How to Build an Insight Function in a Small Company (2026)

An insight function is the standing capability a company uses to gather, interpret and act on evidence about its customers and market — a defined remit, a recurring research cadence, an owned backlog and a shared store of findings. It is not a person who runs the occasional survey. How to build an insight function in a small company comes down to six moves, from naming the decisions to naming an owner, and the whole thing can be stood up in about 90 days.

Most small companies do have customer insight, they just don’t have a function. A founder chats with three users after a churn spike. Sales listens to objection after objection on calls. Support reads the same ticket theme nine times before someone writes it down. All of it is useful and all of it evaporates.

What follows is the version of the build that survives contact with a real company — a 15-person software shop or a 120-person consumer brand — where nobody has spare budget and the founder is the final decision maker. It’s written for the person who will own the function, usually an insights manager, a product lead or an operations person who gets handed the job on a Friday.

What You Need

You need six things before you start. Miss one and the function will quietly die within two quarters, usually because demand eats the calendar or because findings land with nobody accountable for acting on them.

Budget: what each team shape actually costs

At under 50 employees, realistic spend runs from near zero to roughly the cost of one mid-level hire, before tooling. The bands below are the ones I would plan against. Everything is a range because panel costs, agency day rates and salary bands move, and any figure you see quoted elsewhere is a starting point rather than a quote.

Team shape	Indicative cost	Time to decision-quality output	Best when
Founder-led	Under 10k in tooling and incentive payments	Days, ad hoc	Under about 20 people; decisions are still intuitive
Insights team of one	Salary plus 10-20k a year of fieldwork and panel spend	1-3 weeks per study	20-150 people; demand is steady and commercial
Fractional researcher	Monthly retainer, usually a fixed number of days a month	1-2 weeks, capacity-limited	You need 2-5 days a month of senior method, not a full-time seat
First dedicated hire	Salary plus the same fieldwork budget	2-4 weeks once ramped	Insight is a board-level or product-line priority
Agency	Per project, quoted against scope	3-8 weeks	A one-off study, tracker or segmentation you will own yourself afterwards

Skills

Interviewing and survey design, enough statistics to know when a segment is real and when it is an artefact of the base size, and the ability to write a one-page readout a product lead will actually read. If you are missing statistics, that is the easiest gap to plug — but plug it before the research, not during it.

Tooling

A transcript tool, a survey platform with basic quotas, a repository (a structured folder plus a search tool is fine to begin), a booking link for recruiting participants, and whatever analytics you already use. Vendor neutrality matters here: the platform is replaceable in a day, the method is not. If a piece of software ends up being the main thing the function delivers, you have built a reporting function by accident.

Data access

You need read access to product analytics, support ticket exports, CRM fields, billing or subscription data, and the sales call recordings if they exist. Ask for it in writing during the first fortnight. Access to first-party data is what lets a small team compete with a much larger one, because it is the one evidence stream nobody else has.

Decision rights and leadership support

Without a named executive sponsor who will say out loud that findings will be heard, this does not work. You also need a written answer to the question everyone asks in month four: who is accountable for acting on a finding, and by when. If no one is, the insight function becomes a reports department.

A readiness checklist

Small companies tend to be ready to build when three things are true at once: decisions now outpace intuition, the customer base has grown or shifted so the team’s picture of it is stale, and there is at least one leader willing to fund a partial answer. If only one of the three is true, a one-off study will serve you better than a function.

Step-by-Step: How to Build an Insight Function in a Small Company

The sequence matters more than the detail. Each step creates the condition for the next, and skipping the early ones is the reason most small-company insight efforts stall inside a quarter.

Step 1: Start with the decisions the business needs to make

Write down the eight to twelve commercial decisions the company will make in the next two quarters. Not topics like “customer satisfaction” — decisions like whether to launch tiered pricing by Q2, which segment to target for the spring campaign, whether to build the scheduling feature or the reporting feature.

For each one, name a single owner with the authority to make it. Research that has no owner attached is entertainment. This step is also where you set the standing rule that the function works backwards from a decision rather than forwards from a method, which is the difference between an insight function and a research function.

How do you know it worked? Every item on your list can be traced to one named person and one date. If any cannot, the list is too vague and you need another pass.

Step 2: Define a lean insight remit and service model

Write a one-page charter and get the executive sponsor to sign it. It should cover four things: what the function owns, what it explicitly does not, how demand reaches it, and what it delivers.

The “does not” line is the part founders skip and regret. A useful charter says things like: not primary ownership of the product analytics roadmap, not campaign execution, not weekly reporting on demand. Boundaries written down are what protect the strategic work when a stakeholder with a loud voice arrives in month three.

Then set the service model. State intake channels (one form, one link), the response time for triage (say five working days), the standing turnaround for each class of request, and the fixed output formats — a one-page readout, a repository entry, a decision log update. Small functions survive on predictability far more than on volume.

Inside the charter, define the four types of insight the function will produce, because the categories change what you collect:

• Fact insight — a measured description of what is happening. Usage rates, ticket volumes, churn by cohort, funnel drop-off.
• Explanation insight — the motivation behind the fact. Why churn happens, what triggers a purchase, what stops a trial.
• Predictive insight — what the pattern implies next, given what is already known about the base.
• Prescriptive insight — what should change, with the trade-off stated. The only type that closes a decision.

Most small-company research stops at the first two and then argues about the third. Design for the fourth.

Step 3: Choose a small-core team and flexible support network

Almost nobody needs more than one person at the start, and one person carries five roles at once: recruiting, fieldwork, analysis, reporting and stakeholder management. Plan for that openly rather than pretending the calendar holds.

Shape	Strength	Weakness	Where to find it
Founder-led	Fast, direct access to customers, no hiring delay	Stops the moment the founder is pulled into a fire; no institutional memory	Inside; usually the founder plus whoever owns support
Team of one	Continuity, owned backlog, relationships build	Becomes the bottleneck; scope creep arrives fast	Recruit for method, screen for stakeholder handling
Fractional researcher	Senior method without a full salary; scales up and down	Low availability, no institutional memory unless you build it	Independent researchers, boutique agencies, alumni networks
First dedicated hire	Full capacity, credibility with stakeholders	Fixed cost before the backlog justifies it	Specialist job boards, research community referrals

Whatever the shape, surround it with a support network: a panel or participant community for recruiting, a specialist partner for quant or for accessibility and usability work, and a rotation of internal stakeholders who sit in on readouts so the knowledge does not live in one head.

If you are hiring the first person, screen for three things in order: running a research conversation without leading the participant, writing a tight one-page readout, and saying no to a senior stakeholder. Method knowledge is trainable. Those three are much harder to build, and people who have built these functions consistently point out that the first hire’s real job is intake and stakeholder management, not fieldwork.

Step 4: Create a mixed-method research toolkit

Most small companies over-invest in one method. Qualitative explains, quantitative sizes, behavioural data confirms. Your toolkit needs at least one of each, and the sequencing rule is simple: go qualitative to find out what to ask, then quantitative to find out how much of it is true.

Research question	Method	Indicative cost	Typical turnaround	Sample guidance
Why are customers leaving?	Cancelled-account interviews	Almost none beyond incentives	1-2 weeks	Every cancellation in the last 90 days
What do customers actually do?	Diary study or usage and attitude study	Low to mid five figures	3-6 weeks	Match your customer base, not the panel
Will they buy this?	Concept test with forced ranking	Mid five figures	3-4 weeks	Size the base first; 100 same-profile customers beats 400 from a panel
Which of these messages lands?	Claims or messaging test	Mid five figures	2-3 weeks	Recruit by ideal customer profile, not convenience
Where do they drop off?	Behavioural funnel analysis	None beyond analyst time	Days	All users, not a sample
Can they complete this task?	Moderated usability sessions	Incentives only	1 week	5-8 per segment is usually enough to see the blockers
What is the market doing?	Secondary research and win/loss interviews	Analyst time	2-3 weeks	Not a substitute for primary work

Two constraints belong here. The first is the say-do gap: what people say in a survey and what they actually do rarely match, so treat stated preference as a hypothesis to test rather than a finding. The second is the panel trap. Panel samples skew toward the same demographics as the panel, so a small company recruiting purely from a panel will end up studying people who are not its customers.

What AI can and cannot do in a small-company insight function

It is worth answering this directly, because it is a live question and the honest answer is that AI is useful for parts of the work and useless for the part that matters. It can produce a first transcript summary in minutes, help code an open-ended answer set into themes, draft a discussion guide, suggest research questions you have not thought of, and check analysis code. What it cannot do is recruit a sample that matches your customer base, judge whether a question is biased, decide what a finding means for your business, or hold the accountability conversation when a stakeholder pushes back. Use it to compress the mechanical hours. Keep the judgment hours human.

Step 5: Build a simple operating rhythm

A function is defined by its calendar, not its charter. A rhythm that works at small size runs on three layers.

• Weekly — triage of new requests, one research task in flight, a short written update on what changed because of a past finding.
• Monthly — one readout to leadership, one decision log review, one repository entry filed and tagged.
• Quarterly — a repeated study for comparability (a tracker, a concept test, a churn study), plus a review of which decisions changed.

Quarterly repetition is what turns one-off work into a capability. A single study answers a question; the same study run four times gives you a baseline you can move.

Maturity stage	Defining characteristic	How to tell you are here	What to fix next
Ad hoc	Research happens when someone asks loudly	No backlog, no charter, founder runs everything	Write the charter, name the owner
Defined	A remit exists and requests come through one door	Charter signed, intake form live, owner named	Add the repository so findings survive
Repeatable	A cadence runs whether or not anyone asks	Quarterly tracker running, decision log in use	Start measuring adoption and decisions influenced
Strategic	Insight shapes the plan, not just the tactics	Decisions are visibly traceable to findings	Deepen specialist methods and segment work

You will know you have reached repeatable when a stakeholder mentions a finding from six months ago without being reminded. That is the whole test.

Step 6: Turn findings into insight and action

Analysis is not insight. The gap between them is where small-company research usually dies, because a readout that reports what the data shows is easy to circulate and easy to ignore.

Use a four-part formula for every insight statement: claim plus evidence plus so what plus decision. One worked example. A subscription product’s onboarding survey shows a two-point drop after the billing-details step, and 18 of the 22 customers who abandoned during trial had opened support chat in the same session. The data alone says “billing step loses people”. The insight statement reads: customers who need help at the billing step abandon within the trial, and we treat support contact during onboarding as an early churn signal rather than noise. The decision: move the billing-step help prompt ahead of the card form, and add support chat to the onboarding event list so the team sees it in week one.

Two rules make that stick. Attach an owner and a date to every finding before it leaves the room. Then maintain the decision log — one line per finding, with what was decided and what changed — and review it monthly. Findings without a decision recorded are still not insights.

Build the repository at the same time, because it is what stops the same question being researched twice and what preserves institutional memory when the one researcher leaves. Four fields per entry: the question, the method and sample, the finding in one sentence, and the decision it changed. Tag by topic and by date. A plain structured folder works; anything searchable works better.

Common Mistakes

Almost every stalled insight function I have seen failed for one of the following reasons. Each has a fix that costs nothing but discipline.

Letting demand overload the strategic work

The loudest recurring complaint from people building these functions is that ad-hoc requests consume the entire calendar and the strategic research never happens. Fix it with a standing capacity split, for example 60% planned work and 40% intake, plus a written rule that requests over two weeks of effort need a sponsor conversation before they enter the backlog. Say no with an alternative: “not this quarter, but here is the existing finding that covers part of it.”

Producing findings nobody reopens

The report graveyard is real. Outputs are circulated once, never referenced, and quietly abandoned. Fix it with one repository as the single source of truth and no distribution outside it, plus a monthly “what changed because of research” note that lists last quarter’s findings and their decisions. Repetition in front of leadership is what turns an archived report into institutional memory.

Capturing only the loudest voice

If the founder’s favourite hypothesis shapes every study, you have confirmation bias with a budget. Fix it by writing the research question before the study is designed, asking one question that could disconfirm the standing view, and rotating who chooses the sample frame.

Over-reading small samples

Stated results on 30 people from one segment are directional, not conclusive. Fix it by labelling every finding with its confidence level — strong, directional, or inconclusive — and refusing to act as though the third one is the first. Inconclusive results are a legitimate output and should be reported as such.

Recruiting from the panel and calling it the market

Panels are fast and they skew. Fix it by recruiting first-party customers wherever the question allows, matching on ideal customer profile rather than convenience, and documenting the base so a reader knows how far it can be generalised.

Treating one person as five

At small size, one researcher carries recruiting, fieldwork, analysis, reporting and stakeholder management. Fix it by outsourcing what is commodity — panel management, transcription, basic fieldwork — and protecting time for analysis and readouts, which are the only parts that produce insight.

Skipping the repository

Nothing was written down, so when the one researcher left, every prior finding went with them. Fix it with four fields per entry and a rule that no readout goes out without a repository entry filed first.

Waiting for someone else to define success

A new function with no agreed definition of value gets judged on output volume and cut first. Fix it by defining three measures in the charter: how many decisions were influenced this quarter, how often a finding was reused rather than regenerated, and the share of readouts that produced a recorded action. Volume of studies is not one of them.

Frequently Asked Questions

Conclusion

If you take three things from this, make them these. Write the list of decisions the business faces in the next two quarters and put a name against each one, because an insight function with no decision owner is a reports function. Draft the one-page charter with an explicit “does not own” line and get it signed by an executive who will be heard in the room. Then run one small study end to end — intake form, method, one-page readout, named owner, recorded decision, repository entry — because the first complete cycle teaches the organisation how the function works faster than any charter will.

The test is simple. After two quarters, ask whether anyone cites a finding from six months ago without being prompted. If they do, you have a function. If they do not, you have research, which is a different and much less useful thing.