Prospeo Case Study: 3,968 Credits, 948 Verified Contacts

Contact data is usually bought on a price per record and judged on a bounce rate. Both are the wrong measure. What decides the real cost of a list is which API calls bill and which do not, because that determines how much qualification you can afford before committing money to a contact. This is a record of three Provena builds on the Prospeo API between June and September 2026: 3,968 credits spent, 948 verified decision makers kept, and the specific mistakes that cost the most.

We are a paying customer, not a partner or a reseller. Nothing here was reviewed by Prospeo before publication and no part of it is sponsored. Every number comes from our own runs, and the endpoint behaviour was verified live against our account rather than read from documentation, which in one important case was out of date.

Which calls actually cost credits

The pricing question that matters is not what a credit costs. It is where credits are consumed, because everything free can be done across the whole market and everything that bills has to be rationed. We verified the following live on 29 June and again on 8 July 2026.

Person search is free, and so is paginating it. A search that returns results costs nothing, which contradicted both our own earlier notes and the version of the documentation we had been working from. Company search bills one credit per page. Bulk enrichment charges one credit for each matched verified email, in batches of fifty; a contact it cannot match is free, and re-enriching a contact already paid for is free.

That single fact shapes the whole method. You can search the addressable market, page through it and read what comes back without spending anything. Credits are committed only at the last step, on contacts already chosen.

What the free preview already tells you

The free search response is not a stub. It returns the full company object, including employee count, employee range, industry, revenue range, funding and job postings. It returns the contact email verification status and the receiving mail provider. It returns a masked mobile object carrying the number status and country code without revealing the number itself.

That is enough for a complete qualification pass at zero cost. We gate company budget on headcount and revenue, gate the role on seniority and department, drop anyone whose email is not already marked verified, and rank the survivors by mail provider before revealing anything. On one build a technology filter applied at query time moved the on-target rate from roughly 5.6% to roughly 95%, and about 94% of that audience independently resolved to Google-hosted mail when we checked the DNS ourselves.

Mail provider is not a detail. Our dealer deliverability benchmark measured a fifteenfold difference in reply rate between two receiving gateways across 43,302 sends of the same campaign. Seeing the provider before paying for the address lets the list be built around that finding instead of discovering it afterwards. We run that gating as part of our outbound service, and the outcomes it produces are published with their limits in our client results.

The three builds, and what each cost

The first build, in June, sourced net-new decision makers at vendors we already knew: 306 verified contacts for 379 credits, or 1.24 credits each. The second expanded that universe outward through competitor mining and themed discovery across five waves and returned 301 verified contacts at 116 companies for 475 credits, or 1.58 each. All 301 came back on Google-hosted mail, which was an outcome of the gating rather than luck.

The third build looks like an outlier and is not. It returned 341 contacts for 3,114 credits, 9.13 each, because 329 of them carry a verified United States mobile number. A verified email costs one credit. A mobile costs close to ten. That one decision is a sixfold difference in the cost of a list, and it deserves to be made deliberately rather than by leaving a flag switched on.

Across the three builds: 3,968 credits, 948 verified contacts, an average of 4.19 credits each that is really two different products averaged together.

The two mistakes worth publishing

The first cost credits. Targeting companies by name looks like the obvious approach and it is a trap. If the name supplied is not the exact canonical form the provider holds, the filter does not raise an error, it drops silently, and the search returns an unfiltered population. In our case that was roughly 9.3 million people. One vendor name failed where the same name with its legal suffix worked. We now target companies by domain, which is deterministic, and every search carries a guard that aborts when the result count exceeds one hundred thousand, on the reasoning that no correctly filtered query of ours should ever be that large.

The second cost accuracy. An early automated quality gate rejected legitimate vendors because words like fleet, RV and marketing appear as feature descriptions inside perfectly good dealer software listings. It removed real companies before a human saw them. A keyword exclusion applied to a company description is a blunt instrument, and it needs a known-good seed list it is not permitted to overrule.

The limits we ran into

A single query returns at most 25,000 results, which constrains any attempt to enumerate a large market in one pass. We partition the query, usually by state, and paginate each partition; because search is free, the extra requests cost time rather than money.

Mobile coverage on software decision makers is thinner than on operational roles. A separate pull in September produced 425 contacts with a real direct or mobile number across 249 automotive software vendors, and reaching that required combining the API with sources we already held rather than relying on it alone.

Documentation and live behaviour disagreed in several places during these builds, including on the central question of whether search bills. Anyone building against this API should verify endpoint behaviour against their own account and treat the account information endpoint as the only reliable source of a credit balance.

What we would tell another team

Design the pipeline around the billing boundary. Discover, qualify, deduplicate and rank on free calls until the only remaining decision is whether a specific contact is worth one credit. Budget mobile numbers as a separate line rather than as an option on a reveal, because they are an order of magnitude more expensive. Target companies by domain and never by name. Put a hard ceiling on result counts so a silently unfiltered query cannot spend money before anyone notices. The rest of this thinking sits in the outbound fundamentals hub.

Our verdict after 3,968 credits is that the free-search, paid-reveal model is the right shape for disciplined outbound, because it makes careful targeting cheaper rather than more expensive. The risk was never the price of a contact. It is building a pipeline that pays before it qualifies.

Method and limits

Three builds between 19 June and 9 September 2026 on a single Prospeo account. Credit figures come from run logs and are reconciled against the account information endpoint. Endpoint costs were verified live on 29 June and 8 July 2026 and may change; check them against your own account before relying on them. Contact counts are unique verified contacts retained after deduplication against lists we already held, not raw API matches. The fifteenfold gateway reply-rate difference comes from a separate Provena study of 43,302 sends and is cited rather than reproduced here. We are an ordinary paying customer; this study was not sponsored, reviewed or approved by the vendor.

Sources

Provena Prospeo credit economics record, three builds, June to September 2026. Vendor reference: Prospeo API documentation. Companion Provena study: where dealer cold email dies, 43,302 sends by mail gateway.

Frequently asked questions