Provider A charges less per lookup. Provider B finds more contacts. Provider C advertises a larger credit allowance. Which one produces the cheapest usable result for your sales team?
You cannot answer from the price list alone. The calculation depends on the companies you accept, the fields you request, what gets billed when nothing is found, how much review the result needs and whether the charge has actually settled.
This guide gives you a complete 100-record example, a worksheet and a decision rule for adding another provider. Every euro amount and outcome in the example is hypothetical. They explain the arithmetic; they are not Leadbase prices, vendor quotes or measured provider performance.
What data enrichment costs should include
Data enrichment adds information to a known company or person. The work may include identity resolution, web research, a contact lookup and verification. Those are different activities, even when one interface presents them as one job.
Separate four cost layers:
Keep the last layer separate if you cannot observe it yet. Do not set an unknown cost to zero simply to complete a spreadsheet. Report the measured cost and the unmeasured component.
Define a usable result before comparing prices
For this example, a usable contact must match the intended person, current company and required role, and provide the requested professional contact field with evidence acceptable to the team. The account must already fit the campaign's written criteria.
That definition could differ for your workflow. A company-domain result and a direct phone number need different acceptance rules. Write the rule before testing so it cannot drift toward whichever provider returns the most filled cells.
Use four mutually exclusive final outcomes:
- Accepted: the result meets your written rule.
- Incorrect: evidence contradicts a required element.
- Unresolved: a value was returned, but evidence is insufficient.
- No result: no candidate value was returned.
Track technical failures and cancelled work separately from those completed outcomes. A returned email address is coverage; it is not automatically an accepted contact or permission to use the address for outreach.
If account quality is still unclear, start with the ICP template. Buying more fields for unsuitable companies makes a cost-per-contact metric look productive while wasting budget.
What is waterfall enrichment?
A waterfall queries providers in sequence and uses later steps for records not resolved by earlier steps. Clay's waterfall overview describes this sequential provider approach. The important economic question is how much additional usable yield each step creates.
Stop rules are as important as provider order. Decide whether a returned candidate stops the automatic search, whether it needs verification and what happens when two sources disagree. Do not quietly replace a rejected value with a later answer without preserving why the first was rejected.
The example below passes only no-result records to the next provider. Incorrect and unresolved candidates are quarantined for review rather than automatically retried. This keeps the arithmetic clear; other routing policies require their own measured costs and counts.
A complete 100-record waterfall example
Assume the same 100 qualified people need one professional contact field. All lookups in this fictional example are charged per attempt, including attempts returning nothing. There are no retries, subscriptions or separate verification charges in the direct-usage column.
Provider A returns 60 candidates, leaving 40 no-result records for B. B returns 20, leaving 20 for C. C returns eight, leaving 12 without a result. The final account-level outcomes reconcile: 70 accepted + 8 incorrect + 10 unresolved + 12 no result = 100.
There were 160 lookup attempts, but only 100 original requested records. Mixing those denominators is a common reporting error. For the team's purpose, the waterfall produced 70 usable records from 100 requests.
Coverage and usable yield
The 18 returned but unaccepted candidates are not equivalent to the 12 empty results. They create review work and may contain wrong identities or fields. Report both groups instead of hiding them inside a single “not found” percentage.
Add review labor and fixed costs
Suppose the pilot needs 90 minutes of review at an assumed fully loaded rate of €40 per hour. That is €60. Allocate another €20 of setup or fixed cost to the pilot. These are illustrative assumptions, not suggested market rates.
The headline lookup calculation gives approximately €0.46. The broader observed calculation gives €1.60. Neither number includes unmeasured downstream corrections or predicts revenue. Use the one that matches the decision and label what it contains.
For a subscription, show both cash paid and allocated consumption. If you pay for an allowance you do not use, dividing only by theoretical capacity understates the cost of the actual pilot. Conversely, allocating an entire annual implementation to one tiny test can make a scaling decision misleading. State the allocation period and expected volume.
Is the third provider worth adding?
Provider C adds five accepted records for €12 of direct spend. Its incremental direct cost is €2.40 per additional usable record. That is the relevant number for deciding whether to add C after A and B.
Adding C increases coverage and usable yield, while raising average direct cost. That can still be rational when those remaining accounts matter. Add the extra review and operational work before deciding.
Do not use the waterfall average to claim that C's five records cost €0.46 each. Earlier, cheaper results dilute the average. Likewise, do not assume that reordering the providers preserves these yields. Provider overlap and the remaining case mix change at every step.
Choose provider order from your actual segment
Test the same frozen records and the same acceptance rule. Preserve country, role, company type and the input identifiers each provider receives. A provider given a professional profile URL has a different task from one given only a common name.
Start with enough variation to expose problems: small firms, subsidiaries, local-language roles and companies with sparse public information. A small sample is a screen for workflow defects, not a precise estimate of regional accuracy.
For each proposed order, measure incremental accepted records, direct charges, review time and contradictions. If the second provider mostly repeats what the first already found, a strong standalone coverage number may contribute little to the waterfall.
Keep rejected and unresolved records visible. Reclassifying them as empty can make an aggressive provider look cheap because its identity errors disappear from the scorecard.
A copyable enrichment pilot worksheet
Attach original record IDs to provider attempts so duplicates do not inflate accepted yield. Preserve job or operation IDs when troubleshooting billing. Keep sensitive raw contact details out of a broadly shared scorecard; counts and internal references usually suffice.
Reservations, pending billing and retries
Some systems reserve credits before work starts and settle actual usage later. A reservation is not the final charge. A missing charge while settlement is pending is not zero cost.
Leadbase's credit documentation distinguishes estimates, reservations and settled usage. Research may incur cost before a contact lookup returns no result. For reporting, inspect the execution outcome and billing state together in Activity.
If a request times out after submission, first determine whether it was accepted. An automatic repeat can create another paid operation while the original continues. Retain the request identity and follow the documented retry or support path. Include pending spend explicitly when you cannot yet calculate a final cost.
Common mistakes that make enrichment look cheaper
Dividing by filled cells. Use accepted records for a usable-yield decision. Show coverage as a separate metric.
Comparing incompatible fields. A switchboard number, a general mailbox and a person's direct contact are different deliverables. Use the same requested field and acceptance rule.
Ignoring no-result research. A failed contact lookup can follow billable identity or web research. Measure the whole operation you actually purchased.
Double-counting repeated people. One accepted person returned by three providers is one accepted record. Preserve field-level provenance without multiplying the outcome.
Treating credits as a universal currency. Credit definitions and charge rules vary. Translate them into actual settled cost under the terms of your plan, and show the conversion assumption.
Optimizing spend before fixing the question. If reviewers repeatedly reject results because the wrong entity or role was requested, improve the research question and evidence checks before adding suppliers.
Decide what to enrich next
Enrich fields that can change an action. If a fact will neither qualify an account, resolve a necessary role nor support a permitted handoff, ask why you are buying it.
If account fit is clear but timing is uncertain, use the buying-signal checklist to decide which companies deserve fresh research before paying for more contact fields.
In Leadbase, company and people search, typed research and direct contact discovery address different tasks. Use the current catalog and pricing for live capabilities and terms. This guide does not claim that Leadbase implements the fictional three-provider sequence or offers the hypothetical prices above.
This article is published by Leadbase. Our recommendation is to compare accepted outcomes and complete observed costs, including for our own product. A provider with less coverage can be the better purchase if it yields the right records with less review; a more expensive fallback can be justified for a strategically important remainder.
Your next step: copy the worksheet, test a small representative set and reconcile billing before expanding. If too much spend reaches unsuitable companies, return to the ICP template. If the identities are right but evidence is weak, use the AI research audit.
Sources and product documentation checked on 17 October 2026. All calculation inputs and provider outcomes are hypothetical; no live provider comparison is claimed.
