ARTICLE · Research

The research report: turning findings into decisions

Boris Kaptelov · 02.08.2026 · 7 min

The research is done. The hundred-and-twenty-slide report sits in a shared folder. Six months later nobody remembers what was in it, and the company commissions a new study on the same question.

Between a finding and a decision there is a step that almost always gets skipped, and the report stops right before it.

A finding and a recommendation are different things

A finding describes what was discovered: customers leave in month nine because the jobs they bought the product for have been done by then, and it does not cover the new ones.

A recommendation describes an action. To turn the first into the second, you have to add what the data does not contain: priority relative to the company's other initiatives, cost, timeline, an owner, a criterion that will show whether it worked, and an estimate of what happens if you do nothing.

The researcher usually does not add this, because they have no data for it. The client does not add it, because they were expecting a ready-made answer. The report stays stuck in that gap.

Hence the practical conclusion: this step should be named as a separate piece of work and built into the project. It is neither free nor automatic.

A structure that forces you to separate the layers

The simplest discipline, which solves half the problem: every substantive block of the report is built on three items.

Finding. What was discovered. One statement.

Evidence. What it rests on: how many cases, which ones, what exactly they showed, whether any contradict it.

Recommendation. What to do, who should do it, by when, and how we will verify it.

Once the third item is separated from the first two, something usually hidden becomes visible: which part of the recommendation follows from the data, and which is the author's managerial judgment. Both elements are legitimate. Blending them is not.

A separate word about numbers in recommendations. A phrase like "we expect retention to rise by fifteen percent" is not a conclusion drawn from in-depth interviews. It is an estimate, and it requires either a quantitative model or an honest note that it is an assumption. Interviews explain the mechanism, not the size of the effect.

One finding, three different documents

The second reason reports fail to work: they are written for everyone at once, which means for no one.

The CEO needs: what decision is being made, what the options are, what we choose and why, what it costs, what the risk is, what happens if we do nothing. One or two pages. Method goes in the appendix, because the CEO is checking the logic of the decision, not the coding.

The product or commercial team needs: the full set of jobs and barriers, the customer's language verbatim, differences between segments, contradictions within the corpus, priorities from which a plan can be assembled. This is exactly where quotes and details belong.

The people who execute need absolute specifics: what changes in their work starting Monday, what to say to the customer, what to do when an exception comes up. No methodology, no segments, no rationale.

These are not three versions of one text at different lengths. They are three documents, because the decisions behind them are different.

What to demand as evidence

The key property of a report worth demanding up front and writing into the contract: the ability to trace a conclusion all the way back to the data.

Concretely, that means three artifacts, and all three have names.

The analytic memo. It records how a category developed: when it appeared, what changed it, which alternative explanations were considered and why they were discarded, where the researcher had doubts. Without it, a conclusion looks as if it had always existed.

The log of analytic decisions. What was decided about the codes, when the coding scheme changed, which fragments were excluded and on what grounds, how the move was made from fragment to claim.

Triangulation. Comparing different sources: the account against the system record, words against observation, an interview against a document. An important caveat: agreement between sources strengthens confidence, while divergence may point not to an error but to a difference in contexts. Triangulation does not automatically turn subjective data into objective fact.

A practical point people forget. If the work was done in a contractor's system, the trail disappears along with the access once the subscription ends. So the contract should specify what is handed over on completion: transcripts, the codebook, the decision log, the recordings. Otherwise, a year later there will be nothing to check.

And the opposite property, the one people fall for most often: polish. A coherent, confident, well-written text reads as proven, especially now, when a coherent report can be assembled from an incomplete foundation. Why this has become a problem in its own right, and what to ask about the analysis method, we cover in our article on AI. What you should check is not how it is written, but whether you can get to the fragment.

What to do with findings that are not about your decision

Every study turns up interesting things unrelated to the original question. They usually go into the report alongside everything else, dilute it, and lower the odds that anyone gets to the main point.

Three buckets.

First: findings that answer the research question. They go into the decision.

Second: findings that change the picture and require verification. They go into the list of hypotheses for the next stage, with a note on which method will test them. A hypothesis with no named way to test it is not a hypothesis but an opinion.

Third: findings that are interesting yet affect no decision in the foreseeable future. They go into the knowledge base and stay out of the report. Their value is that they sit there and turn up in the next study.

No change is not always a failure

A caveat that determines whether the report will be honest or merely convenient.

A well-grounded outcome of a study can be a decision not to invest. Or confirmation that the current course is right and should not change. Or the conclusion that the data is insufficient and a different design is needed.

All three look like failure, both to the contractor who wants to sell a follow-on and to the internal sponsor who defended the research budget. That is exactly how reports end up with recommendations the data cannot support.

A launch cancelled on the strength of research saves the launch budget. That is a result, and it should be presented as one.

How to close the loop

And the last thing, which separates research as a one-off event from research as a process.

Every finding carried through to action generates a testable prediction: if the explanation is correct, this is what will happen after the change. Churn will shift to such-and-such a point. Conversion at this step will rise. The share of customers reaching a second order will change.

Writing the prediction down at the moment of decision takes five minutes. A quarter later it becomes a test of whether the explanation was right. Without that record, any outcome will be explained after the fact, and no one will ever know whether the method works.

That is why we do not hand over the report and walk away. A study after which no decision changed and none was confirmed differs from a failed one only in that someone paid for it.

In practice this is a distinct format of work: carrying findings through to recommendations with owners and deadlines, assembling three documents for three audiences, and setting a check a quarter out. It takes two to three weeks after the study, whether your own or someone else's. And it is also the core of ongoing support when there are many decisions and they need to be carried through the organization.