How to Test Intent Data Coverage Against Your Real ICP Before Buying

Define coverage as usable ICP coverage, not database size

Coverage is the share of eligible entities in a defined market for which a provider can return the evidence required for a specific workflow. The denominator may be accounts, topics, geographies, company sizes, sites, people, contact fields, platform matches, or qualified outcomes. State it. “We cover millions of companies” cannot tell a buyer how many target manufacturers in one region have fresh signals and usable decision-maker data.

Start with a coverage specification: target-account universe; positive and negative fit rules; required countries and company bands; buying roles; essential fields; topic definitions; acceptable signal age; identity-confidence states; permitted uses; destinations; and the business action. Label must-have, useful, and optional fields. This prevents an impressive demo from winning on information the actual workflow does not need.

Use at least three outputs: raw coverage for any returned record, usable coverage for records passing fit, quality, freshness, rights, and field requirements, and activated coverage for records accepted and matched by the intended workflow. For example: usable account coverage = qualified sample accounts with a usable record divided by all eligible sampled accounts. Never quietly remove no-matches from the denominator.

Five coverage tests and the resources each requires

A reliable intent data coverage assessment combines five methods. The first four isolate where coverage succeeds or fails; the fifth tests whether records survive real activation. These are methods, not a vendor list.

1. Stratified ICP sample test

Best fit: Every material procurement. Draw the sample across the dimensions that change commercial value: region, industry, employee or revenue band, technology, account tier, ownership structure, and buying motion. Oversample hard or valuable strata, then weight results back to the real market mix.

Output: coverage and uncertainty by stratum, plus a weighted total. Record no-match, partial, stale, ambiguous, suppressed, and usable as distinct dispositions. Tools: frozen account universe, reproducible random seed, sampling worksheet, unique IDs, and returned-record reconciliation. Meaningful limitation: bad strata, a biased account universe, or undisclosed denominator changes can make precise percentages misleading.

2. Known-account benchmark

Best fit: Checking whether the provider recognizes accounts and facts the company can independently validate. Include customers, qualified opportunities, closed-lost accounts, disqualified accounts, subsidiaries, rebrands, and deliberately difficult records without revealing the expected labels during matching.

Output: entity-resolution agreement, field accuracy, missingness, duplicate rate, hierarchy handling, and false-positive review. Tools: masked gold set, adjudication rules, blind scoring, correction log, and reviewer evidence. Meaningful limitation: known customers and opportunities are usually easier and more visible than the wider addressable market, so this test can overstate real-market coverage.

3. Topic, geography, and company-size matrix

Best fit: Buyers whose market spans several subjects or uneven data environments. Test exact topic meanings, synonyms, exclusions, observation windows, company bands, and geographies. Ask the provider to state whether a row is observed, modeled, aggregated, inferred, or unavailable.

Output: a matrix of raw signal presence, relevant signal presence, fresh signal presence, and sufficient volume by segment. Tools: topic dictionary, expert relevance rubric, negative topics, time-window rules, and ambiguity review. Meaningful limitation: topic breadth and record volume do not prove that activity is accurate, recent, attributable to a target account, or useful for a decision.

4. Usable identity and match test

Best fit: Workflows requiring person records, business contacts, website/company resolution, CRM enrichment, ad audiences, or downstream integrations. Score account match, person match, role fit, employment recency, required-field completeness, contact validation, permitted use, and suppression independently.

Output: a match ladder from sampled accounts to accepted identities, valid fields, eligible actions, and correct destinations. Tools: field dictionary, identity-confidence states, validation provider, suppression list, rights ledger, and manual adjudication queue. Meaningful limitation: a technical match is not automatically the right person, a current employee, a permitted action, or evidence that the person researched a topic.

5. Activation-reconciliation pilot

Best fit: Finalists that passed offline testing. Route a bounded cohort through the real CRM, audience, reporting, or rep workflow with approvals, rollback, and a comparison group. Preserve the provider row ID so rejections, duplicates, actions, corrections, and outcomes can be traced.

Output: activation acceptance, match, action, adoption, qualified outcome, cost, and failure reasons. Tools: sandbox or limited production path, mapping specification, destination diagnostics, SLA log, outcome taxonomy, and experiment brief. Meaningful limitation: a short pilot may prove usability and expose defects but miss seasonality, long sales cycles, model drift, and revenue effects.

Coverage scorecard and denominator rules

Build a scorecard that does not let one broad metric hide a critical gap. Report account coverage, relevant-topic coverage, fresh-signal coverage, person or role coverage, required-field completeness, validation acceptance, permitted-use yield, destination acceptance, match, duplicate and correction rate, and qualified outcome separately. Add a pass/fail threshold for each must-have stratum before seeing vendor results.

• Raw account coverage = accounts with any returned record ÷ all eligible sample accounts.
• Usable account coverage = accounts passing fit, freshness, quality, required fields, rights, and suppression ÷ all eligible sample accounts.
• Topic relevance rate = adjudicated relevant signals ÷ all reviewed returned signals, including ambiguous signals in the denominator.
• Activation yield = records accepted or matched by the intended destination ÷ all eligible submitted records.
• Coverage-adjusted cost = total operating cost ÷ usable or activated entities, with the entity and time window named.

Show numerator, denominator, missing data, exclusions, confidence interval or uncertainty note, sample design, and field version beside every percentage. Report the weighted market estimate and the unweighted strata. A strong total can conceal zero coverage in a priority country or enterprise band.

Set a decision rule for partial coverage. A provider may pass the core market while failing a smaller strategic segment. The buyer can narrow scope, add a complementary source, keep manual research for the gap, negotiate a correction milestone, or reject the offer. Record the added operating cost of every workaround; otherwise an apparent coverage win becomes a hidden integration and labor problem.

Example of a defensible test design

Suppose a company sells two B2B products across three regions and serves both mid-market and enterprise accounts. The buyer should not draw one convenience sample. Build twelve core strata from product fit, region, and company band, then add separate hard-case sets for subsidiaries, rebrands, private companies, sparse industries, and excluded accounts. Freeze unique account IDs and the market weight of each stratum before any provider sees the file.

Ask every provider to return the same fields: input ID, resolved account, parent and domain, topic and semantics, observation window, score meaning, source or provenance class, identity state, required business fields, validation state, permitted-use metadata, no-match reason, and price/usage unit. Prohibit silent removal of input rows. The analyst can then report raw, relevant, fresh, usable, and activated coverage without mixing different products.

Adjudicate a blind sample of positives, negatives, ambiguities, and no-matches. Two reviewers should apply the same rubric, discuss disagreements, and record the final reason. This is not a claim that manual reviewers reveal absolute truth; it creates a transparent reference point and exposes topic or entity rules that need clarification. Keep a correction set to test whether the provider and internal workflow can accept feedback.

Finally, test one bounded destination. If the source claims strong coverage but the CRM rejects fields, the ad platform cannot match enough accounts, sellers decline the records, or permissions block the intended action, procurement should use activated coverage – not the raw feed – as the economic denominator. Preserve the negative findings; they are often the most valuable output of the assessment.

Run the assessment as a controlled procurement workflow

Procurement should own the evidence request and commercial normalization; RevOps or data operations should freeze the account universe and mappings; marketing defines topics and actions; sales validates roles and usefulness; analytics owns sampling and measurement; security, privacy, and counsel review the actual data flow; an executive decision owner sets thresholds and accepts exceptions.

1. Write the workflow decision and coverage specification before contacting finalists.
2. Freeze the real-ICP universe, segment mix, unique identifiers, sample, known positives, known negatives, and hard cases.
3. Send the same input schema, instructions, observation window, deadline, and prohibited enrichment hints to every provider.
4. Require row-level return codes, provenance, time semantics, identity state, missing-value meaning, rights metadata, and pricing assumptions.
5. Reconcile all input rows, including no-match and rejected rows; adjudicate a blind sample of positive, ambiguous, and negative outputs.
6. Test required fields, freshness, duplicates, corrections, suppressions, and destination mappings before an activation pilot.
7. Run a bounded activation, document operational labor and exceptions, and capture qualified outcomes without claiming causation.
8. Apply the precommitted thresholds by stratum, perform sensitivity analysis, document tradeoffs, and request a scope-matched written quote.

Useful resources are a sampling workbook, field/data dictionary, topic rubric, gold-set adjudication guide, provider return template, no-match taxonomy, coverage matrix, rights/security questionnaire, activation checklist, TCO calculator, scorecard, and decision memo. A spreadsheet can be enough for a small sample; use reproducible code and version control when joins, weighting, or repeated evaluations become material.

Sample testing beats headline counts and generic demos

A stratified sample predicts usable coverage better than a database-size claim because it evaluates the buyer’s denominator. A known-account benchmark is useful for entity resolution but tends to favor visible accounts. A generic demo shows product experience, not coverage. A manual or non-intent approach can be better when the market is tiny, topic demand is sparse, actions are low volume, or the team still needs to learn its ICP.

Compare providers on the same frozen input and required output. Do not allow one to return account signals while another returns validated people and then call the raw row counts equivalent. Normalize granularity, geography, topic, freshness, confidence, field requirements, usage rights, activation, support, term, and internal effort. Quote-required is an acceptable pricing result; guessed ranges are not.

Score the full signal-to-outcome chain

Coverage creates value only through a complete chain. Fit asks whether the entity belongs in the market. Signal quality asks whether the event or topic is relevant. Identity asks what account or person was resolved and at what confidence. Freshness asks whether the evidence is timely enough for the action. Activation asks whether the record is eligible and accepted. Outcome asks whether an accountable team took a qualified action.

Keep the stages separate in the dataset. A provider can have broad company coverage and weak person coverage, accurate identity and stale intent, or fresh signals that cannot be activated in the chosen destination. One blended score hides the reason. Use gates for non-negotiable requirements and weights only for real tradeoffs.

Budget and cost for a defensible coverage assessment

Assessment cost includes account-universe preparation, sampling, secure exchange, provider fees or paid proofs, analyst review, domain and hierarchy cleanup, identity validation, topic adjudication, activation setup, destination spend, engineering, procurement, security, privacy/legal review, decision time, and deletion. Budget by complexity rather than a generic percentage of license value.

Estimate internal hours for each role and price the cost of a false positive, missed priority segment, unusable integration, and delayed decision. A small, clean market may be assessed with a few focused strata and a manual review. A multi-region enterprise should fund reproducible sampling, expert adjudication, security diligence, and a controlled activation. Reuse the artifacts at renewal so the first assessment compounds.

How BrandWell should be tested

BrandWell’s separate agency-reseller intent-data offer should face the same coverage assessment as every other provider. Test configurable topic evidence, website identification, LeadFuze-powered identity/enrichment and validation, required fields, geographic and company-size strata, activation paths, tenant separation, and the agency’s real client segments. Do not use the legacy BrandWell SEO writer as evidence for this product.

BrandWell’s public pricing is custom quote. BrandWell agency plans range from $2,500 to $5,000 per month, depending on topic count, term, and available contractually scoped topic exclusivity. The current written quote and Order Form control. The low end is $2,500/month; current product and pricing approval plus the signed scope control. Topic exclusivity or protection is conditional on specific topics, territory, scope, term, and written availability.

If approved for the reseller, BrandWell can be scoped as a complete white-label sales and delivery engine, with agency-controlled retail pricing and end-client billing. Agencies can purchase BrandWell’s $70 seven-day reseller pilot. It includes agency-branded topic reports and the complete sales playbook under the current written pilot terms. Other product capabilities and any topic exclusivity remain subject to their separate current written scope. Agent-ready workflow instructions may be reviewed and run with Claude or ChatGPT; Moxby is a separate optional browser product for direct execution, not part of the BrandWell platform. Before operational use, complete product, pricing, privacy, security, compliance, legal, and platform-policy review.

Who needs a coverage assessment most

Coverage diligence matters most when the addressable market is narrow, several geographies or languages are required, small or private companies are important, identity fields drive outbound, ad-platform matching is essential, data will be resold to clients, prior vendors underdelivered, or a long contract and integration make reversal costly.

A lighter check can be enough when the buyer needs only account-level prioritization, the workflow is reversible, the contract is small and short, or manual research remains practical. A deep assessment cannot fix an undefined ICP, vague topic taxonomy, no destination owner, weak offer, or missing qualified outcome. Correct those before comparing providers.

Connect coverage to activation and pipeline economics

Tie coverage to economics with a ladder: eligible market, usable records, activated records, acted-on records, accepted opportunities, wins, and gross profit. Multiply projected volume by observed usable and activation yields, then apply conservative action and conversion assumptions. Compare the result with the fully loaded program cost and a manual or non-intent baseline.

Coverage ROI is not “rows returned × assumed lead value.” Use a holdout or staged rollout where feasible and distinguish provider coverage from incremental revenue. At procurement, the strongest evidence may be that the source covers priority strata, survives activation, reduces manual research, and produces enough qualified opportunities to justify continued testing. State the uncertainty rather than inventing a benchmark.

Sampling, privacy, security, and data-quality failure modes

Sampling mistakes include vendor-selected accounts, customer-heavy gold sets, too few hard segments, removed no-matches, changing the observation window, duplicates in the denominator, silent enrichment from forbidden sources, and thresholds chosen after results. Data-quality mistakes include stale employment, domain/parent confusion, modeled fields presented as observed, ambiguous topics, and treating a match as accuracy.

Minimize the data exchanged and restrict access. The FTC’s business security guidance recommends collecting only what is needed, controlling access, and overseeing service providers. Use secure transfer, tenant isolation, retention and deletion rules, incident responsibilities, subprocessor review, audit evidence, and a plan for samples that contain personal information.

The NIST Privacy Framework can organize privacy-risk identification and governance, but it is neither legal advice nor a certification. Review notices, lawful basis or consent where applicable, rights, suppression, sensitive categories, cross-border transfers, contract roles, and the proposed activation with qualified counsel. Intent and identity remain probabilistic evidence, not proof of private research or purchase readiness.

Offer coverage audits as a recurring agency service

Agencies can sell an initial coverage audit and a recurring coverage-health service. The initial engagement freezes the ICP, tests providers or sources, validates identity and fields, estimates activation yield, and produces a decision memo. Recurring work monitors drift by segment, source, topic, freshness, match, correction, destination, client, and cost.

A monthly or quarterly client report should include frozen denominators, new and lost coverage, priority-strata gaps, topic changes, identity and validation quality, suppressions, activation acceptance, operational exceptions, qualified outcomes, cost, decisions, and limitations. Price separately for sample design, additional geographies, new destinations, deep adjudication, engineering, and legal/security evidence. Renewal depends on usable coverage and client action – not database volume.

Common questions

How large should the sample be?

Choose sample size from the decision, segment variability, expected rate, acceptable uncertainty, and review budget. Ensure every must-have stratum has enough observations for a useful judgment. A large biased sample is worse than a smaller representative one; document uncertainty when counts are low.

Should the vendor know the positive and negative controls?

Give the schema and evaluation rules, but keep the expected labels blind where practical. This reduces tuning to the answer. Share enough context for lawful, accurate processing and let the provider explain legitimate assumptions in its return.

Is match rate the same as coverage?

No. Match rate normally describes how submitted identifiers mapped within a system. Coverage starts from the real eligible market and includes no-record cases; usable coverage adds fit, freshness, quality, rights, fields, and actionability. Destination match is one downstream stage.

Can a pilot prove revenue ROI?

A bounded pilot can validate coverage, integration, usability, operational effort, and early qualified actions. It may not observe enough independent opportunities or time to establish incremental revenue. Use explicit continuation and stop rules.