a longitudinal field study design to examine fifty-two strategic decisions in twenty-four companies in order to test whether strategic decision-making processes are related to decision effectiveness. Dean and Sharfman (1993) studied fifty-seven strategic decisions in twenty-four companies using a multiple-informant structured interview protocol and found environmental competitive threat, perceived external control and the uncertainty of the strategic issues related to procedural rationality, with some relationships running opposite to predictions. Elbanna and Child (2007) developed an integrated model of strategic decision-making rationality and found in a study in Egypt that decision, environmental and firm characteristics all shaped it, that no single perspective modelled decision processes adequately, and that the national setting mattered. Fredrickson (1984) examined comprehensiveness in strategic decision processes, following an earlier report of a negative relationship between comprehensiveness and performance in an industry with an unstable environment. Stacey and colleagues (2024) updated the Cochrane review of patient decision aids to 209 randomised trials involving 107,698 participants across seventy-one decisions, and found moderate-certainty evidence that decision aids probably increase congruence between informed values and care choices (risk ratio 1.75), and high-certainty evidence that they improve knowledge (mean difference 11.90 out of 100 across 107 studies and 25,492 participants) and accuracy of risk perceptions (risk ratio 1.94), decrease decisional conflict related to feeling uninformed and to indecision about personal values, reduce the proportion of people who are passive in decision-making (risk ratio 0.72), and produce no difference in decision regret. Lyons and Popejoy (2014) meta-analysed nineteen studies of surgical safety checklists and found an effect size of 1.180 on teamwork and communication, 0.123 on morbidity and 0.268 on compliance with safety measures. Vargas and Servillo (2018) argued that the World Health Organisation surgical safety checklist does not deliver the improvements claimed for it in general surgery. Spengler and colleagues (2009) meta-analysed seventy-five clinical judgement studies in which the experience of 4,607 clinicians was assessed against the accuracy of their judgements and found a small but reliable effect, d = .12, amounting to almost a thirteen per cent increase in decision-making accuracy. Einhorn and Hogarth (1975) examined unit weighting schemes for decision making.

The record that stops it being reopened

The last part of a decision brief is the part almost nobody writes, and it is the part that saves you the most time.

Start with what makes people accept a decision they did not get. An experiment with intact management teams found that when members' input was considered and members had influence, they judged the process fairer, and that perceived fairness carried through into commitment to the decision, attachment to the group and trust in the leader. A long line of procedural justice research makes the same argument for why people cooperate voluntarily rather than hoarding ideas and dragging their feet. A study of sixty-eight strategic decision-making teams in hospitals found consensus building commitment and commitment predicting implementation success, with one delightfully awkward finding attached: commitment slowed implementation down. People who are committed take the time to do it properly.

The design implication is small and specific. A brief that visibly contains the objections people raised, attributed and answered, buys more acceptance than one presenting a tidy unanimous case. The dissenting view reads as a receipt showing the process was fair, rather than as a weakness in your argument.

Then the record itself, and the reason to write it on the day. People who know how something turned out believe it was more predictable than it was, and a meta-analysis of ninety-five studies and 252 effect sizes puts the mean effect at a moderate size and reports that studies which tried to reduce hindsight bias produced effects of the same size. Warning people about it in advance achieves very little.

What that does to a decision is shown most vividly in a legal study. Participants judged whether a municipality should take precautions against a flood. Reviewing the evidence as it stood beforehand, three quarters concluded the flood was too unlikely to justify further precautions. Reviewing parallel material after the flood, a majority judged that same decision negligent, and a debiasing procedure made no significant difference. That is what a decision of yours looks like eighteen months later to somebody who knows what happened, and it is why the only real defence is a record written before the outcome existed.

The last piece is the review. A study of 188 people in forty-seven teams found that teams which ran an after-action review performed better and reported better communication, efficacy and cohesion than teams which did not, and that reviews built on memory did about as well as reviews built on a recording, which matters because most of us will never have one.

Try this. Before you close the file, add four lines to the bottom of the brief: what was decided, who decided it, what we believed at the time that made it the right call, and what would make us revisit it. Send it in the same thread. When somebody reopens the decision in March, you are not arguing from memory against a person who knows the ending.

The research behind this section

Korsgaard, Schweiger and Sapienza (1995) found in an experiment with intact management teams that consideration of member input and members' influence on a decision affected perceptions of procedural fairness, and through them commitment to the decision, attachment to the group and trust in the leader, with perceived fairness partially mediating the effect. Kim and Mauborgne (1998) drew on procedural justice research to argue that when people feel strategic decision-making processes are fair they cooperate voluntarily through trust and commitment, and that when they feel the processes are unfair they hoard ideas and drag their feet. Dooley, Fryxell and Judge (2000) studied sixty-eight intact strategic decision-making teams in hospitals and found that decision consensus helped build decision commitment, which positively affected implementation success, while commitment served to slow implementation speed. Guilbault and colleagues (2004) meta-analysed ninety-five studies of hindsight bias with 252 independent effect sizes, found an overall mean effect size of .39, and found that studies including manipulations designed to reduce hindsight bias did not produce lower effect sizes. Kamin and Rachlinski (1995) found that three quarters of participants judging in foresight concluded a flood was too unlikely to justify further precautions, that a majority of participants judging the same decision in hindsight found it negligent, that hindsight participants gave higher probability estimates for the disaster, and that a debiasing procedure produced no significant improvement. Villado and Arthur (2013) assigned 188 individuals in forty-seven four-person teams to after-action review conditions on a cognitively complex task and found that teams which ran a review had higher team performance, team efficacy, openness of communication and cohesion than teams which did not, with no significant differences between objective and subjective reviews. Keiser and Arthur (2022) meta-analysed the task and training characteristics that contribute to or attenuate the effectiveness of the after-action review. Hogarth and Einhorn (1992) set out the belief-adjustment model of how the order in which evidence arrives changes the belief a person ends up holding.

The one page, line by line

Here is the page itself, with what each line is doing and the evidence that put it there.

The decision, in one sentence, at the top. The reading research says the first block is the only part guaranteed to be read, so the question being answered belongs in it. Name the decision, not the topic: whether to renew with the current supplier for twelve months is a decision, and supplier update is a topic.

The recommendation, immediately after, with an owner and a date. This is the line the recommendation research supports and the line the organisational research warns about, so it carries both: the recommendation earns the speed, and the lines beneath it preserve the scrutiny it would otherwise remove.

The reason, in one sentence. The single reason rather than the list, because the working is what keeps the reader checking rather than accepting, and because a reason a reader can hold is a reason they can argue with.

Two options, side by side, in one table with identical rows. Two rather than one, because two is where the gain appeared and where it stopped. Side by side rather than one after another, because that is the difference between a reader who can weigh implementation time and a reader who can only weigh price. Identical rows, because the empty cell is itself a finding.

A third line naming what you ruled out and why. One line. It shows the search was real, which is what the organisational research says separates a decision from an imposition, without asking anyone to hold a third column in their head.

The numbers, with ranges rather than points. Low, high and the basis, because intervals written without thought are systematically too narrow and a single figure invites a precision nobody has.

Every probability word followed by a bracketed number, and every unknown given a date. Likely (55 to 80%). The bracket doubled agreement in the experiment and takes four keystrokes, and uncertainty with a closing date reads as a plan where uncertainty without one reads as a shrug.

What would have to be true for this to be wrong. Two lines, written in the voice of the person most likely to object, named. This is the pre-mortem, compressed into the only space a one-pager has for it.

The decision record, filled in afterwards. What was decided, by whom, what we believed at the time, and what would make us revisit it. Written on the day, because nobody can reconstruct what they believed once they know how it ended.

Try this. Take the last brief you wrote and mark it against those nine lines. Most papers I have seen have four of them, and the five that are missing are almost always the ones that would have saved the second meeting.

The research behind this section

This section draws together the studies set out above: Shiely, Gallagher and Millar (2024) on what readers do first; Brick, McDowell and Freeman (2020) on tabular presentation; Hsee and colleagues (1999) and Basu and Savani (2017) on joint and simultaneous presentation; Altinger and colleagues (2025) on two alternatives rather than one; Nutt (1999) and Nutt (1993) on the visible search for alternatives; Soll and Klayman (2004) and Haran, Moore and Morewedge (2010) on interval width; Wintle and colleagues (2019) and Mandel and Irwin (2021) on bracketed numerical equivalents; Gallop, Willy and Bischoff (2016) and Bettin and colleagues (2022) on the pre-mortem; Stacey and colleagues (2024) on what a decision aid does; Korsgaard, Schweiger and Sapienza (1995) on procedural fairness and commitment; and Guilbault and colleagues (2004) and Kamin and Rachlinski (1995) on hindsight and the value of a record written before the outcome is known.

The brief check

All of this comes down to one moment. You have a decision to write up, a particular person who has to make it, a certain amount of time before they do, and a page. So here is a tool for that moment.

Put in the decision, who has to make it, how many real options you have, whether you have a view, how reversible it is, how much is still unknown and how long your reader is likely to give it. It will tell you what shape the page should take, where the risk in this brief sits, what to put in and what to leave out, and it will give you a skeleton to copy.

Where to start

If you take one thing from all of this, make it the table. Two options, side by side, identical rows, with your recommendation named above it. That single change does more work than anything else on the page, because it is the difference between a reader who can compare the things that matter and a reader who can only compare the price.

After that, there are three moves worth making.

Put the number in brackets after every probability word, so that likely means the same thing to both of you. Write the two lines beginning "what would have to be true for this to be wrong" in the voice of the person most likely to object. And add the four-line record at the bottom on the day the decision is made, while everybody still remembers what they believed.

I have put the three pages I would use into a download: the one-page brief itself, an option table with the rows already chosen, and a decision record with the pre-mortem lines built in.

A graphic titled The one-page decision brief, with the eyebrow The whole post on one side of A4, with the evidence for each line. Standfirst: Nine lines, in this order. Each one is there because something in the research put it there, and the note beside it says what. Nine rows laid out as a sheet, each with the line name and its instruction on the left, the worked example in the middle, and the evidence in a small panel on the right. The decision, one sentence, at the top: Whether to renew with the current supplier for twelve months, and not: supplier update. Evidence: 98.6% of readers read the first block, and use it to decide whether to read on. The recommendation, with an owner and a date: I recommend we renew for twelve months, signed by the 30th, owner: me. Evidence: Recommendations make decisions faster and more accurate, and make readers less likely to spot a flaw. Hence the next two lines. The reason, one sentence only: Because it holds the price through the rebuild and keeps the exit open. Evidence: The working is what preserves the scrutiny a recommendation otherwise removes. The options, two, side by side: Cost, Time to implement, Exit terms, Who carries the risk. Evidence: Two is where the gain appeared and where it stopped: 44.0% moved with one option, 62.1% with two, and nothing beyond that. What we ruled out, one line, with the reason: Open tender: eleven weeks, and the rebuild starts in six. Evidence: Imposing one answer with no visible search behind it was the commonest failure in 358 real decisions. The numbers, low, high, and the basis: Saving of 62,000 to 110,000 pounds, on last year's volumes. Evidence: Ranges written without thought come out too narrow, sometimes only forty per cent as wide as they should be. How sure, bracketed, with a closing date: Likely (55 to 80%) that the rebuild slips past March; supplier confirms on the 12th. Evidence: Numbers in brackets lifted agreement with the intended meaning from 32% to 66%. If this is wrong, in the objector's own voice: Finance will say the payback assumes prices hold flat. Evidence: Teams running a pre-mortem found better risks and more of the unthinkable ones than teams brainstorming. The record, filled in on the day itself: Decided, by whom, what we believed, what reopens it. Evidence: Three quarters called a risk too small beforehand; afterwards, a majority called the same call negligent. Foot: Shiely and colleagues (2024); Altinger and colleagues (2025); Nutt (1999); Soll and Klayman (2004); Gallop, Willy and Bischoff (2016). — The whole post on one side of A4, with the evidence for every line printed beside it.

There is something quietly powerful about being the person whose papers get decided from rather than deferred. It has little to do with writing more or faster, and a great deal to do with understanding that a decision brief is a piece of equipment: it has a job, it has a user, and the research now tells us what makes it work in their hands rather than in yours. That is a craft, and like every craft in this role, it is learnable and it compounds.

The part of this that shows up in how you are seen at work sits inside what the Connected Leader Assessment calls Judgement and Decision-Making: how you frame a choice for somebody else, how well you hold and communicate what is uncertain, and whether the decisions that pass across your desk get made once and stay made. It is the dimension that decides whether you are handling decisions or shaping them.

The assessment is free, takes about twelve minutes and gives you a read across all five dimensions. If you want to go further on this one, the Skill Sprints are short, focused programmes on a single dimension, and the Irreplaceable Audit is the place to look at how much of your organisation's decision-making quietly runs through your drafting, because that is usually a great deal more than the job description admits.

Stay savvy.

Meg ✌️

Free tools

The brief check, the interactive tool in this post, free
The Decision Brief Pages, the download in this post, free
The Connected Leader Assessment, twelve minutes, free, with a read on Judgement and Decision-Making, the dimension this post sits in
The Irreplaceable Audit, forty-eight activities across four kinds of work, free, for seeing how much of the organisation's decision-making runs through your drafting

P.S. Want to Go Deeper on Decision Briefs?

If this resonated and you want to go further, here is the evidence the piece is built on. These are the papers that shaped my thinking on how people read a document they have to act on, what the shape of the options does to the answer, how uncertainty should be written, and what makes a decision stay decided, and they are well worth your time.

Every DOI below has been checked. Where I could not confirm a figure from the paper itself, I have left the figure out rather than guess.

📃 Hartley, Sydes and Blurton (1996) Whether structured abstracts let readers get information more accurately and quickly. https://doi.org/10.1177/016555159602200503

📃 Hartley (2000) Thirty abstracts compared. The structured versions were more informative and no less accurate. https://doi.org/10.1177/016555150002600408

📃 Hartley (1999) Rewritten in structured form, abstracts were longer, more informative and judged clearer by readers. https://doi.org/10.1016/s0003-6870(99)00004-6

📃 Shiely, Gallagher and Millar (2024) 98.6% of experienced readers read the abstract first, to decide whether to read on at all. https://doi.org/10.1371/journal.pone.0297034

📃 Brick, McDowell and Freeman (2020) 2,305 people. The same facts in a small table beat the same facts in prose on comprehension and on recall six weeks later. https://doi.org/10.1098/rsos.190876

📃 Peters and colleagues (2007) Three studies on comparative quality information. Less is more, and most of all for readers least at ease with numbers. https://doi.org/10.1177/10775587070640020301

📃 Kurtzman and Greene (2016) Thirty-one experiments. People understood more and chose better when the display was less complex. https://doi.org/10.1016/j.pec.2015.07.030

📃 Kerwer and colleagues (2021) 166 people, preregistered. Plain language summaries were understood more correctly and read as more credible. https://doi.org/10.1525/collabra.18898

📃 Petkovic and colleagues (2018) Six trials. Evidence summaries were easier to understand; whether they changed what policy-makers did stayed unclear. https://doi.org/10.4073/csr.2018.8

📃 Oppenheimer (2006) Five experiments. Longer words made authors look less intelligent, through how hard the text felt to read. https://doi.org/10.1002/acp.1178

📃 Masson and Waldron (1994) Plain redrafting of legal contracts improved comprehension reliably, and comprehension was still low. https://doi.org/10.1002/acp.2350080107

📃 Lorch and Lorch (1996) What headings do to what readers recall and how they summarise. https://doi.org/10.1006/ceps.1996.0022

📃 Hartley and Trueman (1983) Nine experiments. Headings helped recall, search and retrieval, wherever they sat on the page. https://doi.org/10.1111/j.2044-8279.1983.tb02551.x

📃 Alpizar, Adesope and Wong (2020) Meta-analysis of the signalling principle: marking what matters in a document. https://doi.org/10.1007/s11423-020-09748-7

📃 Richter, Scheiter and Eitel (2016) A comprehensive meta-analysis of signalling between text and pictures. https://doi.org/10.1016/j.edurev.2015.12.003

📃 Delgado and colleagues (2018) Fifty-four studies, 171,055 people. Paper beat screen, and the gap widened under time pressure. https://doi.org/10.1016/j.edurev.2018.09.003

📃 Clinton (2019) Thirty-three experiments comparing reading from paper and from screens. https://doi.org/10.1111/1467-9817.12269

📃 Eppler and Mengis (2004) Thirty years of research on information overload, and what actually causes it. https://doi.org/10.1080/01972240490507974

📃 El-Sayed and colleagues (2021) More readable disclosures improved judgements; with complex ones, the order of the information changed the verdict. https://doi.org/10.1108/jaar-06-2020-0115

📃 Hogarth and Einhorn (1992) The belief-adjustment model: why the order evidence arrives in changes where you end up. https://doi.org/10.1016/0010-0285(92)90002-j

📃 Cowan (2001) The capacity limit people work with is about four chunks, not the seven everybody quotes. https://doi.org/10.1017/s0140525x01003922

📃 Levin, Schneider and Gaeth (1998) The typology that keeps risky choice, attribute and goal framing apart. https://doi.org/10.1006/obhd.1998.2804

📃 Kühberger (1998) 136 papers, nearly 30,000 people, 230 effect sizes. Framing is real and smaller than the famous examples suggest. https://doi.org/10.1006/obhd.1998.2781

📃 Ding, Li and Qiao (2025) A 2025 meta-analysis revisiting framing effects and risky decisions. https://doi.org/10.1016/j.socec.2025.102351

📃 Best and Charness (2015) Eighteen studies, 3,232 people, on whether framing lands differently at different ages. https://doi.org/10.1037/a0039447

📃 Zhen and Yu (2016) 159 people, two classic framing measures, no correlation between them. https://doi.org/10.1038/srep30071

📃 Hsee (1996) The evaluability hypothesis: why an option scores differently alone and beside its rivals. https://doi.org/10.1006/obhd.1996.0077

📃 Hsee and colleagues (1999) Joint and separate evaluation reviewed. Hard-to-judge attributes only count when there is something to compare them with. https://doi.org/10.1037/0033-2909.125.5.576

📃 Basu and Savani (2017) Seven experiments. People picked the better option more often when the options were shown together. https://doi.org/10.1016/j.obhdp.2017.01.004

📃 Basu and Savani (2019) A review of what changes when options are seen one at a time rather than all at once. https://doi.org/10.1177/0963721418806646

📃 Jachimowicz and colleagues (2019) Fifty-eight studies, 73,675 people. Defaults moved decisions (d = 0.68), most when read as endorsement. https://doi.org/10.1017/bpp.2018.43

📃 Buchanan and colleagues (2023) People treat a preselected option as a recommendation, and assume it is what most others chose. https://doi.org/10.1111/ijcs.12847

📃 Xiao and colleagues (2021) Two preregistered replications. The status quo bias held in three of four classic scenarios. https://doi.org/10.15626/mp.2020.2470

📃 Johnson and colleagues (2012) The catalogue of choice architecture tools: how you structure the task and how you describe the options. https://doi.org/10.1007/s11002-012-9186-1

📃 Frederick, Lee and Baskin (2014) The decoy effect mostly lives in stylised number tables and largely vanishes when people meet the real thing. https://doi.org/10.1509/jmr.12.0061

📃 Mousavi, Adamopoulos and Bockstedt (2023) In a personalised setting, a decoy reduced demand for the option it was meant to flatter. https://doi.org/10.1287/isre.2022.1197

📃 Altinger and colleagues (2025) 402 doctors. Two alternatives rather than one raised the move off the existing plan from 44.0% to 62.1%; more than two added nothing. https://doi.org/10.1001/jamanetworkopen.2025.42949

📃 Fink, Newman and Haran (2024) Three experiments. Several recommendations rather than one raised acceptance, through the need for autonomy. https://doi.org/10.1016/j.chb.2024.108244

📃 Gegoff and colleagues (2025) Recommendations made decisions faster and more accurate, and made people less able to spot bad information. https://doi.org/10.1177/00187208251318465

📃 Nutt (1999) 358 decisions. Half failed, and the failures traced to imposed solutions and a limited search. https://doi.org/10.5465/ame.1999.2570556

📃 Nutt (2004) Three ways to escape the limited-search trap during a strategic decision. https://doi.org/10.5465/ame.2004.15268668

📃 Nutt (1993) 168 decision cases. Imposing a fully developed solution was widely used and seldom successful. https://doi.org/10.1287/mnsc.39.9.1071

📃 Nutt (1998) How alternatives are evaluated when strategic choices are made. https://doi.org/10.1016/s0305-0483(97)00068-6

📃 Goodman and colleagues (2013) Three experiments. Recommendation signs made choosing harder for people who already knew what they wanted. https://doi.org/10.1016/j.jcps.2012.06.003

📃 Keeney (1996) Value-focused thinking: start from what you are trying to achieve and the better options appear. https://doi.org/10.1016/0377-2217(96)00004-5

📃 Iyengar and Lepper (2000) Six choices beat twenty-four or thirty for purchases, essays written and satisfaction afterwards. https://doi.org/10.1037/0022-3514.79.6.995

📃 Scheibehenne, Greifeneder and Todd (2010) Fifty experiments, 5,036 people. The average choice overload effect was close to zero. https://doi.org/10.1086/651235

📃 Chernev, Böckenholt and Goodman (2015) Ninety-nine observations, 7,202 people. Four conditions decide whether more options hurt. https://doi.org/10.1016/j.jcps.2014.08.002

📃 Chernev, Böckenholt and Goodman (2010) The reply: the question is not whether choice overload happens but when. https://doi.org/10.1086/655200

📃 Besedeš and colleagues (2015) A tournament structure cut choice overload while keeping the choice set large. https://doi.org/10.1162/rest_a_00506

📃 Diehl (2005) In a best-first list, showing the top fifty rather than the top fifteen made choices worse. https://doi.org/10.1509/jmkr.2005.42.3.313

📃 Schneider and colleagues (2019) Sixteen ranked options, several architectures. Joint presentation beat separate presentation on decision quality. https://doi.org/10.1287/deca.2018.0379

📃 Beyth-Marom (1982) Forecasters translating their own probability words. High variability, and higher still in context. https://doi.org/10.1002/for.3980010305

📃 Wintle and colleagues (2019) 924 people. Numbers bracketed in the text lifted agreement with the intended meaning from 32% to 66%. https://doi.org/10.1371/journal.pone.0213522

📃 Budescu and colleagues (2014) How the same probability statements were read in countries around the world. https://doi.org/10.1038/nclimate2194

📃 Budescu, Por and Broomell (2012) Testing ways of communicating uncertainty in the IPCC reports. https://doi.org/10.1007/s10584-011-0330-3

📃 Mandel and Irwin (2021) A re-analysis with a fairer measure of agreement. Bracketed numbers still won. https://doi.org/10.1371/journal.pone.0248424

📃 Mandel and Barnes (2014) More than 1,500 intelligence forecasts over six years, with high discrimination and calibration. https://doi.org/10.1073/pnas.1406138111

📃 Teigen (2023) What verbal terms and numeric ranges each convey beyond the strength of an expectation. https://doi.org/10.1007/s12144-022-03985-0

📃 Soll and Klayman (2004) The range you write is systematically too narrow, sometimes only forty per cent as wide as it needs to be. https://doi.org/10.1037/0278-7393.30.2.299

📃 Haran, Moore and Morewedge (2010) Three experiments. Estimating the whole spread of outcomes cut overprecision, and the effect carried over. https://doi.org/10.1017/s1930297500001637

📃 Ferretti, Montibeller and von Winterfeldt (2023) Hypothetical bets and counterfactuals did little; mechanically widening the range and asking again worked better. https://doi.org/10.1016/j.ejor.2022.04.008

📃 Cox (2008) Risk matrices compare only a small fraction of pairs correctly, and can be worse than useless. https://doi.org/10.1111/j.1539-6924.2008.01030.x

📃 Harris and Corner (2011) Three studies. The same probability word is read as a higher probability when the event is severe. https://doi.org/10.1037/a0024195

📃 Duke (2025) 1,315 people. Readers used confidence labels to judge likelihood, and the labels did not improve interpretation. https://doi.org/10.1080/02684527.2024.2392376

📃 Miller, Spengler and Spengler (2015) Thirty-six studies, 1,485 clinicians. Confidence tracked accuracy better than the field had assumed. https://doi.org/10.1037/cou0000105

📃 Schweiger, Sandberg and Ragan (1986) Dialectical inquiry, devil's advocacy and consensus compared in groups making strategic decisions. https://doi.org/10.2307/255859

📃 Schweiger, Sandberg and Rechner (1989) The same three approaches compared longitudinally with fast-advancing middle managers. https://doi.org/10.2307/256567

📃 Schwenk (1989) A meta-analysis that revisited the devil's advocacy studies and reached a different conclusion from the earlier review. https://doi.org/10.1002/smj.4250100309

📃 Priem and Price (1991) What people expect from structured disagreement matched the feelings it produces, not the results. https://doi.org/10.1177/105960119101600207

📃 Schulz-Hardt and colleagues (2000) Three experiments, one with managers. The more a group already agreed, the harder it looked for agreeing information. https://doi.org/10.1037/0022-3514.78.4.655

📃 Gallop, Willy and Bischoff (2016) Pre-mortem against brainstorming. Better risks, better changes to the plan, and more of the unthinkable ones. https://doi.org/10.1080/19488289.2016.1240118

📃 Bettin and colleagues (2022) A design team used a pre-mortem early and found risks it had not anticipated. https://doi.org/10.1177/1071181322661307

📃 Wippold, Garcia, Domlyn and Wong (2025) The theory behind the implementation pre-mortem and how to analyse what it produces. https://doi.org/10.1186/s43058-025-00768-z

📃 Wippold and colleagues (2025) Twenty-three people told a programme had failed. The commonest reason was fit with the setting. https://doi.org/10.1093/tbm/ibae060

📃 Veinott and Lehman (2024) People and a language model evaluating the same plan in a pre-mortem. https://doi.org/10.1007/978-3-031-76827-9_15

📃 Dean and Sharfman (1996) Fifty-two strategic decisions in twenty-four companies, followed over time, to ask whether process matters. https://doi.org/10.2307/256784

📃 Dean and Sharfman (1993) Fifty-seven decisions in twenty-four companies on why some procedures are more rational than others. https://doi.org/10.1111/j.1467-6486.1993.tb00317.x

📃 Elbanna and Child (2007) Decision, environmental and firm characteristics all shape how rational a decision process is. https://doi.org/10.1111/j.1467-6486.2006.00670.x

📃 Stacey and colleagues (2024) 209 trials, 107,698 people. Decision aids raised knowledge and accurate risk perception, and added no regret. https://doi.org/10.1002/14651858.cd001431.pub6

📃 Lyons and Popejoy (2014) Nineteen studies. Surgical checklists improved teamwork and communication most, and outcomes by a smaller margin. https://doi.org/10.1177/0193945913505782

📃 Vargas and Servillo (2018) The counterweight: what the WHO checklist does and does not deliver in general surgery. https://doi.org/10.1016/j.bja.2018.02.003

📃 Chang and colleagues (2018) Structured analytic techniques treat two-sided biases as one-sided, and decomposition may add noise rather than remove it. https://doi.org/10.1080/02684527.2017.1400230

📃 Coulthart (2017) An evidence-based evaluation of twelve core structured analytic techniques. https://doi.org/10.1080/08850607.2016.1230706

📃 Spengler and colleagues (2009) Seventy-five studies, 4,607 clinicians. Experience improved accuracy by about thirteen per cent. https://doi.org/10.1177/0011000006295149

📃 Einhorn and Hogarth (1975) Treating your criteria as equally important does surprisingly well. https://doi.org/10.1016/0030-5073(75)90044-6

📃 Fredrickson (1984) Comprehensiveness in strategic decisions, and the warning that thoroughness can cost in an unstable industry. https://doi.org/10.2307/256039

📃 Korsgaard, Schweiger and Sapienza (1995) Intact management teams. Being heard made the process feel fair, and fairness carried commitment and trust. https://doi.org/10.2307/256728

📃 Amason (1996) Two samples of top teams. Conflict's consistent and contradictory effects on decision quality and acceptance. https://doi.org/10.2307/256633

📃 Dooley, Fryxell and Judge (2000) Sixty-eight teams. Consensus built commitment, commitment built success, and commitment slowed things down. https://doi.org/10.1177/014920630002600609

📃 Kim and Mauborgne (1998) Fair process buys voluntary cooperation; unfair process buys hoarded ideas and dragged feet. https://doi.org/10.1002/(sici)1097-0266(199804)19:4<323::aid-smj976>3.0.co;2-f

📃 Villado and Arthur (2013) 188 people in forty-seven teams. Reviewing afterwards raised performance, and memory-based reviews did as well as recordings. https://doi.org/10.1037/a0031510

📃 Keiser and Arthur (2022) What makes an after-action review work better or worse. https://doi.org/10.1007/s10869-021-09784-x

📃 Kamin and Rachlinski (1995) Three quarters called the risk too small beforehand; a majority called the same call negligent afterwards. https://doi.org/10.1007/bf01499075

📃 Guilbault and colleagues (2004) Ninety-five studies. Hindsight bias is moderate, and attempts to talk people out of it did not work. https://doi.org/10.1080/01973533.2004.9646399

📃 De Wit, Greer and Jehn (2012) 116 studies, 8,880 groups. Task conflict looked best where it stayed away from the personal. https://doi.org/10.1037/a0024844

📃 De Wit, Jehn and Scheepers (2013) Task conflict, information processing and the damage relationship conflict does to both. https://doi.org/10.1016/j.obhdp.2013.07.002

P.P.S. Why Do I Even Have the Nerve to Write This?

Fair question.

I am someone who has spent years working alongside remarkable executives, navigating chaos, translating vision into structure and figuring things out, sometimes beautifully and sometimes the very hard way. I have had the privilege of being the steady right hand to leaders who move fast, take risks and expect a lot. That environment has shaped me more than any classroom ever could.

I also happen to have an MBA, and I am studying psychology because people fascinate me. How we work. Why we disconnect. What actually holds us together when the pressure rises.

But really, none of that is the point.

I am here because I have lived the erosion and rebuilt from it. I have seen the cost of carrying too much and the relief that arrives when you finally stop disappearing inside your own competence.

So I share what I have learned in case it helps someone else. Take what is useful. Leave the rest. And remember this one truth that professionals like you often forget:

You are probably doing far better than you give yourself credit for.

The One-Page Decision Brief: What the Research Says About Writing Something an Executive Can Decide From.

What you will get out of this

What the first thirty seconds decide

How you lay the options out is part of the decision

A recommendation, a menu, or both

How many options belong on one page

Saying how sure you are

Putting the objection in the paper

What a structure actually buys you

The record that stops it being reopened

The one page, line by line

The brief check

Where to start

Free tools

P.S. Want to Go Deeper on Decision Briefs?

P.P.S. Why Do I Even Have the Nerve to Write This?

Recent Posts

Comments