Intelyza Team, October 20, 2026, 3 min read

A framework for ICP and go-to-market research

A method for structuring ideal customer profile and go-to-market research: hypotheses, evidence sources, scoring and decision rules.

Most go-to-market research fails in the same way. A team collects a large pile of market data, produces a deck of observations, and then argues about what to do. The research was never built to decide anything. This note describes a method that starts from the decisions and works backward. It is a framework for structuring the work. It contains no market findings, and any numbers you gather will be your own.

Start with the decisions

Before reading anything, list the decisions the research must support. Typical ones:

  • Which segment do we sell to first?
  • Which role buys, and which role blocks?
  • What do we say, and which problem do we lead with?
  • Which channel do we use to reach them?
  • What do we do when the offer is not a fit?

Each decision should have an owner and a date. If a finding cannot change a decision, drop the question that produced it.

Write the ICP as hypotheses

An ideal customer profile is a set of testable claims, not a description. Write each as a statement that can be proven wrong:

  • "Companies with multiple legal entities and a manual close feel this pain most."
  • "The finance controller feels the problem, and the CFO signs."
  • "Companies that recently changed ERP are more open to change."

Group hypotheses into four layers: firmographics (industry, size band, geography, systems used), situation (a trigger event, a regulatory change, a recent hire), buying group (who feels it, who approves, who blocks) and fit (what makes success likely or unlikely).

Choose evidence by strength

Not all evidence is equal. Rank your sources and say which hypotheses each can test.

Strongest: behavior. Closed-won and closed-lost records, sales call recordings, product usage, support tickets and what prospects actually asked for. This shows what people did.

Strong: direct conversation. Structured interviews with customers, lost prospects and people who fit the profile but never bought. Interview about recent events, not opinions about the future.

Moderate: market signals. Job postings, technology stack data, public filings, review sites, conference agendas and community threads. These show where a problem exists, but not whether it is urgent.

Weakest: opinion. Surveys of people who do not buy, analyst summaries and internal belief. Use these to generate hypotheses, not to confirm them.

If you have no customers yet, lean on interviews and market signals, and label every conclusion as provisional.

Build a scoring model you can defend

Convert hypotheses into criteria and score each candidate segment or account list. A workable structure:

  1. Pick five to eight criteria, each with a definition and a source.
  2. Mark each as must-have (disqualifies if absent) or weighted.
  3. Score with a simple scale and record the evidence behind each score.
  4. Review the ranking against sales intuition. Where they disagree, find out why, and revise whichever turns out to be wrong.

Keep the model simple enough to explain in a minute. A model nobody understands will not be used by sales.

Test before you scale

Turn the top segment into a small outbound or content test with a defined message, a defined list and a stated success signal, such as replies from the right role or meetings with real budget owners. Define in advance what result would make you drop the segment. Run it long enough to see a pattern, then compare it with the runner-up segment.

Structure the output

A useful research output has five parts: the decisions and their status, the ICP hypotheses with confirmed, rejected or open marked, the scoring model, the evidence log with links to sources, and the next tests. Keep the evidence log. It lets a new team member, or you in six months, see why a call was made.

Common failure modes

  • Confirmation bias. Collecting evidence for the segment you already chose. Assign someone to argue the alternative.
  • Averaging buyers. Merging the user, the approver and the blocker into one persona. Keep them separate.
  • Treating firmographics as the whole ICP. Size and industry matter less than situation and trigger.
  • No decision rule. Research that ends in "interesting" and no change of action.
  • Stale conclusions. Markets move. Date the findings and set a review point.

Where to go from here

Run the method on a single decision first, such as choosing the first segment. Keep the template and reuse it. The value comes from the habit of stating hypotheses, collecting evidence and changing your mind when the evidence says to.

Related reading

AI4 min read

RAG readiness checklist before you build

A checklist to run before building a retrieval-augmented generation system: sources, permissions, evaluation, failure behavior and ownership.

Want help putting this into practice?

Send a few lines about the work. A person reads it and replies within 24 hours.