Put five people around a table and ask them a difficult question, and the conversation often starts with answers.
One person thinks the number is too high. Another thinks it is too low. Someone sounds confident, someone changes their mind, and eventually the group settles somewhere in the middle.
New research suggests there may be a better way to use the people in the room.
Across three studies involving 900 participants and 1,140 online group discussions, researchers found that groups produced more accurate estimates when they broke complicated questions into smaller parts and worked through those parts together.
The findings, published in Nature Communications on 5 August 2026, suggest that the value of group discussion may depend less on simply collecting different opinions and more on how people reason together.
The crowd is not automatically wise
The idea of the “wisdom of crowds” is familiar.
Ask enough people to estimate something and, under the right conditions, combining their answers can produce a surprisingly good result.
But discussion changes the process.
Once people begin talking, they influence one another. A confident participant can dominate. Groups can converge too quickly around an attractive answer, while useful minority opinions disappear.
The new research investigated whether groups perform better when they do something different: instead of immediately trying to agree on the final answer, they first decompose the problem.
The researchers call the strategy Collective Fermi Estimation.
A difficult question is divided into easier questions. Participants make approximate estimates for those components and then combine them to reach the larger answer.
Instead of asking only, “What is the answer?”, the group begins by asking, “What would we need to know to work it out?”
Three studies tested the same idea
The researchers examined the approach across three studies conducted through online text-chat discussions.
In the first study, involving 500 participants, groups were not instructed to use a particular reasoning strategy. The researchers instead analysed their conversations afterwards.
Groups that spontaneously broke problems into smaller components tended to produce lower estimation errors.
That relationship alone could not show that problem decomposition caused the improvement. Better groups might simply have been more likely to use the technique.
The second study therefore changed the instructions.
Among 240 participants, some groups were explicitly encouraged to break the problem into components, while others were told to discuss and combine their original estimates.
The groups instructed to decompose the problem performed better.
A third study involving 160 participants then asked whether people gained the same advantage by decomposing a problem individually before joining the group.
They did not.
The benefits were greater when the problem was broken apart collectively.
The discussion itself appears to matter
This is an important distinction.
The finding is not simply that complex problems become easier when divided into smaller ones. People have used that principle in mathematics, engineering and planning for centuries.
What the study adds is evidence that doing the decomposition together can improve collective judgement.
Different people may notice different parts of a problem.
One participant might identify a missing assumption. Another may know how one component should be estimated. Someone else may recognise that an apparently minor factor substantially changes the final result.
The group therefore gains something that is harder to achieve when everyone independently completes the same reasoning process and only compares final answers afterwards.
Good discussion, in this view, is not merely a competition between opinions.
It is a way of constructing the problem together.
Why this matters beyond estimation games
Many important social decisions are made in groups.
School governing bodies allocate limited resources. Community organisations plan projects. Municipal committees assess competing priorities. Workplace teams make forecasts. Families make financial decisions together.
These decisions are considerably more complicated than the numerical estimation tasks used in the experiments.
The study therefore does not prove that dividing every political, ethical or community disagreement into smaller questions will automatically produce agreement.
But it does suggest a useful distinction between two kinds of conversation.
One begins with positions.
The other begins with the structure of the problem.
A committee discussing whether a project is affordable, for example, could spend its time arguing over a final figure. Alternatively, it could separately examine the number of people affected, expected costs per person, implementation requirements and possible uncertainties before returning to the larger decision.
The second approach does not remove disagreement.
It may simply make disagreement more useful.
A useful lesson for a divided society
The findings have particular relevance in South Africa, where many public questions quickly become arguments between established positions.
People may disagree about transport, schooling, housing, public spending or community priorities before they have agreed on what smaller questions need to be answered.
Breaking a problem into components cannot resolve conflicts over values or interests.
If two groups fundamentally disagree about what society should prioritise, better arithmetic will not make that disagreement disappear.
But not every disagreement is fundamentally about values.
Some are partly disagreements about assumptions, evidence and how different pieces of a problem fit together.
Making those pieces visible may help groups identify where they actually disagree.
Talking more is not necessarily the answer
There are also limits to the research.
Participants interacted in controlled online text chatrooms, and the tasks involved problems with answers that could ultimately be judged as more or less accurate. Real social decisions frequently involve uncertainty, competing values and outcomes for which there is no single correct answer.
The research therefore should not be read as a formula for solving political conflict or guaranteeing better committee decisions.
Its contribution is narrower.
Groups did not become wiser simply because more people spoke.
They became more accurate when their conversation helped transform one difficult problem into several manageable ones.
That is a useful reminder for any society that spends a great deal of time debating.
Sometimes progress begins not with finding a better answer, but with asking whether the question has been broken down properly in the first place.
Source Information
Study Title: Collective problem decomposition improves the wisdom of deliberative crowds
Authors: Federico Barrera-Lemarchand, Victoria Lescano-Charreau, Julieta Ruiz, Nuria Cáceres, Facundo Carrillo, Mariano Sigman and Joaquin Navajas
Journal: Nature Communications
Published: 5 August 2026
DOI: 10.1038/s41467-026-76365-y
