A B2B software company says its ideal customer profile is “companies with 100 to 5,000 employees that want to improve operations.” The total market looks enormous. Marketing can produce thousands of accounts. Sales can always find another prospect. The problem is that nobody knows which accounts deserve attention first.

After two quarters, the symptoms are obvious. Outbound reply rates vary wildly by rep. Discovery calls begin with basic education because some prospects barely recognize the problem. Product demos are customized for industries that use the workflow in completely different ways. Pipeline looks large, but opportunities stall after a first meeting because procurement, IT, finance and the business sponsor do not agree on the value.

The team initially concludes that it needs more leads.

It actually needs a narrower decision system.

This hypothetical case is useful because an ICP is not a market-size statement. It is a prioritization model: a practical answer to which accounts are most likely to have the problem, be able to act, fit the product and justify the cost of selling to them?

Current 2026 B2B research reinforces the complexity around that decision. Gartner describes buying groups, self-directed research and AI-assisted evaluation as important features of modern B2B buying. Salesforce’s 2026 State of Sales research likewise emphasizes data quality and changing sales workflows. Those market signals do not tell a company what its ICP should be. They make a vague ICP more expensive because teams can automate poor targeting faster.

The original ICP failed four tests

The statement “100–5,000 employees, operations improvement” sounds precise because it has a numeric range. Operationally, it fails four tests.

Problem intensity. It does not say which operational problem is painful enough to create a buying event.

Ability to buy. It does not distinguish companies with budget, systems, people and executive sponsorship from companies that merely match headcount.

Product fit. It does not say which processes, integrations, compliance needs or operating models make the product useful.

Cost to serve. It does not account for complex implementation, security review, data migration or support burden.

The result is a market definition pretending to be a sales decision rule.

Branch one: if the account has the right size but no trigger, do not force intent

The first decision-tree branch concerns a common mistake: treating demographic fit as purchase intent.

An 800-person company may fit the headcount range perfectly and still have no reason to change its current workflow. A 220-person company may be in acute pain because it just expanded into three countries, changed ERP systems or acquired another business.

A stronger ICP therefore separates stable fit from timing signals.

Stable fit might include:

  • industry or operating model;
  • process complexity;
  • number of locations;
  • data or integration environment;
  • regulatory or audit burden;
  • economics of the problem; and
  • service requirements.

Timing signals might include:

  • rapid hiring in a relevant function;
  • merger, acquisition or geographic expansion;
  • new executive ownership;
  • system replacement;
  • compliance deadline;
  • repeated operational failure; or
  • a public initiative that creates the problem the product solves.

Cost of ignoring this branch: sellers spend time manufacturing urgency where none exists.

When this branch does not apply: a product with extremely low friction and broad self-service adoption may not require strong trigger-based prioritization. The sales model should match the motion.

Branch two: if the problem is real but the buying group cannot align, change the sales work

The team’s original model focused on one “decision-maker.” Deals stalled because the business sponsor liked the product while IT worried about integration, procurement challenged terms, and finance could not see the economic case.

Gartner’s 2025 research on B2B buying groups reported significant internal conflict among surveyed buyer teams and found that group-level relevance can help consensus. Its 2026 research continues to emphasize buyer self-service, AI-assisted research and the role of sellers in validation. The exact percentages belong to those samples; the operating lesson is broader.

An ICP should therefore include buying-group feasibility, not only company fit.

Ask:

  • Who owns the pain?
  • Who owns technical approval?
  • Who owns budget?
  • Who can veto security, legal or procurement terms?
  • Who benefits but never attends sales meetings?
  • Does the organization have a decision process that the vendor can support?

If a segment consistently requires six months of security work for a $15,000 annual contract, the issue may not be “sales execution.” The segment may be structurally unattractive.

Cost of ignoring this branch: pipeline becomes full of accounts that can like the product but cannot efficiently buy it.

When this branch does not apply: in a simple owner-led small-business purchase, the buying group may be one or two people. Do not import enterprise complexity into a transactional motion.

Branch three: if the customer can buy but implementation destroys margin, redefine “ideal”

The company discovered that one attractive industry required custom integration, data cleansing and high-touch onboarding. Win rates looked reasonable, but gross margin after services was poor.

This is where an ICP must meet unit economics.

The revenue team adds four fields to the segment score:

Factor Good signal Warning signal
Implementation standard configuration custom workflow every deal
Data clean/common systems extensive remediation
Security/legal repeatable review unique terms and controls
Support predictable continuous specialist support

An account is not ideal simply because it will pay the license price.

The team also distinguishes strategic exceptions from accidental exceptions. A lighthouse customer may justify extra implementation because it opens a category, creates a product capability or produces a reference. But that should be an explicit investment decision.

Cost of ignoring this branch: bookings grow while delivery load and churn risk quietly grow faster.

When this branch does not apply: early-stage companies may intentionally learn from difficult customers. The key is to label that work as product discovery rather than pretend it is already a scalable ICP.

Branch four: if a segment is broad, split it by problem pattern before adding more personas

The company’s first reaction to poor targeting was to create 14 buyer personas. That made content production heavier without fixing account selection.

The better move was to split accounts by problem pattern.

For example:

Segment A — scaling complexity. Multi-site growth has broken manual coordination.

Segment B — control requirement. Audit, compliance or governance creates a need for standardized workflows.

Segment C — system transition. A technology or organizational change creates a temporary window to redesign the process.

Segment D — efficiency pressure. Leadership has a quantified cost or cycle-time target.

Each segment can contain several buyer roles. The segment explains why the company might buy; personas explain how different people experience the decision.

Gartner’s March 2026 research on B2B segmentation and enterprise personas explicitly frames modern ICP work around buying groups and changing environments. That is more useful than multiplying role descriptions without a company-level reason to act.

Cost of ignoring this branch: marketing produces more personalized content for accounts that still should not be prioritized.

When this branch does not apply: if the product genuinely serves one homogeneous use case, additional segmentation can create unnecessary complexity.

The team rebuilt the ICP as a scored hypothesis

The new ICP was not a 40-field database filter. It was a small model that could be tested.

The team scored five dimensions from 0 to 2:

  1. Problem fit: no visible problem / possible problem / verified recurring problem.
  2. Operational fit: major mismatch / manageable / strong standard fit.
  3. Buying feasibility: unclear or blocked / possible / known process and sponsor.
  4. Economic fit: poor cost-to-serve / uncertain / attractive.
  5. Timing: no trigger / weak signal / active trigger.

The maximum score is 10. The company does not claim that “8+ means guaranteed buyer.” It uses the score to choose where human research and outreach should go first.

A scoring model is valuable only if the team periodically compares scores with actual outcomes. If high-scoring accounts do not convert, either the model is wrong or the execution is.

The decision tree the SDR team actually used

Instead of a generic list, the SDR workflow became:

1. Is there stable company fit?
No → do not prioritize.
Yes → continue.

2. Is there evidence of the problem or a plausible trigger?
No → nurture/research rather than aggressive outreach.
Yes → continue.

3. Can we identify a credible problem owner?
No → research the buying group.
Yes → continue.

4. Is there a known product or delivery mismatch?
Yes → disqualify or route to strategic review.
No → continue.

5. Is there enough evidence to make a specific outreach hypothesis?
No → gather evidence.
Yes → contact with the hypothesis, not a generic pitch.

This reduced activity volume. It also made the activity easier to learn from.

AI changed the cost of research, not the standard of evidence

The company began using AI to summarize public account information, group signals and draft research briefs. That improved speed.

But the team adopted one rule: AI-generated evidence is a lead to verify, not a fact to score blindly.

That matters in 2026 because buyer and seller workflows increasingly include AI. Gartner reported in May 2026 that surveyed B2B buyers used multiple information sources and that many used generative AI during purchases, while still valuing sales representatives for validation at key points. Salesforce’s State of Sales research also emphasizes data quality as teams introduce agents.

For ICP operations, the implication is simple. Faster research makes data hygiene and source discipline more important, not less.

The CRM therefore stores:

  • the signal;
  • the source URL or internal evidence;
  • the date observed;
  • confidence level;
  • whether it was verified by a person; and
  • the next hypothesis to test.

The team does not store “AI says high intent” as a decision-grade field.

The metric that exposed the old ICP

The original dashboard celebrated leads, meetings and total pipeline. The reset added conversion by ICP score and problem segment.

Within a quarter, the company could see that high-fit accounts did not merely book more meetings; they also moved through technical validation faster and required fewer custom implementation exceptions.

That changed resource allocation.

Marketing stopped buying broad lists simply because they fit headcount. SDRs spent more time on evidence-rich accounts. Account executives received fewer but more coherent opportunities. Product teams saw which integration demands were common enough to standardize.

The ICP became a cross-functional operating model instead of a marketing document.

What not to do with the new ICP

A sharper profile creates its own risks.

Do not freeze it forever. Markets and products change.

Do not turn it into a discriminatory or unlawful proxy. Employment, consumer, credit and other regulated contexts can impose legal limits on data use; B2B targeting also needs appropriate privacy and data-governance practices.

Do not treat a score as truth. It is a hypothesis.

Do not hide strategic bets. If leadership wants to enter a new segment, label the accounts as an experiment rather than manipulating the score until they look “ideal.”

Do not let enrichment vendors become the source of record for facts they cannot substantiate.

And do not optimize so tightly that the company stops learning from adjacent demand.

What changed the outcome

The revenue team originally thought its problem was insufficient top-of-funnel volume. The real problem was that “ICP” described almost everyone who could theoretically use the product.

The reset made four changes:

  • company fit and timing signals were separated;
  • buying-group feasibility became part of qualification;
  • cost-to-serve entered the definition of “ideal”;
  • segments were built around problem patterns, not an explosion of personas.

The result was not a magical list of perfect buyers. It was a more disciplined allocation of expensive seller time.

That is what an ICP should do.

Bottom line

An ICP is useful when it helps a team say not now, not this segment, or not without more evidence as confidently as it says “pursue.”

In this hypothetical, a broad ICP created lots of activity because almost every account qualified. It produced weak learning because the reasons for buying were mixed together.

A stronger ICP is a scored, testable hypothesis about problem, fit, buying feasibility, economics and timing. Pair it with a decision tree, store evidence with dates and sources, and measure outcomes by segment rather than only by volume.

The objective is not to shrink the market on a slide. It is to focus real sales effort where the organization has the best combination of customer value and repeatable economics.

Sources

Related Reading