Build the research note before writing the research headline

Turn internal knowledge into a defensible story by deciding what the evidence represents, how it was collected and which conclusions remain outside its scope.

The marketing team finds an interesting pattern in customer records. The first draft says the industry is changing. The data analyst points out that the records cover only the company’s customers, and that a recent product launch changed which customers entered the system.

The story may still be useful. It needs a narrower claim and a clearer method before it needs a headline.

A short research note creates that discipline. It records the question, evidence, analysis choices and limitations so the article can be written from a stable foundation.

Start with a question worth answering

Choose a question whose answer could help a reader make a decision. A change in implementation time, the most common reason for a service delay or the conditions associated with successful adoption may be useful.

Avoid questions designed only to place the company’s product at the top of a ranking. If every plausible result must support the same sales message, the exercise is closer to promotion than open investigation.

Write down what would count as a surprising or inconvenient result. That makes it harder to discard evidence simply because it weakens the preferred narrative.

Agree who will make the final analytical judgement before the promotional team becomes attached to a particular chart.

Define the unit and population

Is each record a person, account, transaction, support ticket or project? Multiple records from the same customer may not represent independent experiences.

State who is included, who is excluded and the period covered. Explain whether the data represent all eligible records or a selected sample, and why the selection was made.

Gong’s published research-method description illustrates the value of explaining how operational records become analytical observations. Your process will differ, but readers still need to understand what the dataset can see.

A survey of your newsletter subscribers should not be called a survey of the entire profession. A dataset of completed projects may exclude the projects that failed before completion. Those boundaries can materially change interpretation.

Inspect data quality before searching for a story

Check missing values, duplicates, inconsistent categories and changes in collection. If one office records a field differently, a regional comparison may measure administrative practice rather than customer behaviour.

Keep an unchanged source copy and a record of cleaning decisions. Someone should be able to reconstruct how the published counts were produced without relying on the author’s memory.

Examine whether the data were collected for the purpose now proposed. Operational fields may be convenient but imprecise. A sales-stage label can mean different things across teams or years.

When the underlying measure is weak, say so or choose another question. An elaborate chart cannot improve the meaning of an unreliable field.

Write the narrowest defensible finding

Imagine a hypothetical consultancy reviewing 80 completed projects. Projects with a named customer owner reached an agreed milestone sooner. The observation is interesting, but it does not prove that naming an owner alone caused the difference.

Larger or more organised customers may both assign owners and progress faster. Project complexity, team experience and scope could also matter.

A responsible finding states the observed relationship within the reviewed projects and discusses plausible explanations. The article can recommend examining ownership as a practical hypothesis without promising a universal time saving.

Numbers should retain their denominators and definitions. “Most projects” is less informative than a clearly defined count, but a precise percentage is not necessarily more reliable than the underlying data.

Make uncertainty readable

The ONS publication guidance includes sources, methods and quality information to help readers interpret statistics. A commercial research note can adopt that habit in a shorter form.

Explain material limitations close to the finding they qualify. Do not hide the most important qualification in a distant appendix while the headline makes a broader claim.

Distinguish sampling uncertainty from other problems such as non-response, selection and measurement error. A large sample or narrow confidence interval does not automatically solve those other limitations.

Ask a reviewer unfamiliar with the project to describe what the finding means. If they infer more than the evidence supports, revise the presentation.

Decide what can be shared

Check contractual permissions, confidentiality and privacy requirements before publication. Aggregation does not always prevent identification when groups are small or the circumstances are distinctive.

Use the least detailed information needed to support the useful finding. Where appropriate, publish a method summary and aggregate tables rather than raw customer records.

Give spokespeople the same approved interpretation. A careful report can be undermined by an interview that turns a correlation into a promise or a customer sample into an industry census.

Keep a correction and update route

Record the research date, version and owner. If an error is found, correct the underlying analysis and the derivative assets, including charts, sales slides and social summaries.

For a recurring series, preserve comparable definitions or explain why they changed. A new sample should not silently become evidence of a market trend.

The research note is complete when another informed person can understand the question, follow the method and see the limits. At that point, the headline becomes easier to write because the team knows exactly which story it has earned the right to tell.