Outsourcing an ideal customer profile sounds efficient: give an agency, data provider or sales partner your product, ask for “the best companies to target,” and receive a polished segment deck plus a lead list.

The failure mode is equally simple. The vendor starts with the database fields it already sells, not the buying problem you need to solve. You receive thousands of companies that match industry and headcount filters but have no evidence of need, timing or ability to buy. Marketing calls it an ICP. Sales calls it a bad list.

A useful ICP partner should help turn market evidence into a testable targeting system. Before paying for a large data pull, enrichment contract, outbound program or consulting project, ask the questions below.

Start with one concrete case

Imagine a B2B company sells a $30,000 annual workflow product to multi-location service businesses. Its previous list was “U.S. companies, 50–500 employees, revenue above $10 million.” The list produced meetings, but close rates were weak. A closer look shows that the customers who retained best had three additional traits: several operating locations, a recent process change, and a centralized operations owner.

A weak vendor says, “We can give you 20,000 more companies in that size band.”

A stronger vendor asks, “Can we validate whether those three traits predict better conversion or retention, and can we observe them reliably?”

That difference is the whole assignment.

1. How do you define an ICP, and how is it different from a TAM?

Ask the vendor to explain the difference without slides.

A total addressable market describes the broad universe that might theoretically buy. An ICP is narrower: it identifies the kinds of organizations that are most worth prioritizing for a specific offer and sales motion.

Salesforce's current partner guidance describes an ICP using a combination of firmographic, technographic and behavioral criteria to focus sales and marketing effort. You do not have to adopt any vendor's exact framework, but the distinction between “market size” and “priority account pattern” is essential.

Why ask: if the provider equates “large searchable market” with “good ICP,” expect quantity to dominate quality.

2. Which parts of the profile come from our customer evidence, and which come from your database?

A vendor should be able to label provenance.

Examples:

  • industry: external company data;
  • employee band: external data, modeled or self-reported;
  • number of locations: public records or company website;
  • technology use: observed signal or inferred;
  • recent funding: public announcement or provider feed;
  • high retention: your CRM;
  • painful manual process: discovery notes, survey or call review.

Why ask: ICP work becomes dangerous when inferred fields are presented with the same confidence as verified internal outcomes.

Request a data dictionary with source, refresh date, confidence and missing-data treatment.

3. What customer outcomes are we optimizing for?

Conversion is not the only answer.

Choose the outcome before the profile:

  • first meeting;
  • opportunity creation;
  • close rate;
  • gross margin;
  • implementation success;
  • expansion;
  • retention;
  • payment quality;
  • sales-cycle length.

A segment that books meetings easily but churns quickly can be worse than a smaller segment with slower initial response and strong lifetime economics.

Why ask: vendors tend to optimize what they can measure. If the objective is not explicit, the project drifts toward list size and reply rate.

4. What is the minimum sample needed before we trust a pattern?

ICP projects often overfit a handful of favorite customers.

If five customers share a trait, that is a clue—not proof that the trait predicts success. Ask how the vendor handles small samples, outliers and survivorship bias. Ask whether it will compare good customers with lost, churned or low-margin accounts.

Why ask: studying only winners can turn accidental similarities into “must-have” criteria.

A practical method is to label early criteria as hypothesis, supported, or rejected, rather than pretending the first workshop creates permanent truth.

5. How do you separate fit from timing?

A company can fit perfectly and still have no reason to buy this year.

Ask which signals indicate timing:

  • hiring for a relevant function;
  • opening locations;
  • changing systems;
  • regulatory or operational change;
  • new leadership;
  • funding or budget event;
  • merger;
  • public project;
  • technology migration;
  • explicit research or inbound behavior where lawfully available.

Why ask: firmographic fit tells you “who.” Triggers help answer “why now.”

Do not require every account to have a trigger. Instead, keep fit score and timing signal as separate fields so sales can understand the difference.

6. What exclusion criteria will you build?

Good ICPs contain a “do not target” section.

Examples:

  • too small to support implementation economics;
  • prohibited geography;
  • incompatible technology or operating model;
  • procurement route you cannot serve;
  • customer segment with chronic low retention;
  • direct competitor;
  • existing customer;
  • strategic account already owned by another team.

Why ask: exclusions protect sales capacity. A vendor paid per record has a natural incentive to broaden the list unless the contract rewards quality.

7. Can we inspect a sample before buying the full dataset?

Request 50–100 records or another statistically useful pilot based on the project.

For each field, measure:

  • fill rate;
  • accuracy on a manually checked sample;
  • age;
  • source transparency;
  • duplicate rate;
  • company/entity matching quality;
  • contact relevance if contacts are included.

Why ask: a sales deck about “95% accuracy” is not the same as accuracy on your exact segment and required fields.

Record errors by type. A wrong employee count may be tolerable for one use case; a wrong parent/subsidiary mapping may destroy account ownership.

8. How often are the signals refreshed?

A profile may be strategically correct while the underlying data is stale.

Ask refresh frequency for company size, funding, technology, locations, contacts, job changes and trigger events. Ask how deleted, merged or closed companies are handled.

Why ask: a field that changes monthly should not be sold as though an annual refresh were equivalent to current observation.

Also ask for a timestamp in the output. “Current” is not a date.

9. How will you validate the ICP against sales behavior?

The work should not end when the spreadsheet arrives.

Define a test:

  • select two or three priority segments;
  • build a controlled account sample;
  • use comparable messaging and sales capacity;
  • record connection, meeting, opportunity, win and disqualification;
  • compare by segment;
  • review qualitative objections.

Why ask: ICP is an operating hypothesis. Sales results should be able to falsify it.

McKinsey's 2026 B2B growth research reports a complex buying environment in which decision makers use many channels; the practical implication is that targeting quality should be evaluated across the real sales journey, not only one email metric.

10. Who owns the model after your engagement ends?

Ask whether your company receives:

  • field definitions;
  • scoring logic;
  • exclusions;
  • source notes;
  • query rules;
  • transformation code where relevant;
  • version history;
  • training;
  • the right to modify the model.

Why ask: if the ICP only exists inside the vendor's proprietary dashboard, every adjustment becomes another paid project.

A good partner can use proprietary data while still leaving you with a portable decision framework.

11. How does the output enter our CRM and workflow?

A beautiful profile is useless if reps never see it.

Define:

  • account fields;
  • score or tier;
  • trigger fields;
  • source/date;
  • owner routing;
  • refresh rules;
  • disqualification reason;
  • feedback capture.

Why ask: the ICP is not a PDF; it is a set of decisions inside the sales system.

Make the rep-facing output simple enough to use during prospecting: “Why this account, why now, what evidence, what message.”

12. How do you prevent the score from becoming fake precision?

A score of 87/100 looks scientific even when the weights are arbitrary.

Ask the vendor to show:

  • which fields are hard gates;
  • which fields add evidence;
  • which are optional;
  • how missing data is treated;
  • why each weight exists;
  • what happens when two signals conflict.

Why ask: scoring should help prioritize, not hide uncertainty.

Often three tiers—high, medium, exploratory—with visible reasons are more useful than a complicated decimal score.

13. What compliance and usage limits apply to the data?

Ask where company and contact data come from, what rights permit the intended use, what contractual restrictions apply, and what your organization must do to meet applicable privacy, marketing and sector rules.

The answer depends on geography, data type and channel. A provider saying “our data is compliant” does not remove the buyer's responsibility to use it lawfully.

Why ask: data provenance is not just a legal checkbox; it affects whether the targeting system can be safely operationalized.

14. How will you measure vendor success?

Do not pay only for record volume.

Potential acceptance metrics:

  • agreed sample accuracy;
  • required-field fill rate;
  • duplicate rate;
  • valid entity match;
  • percentage of records meeting hard ICP criteria;
  • documented source and timestamp;
  • pilot conversion quality;
  • handoff documentation.

Why ask: commercial terms shape delivery behavior. If payment is tied to “100,000 records delivered,” the easiest way to succeed is to broaden the definition.

15. What is the smallest pilot that can disprove the idea?

A strong vendor should be willing to learn early.

Pick one product, one geography and two or three competing hypotheses. For example:

  • ICP A: 50–200 employees, 5+ locations, recently hired operations leader;
  • ICP B: 200–1,000 employees, centralized procurement, recent technology change;
  • control: current broad firmographic list.

Run a bounded outreach or sales test and compare meaningful outcomes.

Why ask: the goal of the pilot is not to confirm the consultant's workshop. It is to discover whether the proposed profile improves resource allocation.

A field-by-field acceptance sheet

Before signing, agree on the output.

Field Source Refresh Required? Validation
Industry Provider/public data Quarterly or stated cadence Yes Manual sample
Employee band Provider/public data Dated Usually Compare multiple sources where needed
Locations Website/public/business data Dated If material Spot check
Trigger Event-specific source Event/date Optional Link/evidence
Existing customer outcome CRM Current extract Yes for modeling CRM audit
Exclusion reason Internal rules Versioned Yes Sales ops review

The exact fields change by business. The discipline should not.

Public data is a baseline, not an ICP

U.S. Census business data and BLS QCEW can help estimate how many establishments and jobs exist by industry and geography. That is useful for sanity-checking market claims. It is not a substitute for customer evidence.

Bottom line

The right ICP vendor does not sell certainty. It helps you create a repeatable system for deciding which accounts deserve scarce sales attention.

That system connects internal customer outcomes, external company evidence, timing signals, exclusions, data provenance and a live feedback loop. Start with a small pilot. Demand field-level evidence. Keep the logic portable. And judge the project by better sales allocation—not by how many rows appear in the final CSV.

Sources

Related Reading