How to Score ICP Fit Without Hiding Unknowns

Who this framework is for

This framework is for manufacturers, exporters, industrial suppliers, and B2B teams building or reviewing a target-account list. It is especially useful when public data are incomplete, accounts have several facilities, or product fit depends on operating conditions that cannot be inferred from industry labels alone. It does not map the full buying committee; that belongs in the next decision stage.

Contents

1 Define what the score is allowed to decide 2 Use five standardized evidence states 3 Put hard gates before weighted criteria 4 Build the evidence-aware scorecard 5 Calculate fit and coverage separately 6 Translate the output into an action 7 Illustrative example 8 Common mistakes 9 What this does not mean 10 Common questions 11 Sources and methodology

1. Define what the score is allowed to decide

An account-fit score should answer a bounded question: Based on the evidence available on the assessment date, how closely does this account match the defined application, commercial relationship, operating conditions, and service model? It should not claim that the account is in market, that a named person will reply, or that revenue will follow. Industry classification, facility scale, public projects, and company filings can support account research, but none is proof of current buying intent. Define the unit of analysis before scoring. The U.S. Census Bureau explains that an establishment is a single physical location, while a company can consist of one or more establishments. County Business Patterns methodology The Bureau of Labor Statistics similarly publishes QCEW data at establishment, industry, and geography levels. BLS QCEW overview If a product must be installed at a plant, a parent-company employee count cannot substitute for evidence about that plant. Write the score's permitted use in the ICP record. Examples include research prioritization, eligibility for a bounded outreach test, or routing to a technical review. Do not reuse a research score as an automatic billing, qualification, or intent decision.

2. Use five standardized evidence states

SellersGPT recommends the same controlled states for every criterion and gate: Evidence state Meaning Weighted-model treatment confirmed_fit Credible evidence satisfies the written rule for the defined unit and date Include as positive assessable weight when the criterion is compensating confirmed_non_fit Credible evidence fails the written rule for the defined unit and date Include as negative assessable weight when the criterion is compensating contradicted Two or more credible sources materially conflict and the difference is not resolved by scope, unit, or date Exclude from the assessable numerator and denominator; report separately and resolve or escalate unknown Required evidence has not established the claim Exclude from the assessable numerator and denominator; report separately and assign research not_applicable The approved ICP logic makes the field out of scope for this record Optional; remove before renormalizing compensating weights and record the reason "Not found" is not the same as "does not exist." The difference matters when company websites omit facility detail or aggregated official data protect individual businesses. County Business Patterns applies disclosure protection and does not publish individual employer operations. County Business Patterns methodology not_applicable must never be used to hide an unknown observation. Data quality is purpose-specific. The U.S. Government Accountability Office describes reliability in terms of accuracy, completeness, and applicability for the intended purpose, using a risk-based assessment. GAO, Assessing Data Reliability SellersGPT adapts that principle to account research: a source can be credible yet still be too aggregated, stale, or conceptually mismatched for a criterion. For each observation record:

• the exact claim;

• source URL or document reference;

• source owner;

• observed or filed period;

• retrieval date;

• account or facility to which it applies;

• evidence state;

• reviewer note and next action.

3. Put hard gates before weighted criteria

A hard gate is a requirement whose failure makes the account ineligible for the current motion, regardless of its other strengths. Typical gates may include application compatibility, a mandatory product requirement, serviceable geography, minimum order economics, or a prohibited relationship conflict. Do not make a criterion a gate merely because it is important. A gate should meet three tests: 1 Necessity: the commercial motion cannot responsibly proceed without it. 2 Verifiability: the pass rule and acceptable evidence are written. 3 Consequence: a fail or unresolved state changes the next action. Keep gates in a separate register with no point weights. Apply the standardized evidence states first, then derive one of three gate outcomes:

• Pass: every applicable gate is confirmed_fit; a gate may be not_applicable only when the approved logic explicitly permits it and records why.

• Fail: at least one gate is confirmed_non_fit.

• Unresolved: at least one gate is unknown or contradicted, and none has already failed. Fail means exclude from that motion. Unresolved means hold, research, or route to a specialist. Neither outcome can be changed by a weighted score. A gate must never appear in the weighted numerator, denominator, or normalization base.

4. Build the evidence-aware scorecard

Use criteria that explain why the account could buy the offer, not simply what is easy to obtain. First maintain the unweighted gate register: Gate Example question Pass evidence Unresolved action Application compatibility Does the documented application meet every non-negotiable technical requirement? Facility specification, tender, product record, or verified technical review Hold for application evidence or specialist review Serviceability Can the required delivery, commissioning, warranty, and support obligations be met? Approved service policy, location, partner, parts, and response evidence Hold or route for service design Mandatory access Are required registrations, qualifications, or channel conditions satisfied? Official requirement plus account-level evidence Hold for qualification or channel review Entity identity Are the legal entity, facility, and relevant buying unit resolved? Filing, official registry, and first-party location evidence Stop and resolve identity Only after gate evaluation should the team apply a compensating scorecard. A practical example totals 100 points: Compensating criterion Example question Weight Acceptable evidence Relationship fit Is the account's commercial role suited to this motion? 25 First-party business description, partner directory, filing Operating conditions How strongly do documented site conditions support the value case? 25 Specification, permit, facility document, verified call Commercial scale How well do likely order or lifecycle economics meet the target range? 20 Installed-base proxy, line count, project specification Timing evidence How strong and current is the bounded event relevant to purchase timing? 15 Tender, expansion notice, maintenance window Lifecycle potential How well does the documented lifecycle support repeatable value? 15 Platform plan, maintenance policy, installed-base or replacement evidence The weights are an editorial example, not a universal model. The SBA's size standards show why a single easy metric should not dominate: official definitions of "small" vary by industry and may use employees or receipts, including affiliate rules in the applicable program. SBA small business size standards dataset A workforce threshold useful in one segment may be irrelevant in another. For public companies, EDGAR provides filings by company and form. SEC Search Filings A filing may confirm legal identity, segment description, facilities, or risk factors. It still needs an account-and-period citation; public-company evidence should not be generalized to private firms.

5. Calculate fit and coverage separately

For compensating criteria, first remove any approved not_applicable criterion and record its rationale. Renormalize only the remaining compensating criteria to 100. Hard gates never enter this operation. Assessable weight = confirmed_fit weight + confirmed_non_fit weight Fit among assessable evidence = confirmed_fit weight / assessable weight x 100 Then show unresolved exposure against the 100-point compensating model: Evidence coverage = assessable weight / 100 x 100 Unknown exposure = unknown weight / 100 x 100 Contradicted exposure = contradicted weight / 100 x 100 The three exposure measures reconcile: evidence coverage + unknown exposure + contradicted exposure = 100%. Do not add contradicted weight to either fit or non-fit until the credible-source conflict is resolved. Never display a bare score without its denominator and evidence state. Display something like: Gate status: unresolved. Compensating fit among assessable criteria: 52.9%. Evidence coverage: 85%. Unknown exposure: 15%. Contradicted exposure: 0%. Assessment date: August 24, 2026. This makes a well-supported moderate score distinguishable from a high score built on little evidence.

6. Translate the output into an action

SellersGPT recommends decision bands based on gates and evidence coverage, not score alone: Gate status Evidence condition Decision Fail Any Exclude from this ICP motion; retain reason Unresolved Any Hold or research the gate; do not activate Pass Low coverage Research before prioritization Pass Adequate coverage and strong assessable fit Eligible for bounded validation Pass Adequate coverage and weak assessable fit Deprioritize or exclude with recorded reason "Adequate" must be set by the team before reviewing accounts and should reflect decision risk. A low-cost email research test and a distributor appointment do not require the same evidence depth. Every high-impact unknown needs an owner, evidence request, and due date. If the information cannot reasonably be obtained, the team should decide whether to redesign the criterion, accept the uncertainty for a reversible test, or keep the account on hold.

7. Illustrative example

Illustrative example - fictional. Northstar Controls, a fictional exporter of washdown-rated sensors, assesses a fictional food-processing plant. It evaluates four unweighted gates first. Application compatibility, serviceability, and entity identity are confirmed_fit. Mandatory access documentation is unknown. Gate status is therefore Unresolved, and the decision is Hold - resolve access gate before any activation. For research prioritization, the team records a separate 100-point compensating model: Compensating criterion Weight Evidence state Relationship fit 25 confirmed_fit Operating conditions 25 confirmed_non_fit Commercial scale 20 confirmed_fit Timing evidence 15 confirmed_non_fit Lifecycle potential 15 unknown Total 100 The calculation is reproducible:

• confirmed_fit weight = 25 + 20 = 45.

• confirmed_non_fit weight = 25 + 15 = 40.

• Assessable weight = 45 + 40 = 85.

• Fit among assessable evidence = 45 / 85 x 100 = 52.9%.

• Evidence coverage = 85 / 100 x 100 = 85%.

• Unknown exposure = 15 / 100 x 100 = 15%.

• Contradicted exposure = 0 / 100 x 100 = 0%. The compensating result does not alter the unresolved gate. The team assigns a researcher to obtain the applicable facility requirement and proof of compliance. If that gate becomes confirmed_fit, the account can be interpreted under the compensating model; if it becomes confirmed_non_fit, the account is excluded from this motion. The 52.9% result is not an intent score, forecast, qualified lead, or expected conversion rate.

8. Common mistakes

• Scoring unknown as zero: this confuses missing evidence with confirmed_non_fit.

• Giving unknown a neutral midpoint: this creates fictional information and compresses real differences.

• Letting a total override a gate: a strong sum cannot make an incompatible application safe or eligible.

• Scoring the parent instead of the buying site: enterprise scale may hide a non-fitting facility.

• Using employee count as a universal proxy: official size definitions themselves vary by context and industry.

• Treating clicks or replies as fit: activity is not evidence of application, relationship, or operational suitability.

• Leaving sources undated: a correct fact can become stale after a plant sale, expansion, or product change.

• Hiding contradictions: selecting the more convenient source is not resolution.

9. What this does not mean

This framework does not mean every account needs perfect public information before any research contact. It means the decision must match the evidence and remain reversible when uncertainty is material. It does not turn official establishment statistics into account-level demand, designate a universal decision maker, or replace technical, regulatory, commercial, or human review. A score organizes evidence; it does not create evidence.

10. Common questions

Should unknown reduce the fit score? No. unknown should reduce confidence and coverage, not masquerade as confirmed_non_fit. Keep it outside the assessable-fit denominator and display its weight separately. Can we rank accounts with unresolved gates? You may rank them in a research queue by the value and cost of resolving the gate. Do not rank them in an activation queue as if they were eligible. How many criteria should an ICP score contain? Use the smallest set that materially changes the decision. More fields can increase collection work without improving discrimination. Each criterion should have a definition, evidence rule, and action consequence. Can a contact title be a fit criterion? Only as a limited contact-research field after the account ICP is established. A title does not prove that the account fits, and full stakeholder mapping belongs to Buying Committees. How often should scores be refreshed? Set freshness by volatility and consequence. A legal entity or plant status may need event-driven review; a stable process classification may use a longer cadence. Record the observation date for each field rather than assigning one blanket date to the account.

11. Sources and methodology

This article uses official U.S. sources to define data scope, establishment concepts, formal size rules, filing access, and data-reliability principles. The scoring model, evidence states, formulas, gates, and action bands are SellersGPT editorial tools. They are recommendations for transparent B2B research, not official statistical methods. 1 U.S. Government Accountability Office, Assessing Data Reliability (GAO-20-283G). https://www.gao.gov/products/gao-20-283g. Accessed August 24, 2026. 2 U.S. Census Bureau, County Business Patterns Methodology. https://www.census.gov/programs-surveys/cbp/technical-documentation/methodology.html. Accessed August 24, 2026. 3 U.S. Bureau of Labor Statistics, Quarterly Census of Employment and Wages Overview. https://www.bls.gov/cew/overview.htm. Accessed August 24, 2026. 4 U.S. Small Business Administration, Small Business Size Standards Dataset. https://data.sba.gov/dataset/small-business-size-standards. Accessed August 24, 2026. 5 U.S. Securities and Exchange Commission, Search Filings. https://www.sec.gov/search-filings. Accessed August 24, 2026.