Two, for almost everyone, and the interesting question is which two. Every provider you add can only reach the records the earlier ones missed, so the second one earns its place and the fourth one is a subscription. Here is the arithmetic, what a credit costs, and a test you can run this week.
By Kshitij Maheshwari, co-founder · Updated August 2026 · 13 min read
One number for "data providers" hides the decision that matters, because a work email and a mobile number are different problems with different economics.
| Data type | Week one | Thousands of records a month | What decides it |
|---|---|---|---|
| Work email | One, plus a verifier | Two: a verified-only primary, plus one fallback for the misses | How much of your list sits outside the US and Western Europe |
| Mobile number | Zero, unless you already dial | Two, and only with a real calling motion behind it | Whether you call at all, then right-person accuracy rather than coverage |
| Company attributes | Whatever your primary includes | One. Company data is far more concentrated than person data | How exotic your filter is. A common firmographic needs no second source |
| Verification | One, always | One, always | Nothing. It is not a provider and it is not optional |
These counts are our recommendation, not a measured benchmark. No published dataset supports one universal number, which is why the 200-row test further down matters more than any table. Every credit price here was read off the vendor's own pricing page in August 2026.
Diminishing returns across providers are a property of set unions rather than a market observation.
The saturation curve is what you get when each provider you add can only contribute the records the earlier ones missed, so every one costs the same and returns less than the last.
Your fourth provider's marginal return cannot exceed your third's. Not usually. Ever.
None of this argues against chaining. A waterfall is how you query several sources, not how many you should own: for the cascade mechanics and the per-record chain cost, see how a waterfall actually chains them. This page answers when to stop adding.
The second provider earns its place. The fourth one is a subscription.
Pick a hit rate per provider and the rungs follow, so read this as arithmetic, not measurement.
The number every vendor quotes you
The rung the arithmetic argues for
Only your weakest field still has room here
Same price as the second, a fraction of the return
Three assumptions run it: each provider finds a fixed share of your list, providers are independent, and each is tried against the whole list. Independence is the generous one: providers resell each other and tend to miss the same records, so every gain above is a ceiling, and your real curve is flatter.
One measurement checks that, and it points the same way. Elric Legloire's 2026 B2B mobile data benchmark, published June 2026, ran 1,400 US contacts, spread across roles, seniority levels and company sizes, through ten providers, confirming the right person by exact name match in IPQS. He funded it himself at $2,521, took no vendor money, and published the raw rows. He also sells an outbound engagement, worth knowing when you read him.
His two strongest single providers hit 51% and 42% alone, and an independence model like the ladder above puts them at 72% together. The best measured pair reached 60%. US mobile data only, so do not carry it onto email or Europe, but it lands under the model, not over it.
Headline plan prices drift every quarter, but per-unit credit costs barely move, and they are what this decision runs on.
LeadMagic, FullEnrich, Findymail and BetterContact all price a work email at 1 credit on their own pages, August 2026, which is why a second email source is cheap to test. See our email finder roundup.
The same pages put a valid mobile at 5 credits (LeadMagic), 10 (FullEnrich, Findymail, BetterContact) and 30 (Datagma) against 1 for an email, August 2026. Both ends meet in LeadMagic against Datagma.
Hunter prices a verification at 0.5 credits against 1 to find an email, LeadMagic 0.25 against 1, and Findymail splits finder and verifier credits into two pools (August 2026).
Clay states that an enrichment returning no result is not charged, LeadMagic advertises zero charge on failed matches, and Dropcontact re-credits a miss (August 2026). See our waterfall tool roundup.
Rollover is capped everywhere: Clay banks twice your monthly credits on Launch and Growth, Findymail twice your allowance, LeadMagic's entry tier nothing (August 2026).
Verification is a line item, so budget for a dedicated email verifier and stop counting it as a provider. Free misses make trying another source almost costless while making buying one feel free.
Every published answer here tracks what its publisher sells, and side by side the pattern is hard to unsee.
"A second provider recovers 15 to 25% of what the first missed. A third recovers 8 to 12%. A fourth adds 3 to 5%."
"1,400 US contacts, ten providers, the right person confirmed by exact name match, raw rows published."
That ladder on the left is the most repeated number in the category. It comes from Unify's own analysis of waterfall enrichment, which publishes no sample size and no method behind it (checked August 2026), and every page repeating it cites Unify or something that does. Naming Unify is not the point. The pattern is.
Hunter's waterfall post, updated July 2026, caps the chain at 2 to 3 because many tools license the same upstream sources. An argument, not a measurement: it cites no data.
ZoomInfo's waterfall page, updated July 2026, calls two to five a reasonable range for most teams. It sells a platform that hits many suppliers for you, so a wide band costs it nothing.
Lantern's waterfall guide says to identify three to five providers with complementary strengths, and sells an agent querying 150+ of them. Its coverage percentages carry no method (August 2026).
Cleanlist's product page is titled "Waterfall Enrichment, 15+ Data Providers, One API", and its answer is fifteen or more. The recommendation and the product are the same number.
We say two. We run outbound as a service and we use Clay extensively, so we buy credits rather than sell them. Read our row the same way as the others.
There is a mechanical reason those answers sit closer together than they look: providers resell each other. BetterContact's own pricing page, checked August 2026, names RocketReach, Apollo, ContactOut, Datagma, People Data Labs, Enrich.so, Prospeo and Hunter among its sources, several of which are sold to you separately. Adding a provider is often buying a second door to the same room.
Coverage always looks like progress, because it only ever goes up. Cost per new record is the number that tells you to stop.
Want your list built on two sources that actually fit your market?
Book a Fit CheckNobody can tell you your number, because it depends on your list, and an afternoon of credits settles it.
Draw them from the list you intend to contact, not your best accounts.
Exact setting
Sample: 200 rows, random draw from the live list
Gotcha: Random matters more than large. A hand-picked sample flatters whoever is best on your favorite accounts.
Record three things per row: found or not, the value returned, the credits spent.
Exact setting
Columns: row_id, A_found, A_value, A_credits
Gotcha: Misses are free at most vendors, so the trial costs a fraction of the subscription.
All 200 rows, not only the misses, because you need to see the overlap.
Exact setting
Columns: B_found, B_value, B_credits, both_found
Gotcha: Running it on the misses alone shows the lift and hides the overlap, which predicts provider three.
On 40 to 50 returned records, confirm by name that it belongs to the person you meant.
Exact setting
Subsample: 40 to 50 returned records, exact-name check
Gotcha: The step everyone skips. A returned record is a claimed match, not a correct one.
Buy whichever candidate recovers most per euro, then test the next against both.
Exact setting
Rank by: unique records added, divided by euros spent
| Record this | How you get it | What it tells you |
|---|---|---|
| Coverage of A | Found by A, over 200 | Your baseline, the least useful of the four |
| Incremental unique of B | Found by B, not by A, over 200 | The only number that decides anything |
| Overlap | Found by both, over found by either | How independent your two sources really are |
| Cost per incremental usable record | B's credits in euros, over B's verified uniques | The input to both stopping rules |
The first stopping rule uses that last number. Stop when the next provider's cost per incremental usable record exceeds what a usable record is worth to you, which comes from your funnel: what you would pay for one more contactable person in your ICP.
The second needs no funnel math. Stop when the gain is too small for your own test to see: 200 rows resolve a few points of gain to within four points either way, so if the test cannot separate it from noise, neither will your reply rate.
Three questions dissolve most of the numbers in this category, and asking them takes seconds.
A second provider recovers 15 to 25% of the records the first one missed.
On what sample, verified how, and what does the publisher sell?
The best-documented dataset here measured US mobile numbers, and its curve gets quoted as though it described enrichment in general. A mobile curve does not predict an email curve, and a US curve does not predict a European one.
Provider count is one call inside the rest of the enrichment job, and the one small teams over-think.
Buy this
One provider, plus a verifier
What decides it
The trade
A second subscription buys more sources per account. More accounts is the better buy.
Buy this
One, then two only once the test says so
What decides it
The trade
Below a few hundred rows a month, researching the misses by hand beats a subscription.
Buy this
Two on the field that matters
What decides it
The trade
Two maintained well beat five maintained badly. Re-test before you renew anything.
Six habits cover almost every over-bought data stack, plus one you cannot undo.
Vendors price verification at a fraction of finding, because it is a different job. Count it as one of your two and you have one real source, not two.
Mobile is the most expensive record here. If nobody dials, the right number is zero and the credits belong on email.
Nominal provider count overstates source diversity. Read a candidate's own documentation for its upstream sources first.
A second source returning twenty rows and getting six wrong has not handed you twenty contacts. Coverage rises either way, which is why it is a bad target.
A provider bought before your first campaign optimises the wrong variable. Learn whether anyone replies first.
Coverage shifts as upstream sources change hands, so last year's winner is this year's dead weight.
Three published terms compound. Misses are free, so trialling feels costless. Rollover is capped or absent, so credits you never work vanish. And entry tiers are discounted for annual commitment, so the year is bought before the test is run.
Co-founder of Real Good GTM. He has been the first business hire and Chief of Staff at seed-stage B2B startups, building outbound pipeline before any playbook existed. This post comes from buying credits rather than selling them, and from re-running this test on client lists every quarter.
Connect on LinkedInWhat comes next: whether to buy at all, how to build the list yourself, and which two tools go in the slots.
The question before this one: whether a bought list belongs in your motion at all.
Read the postThe job around the providers: sourcing, filtering, and the legal ground rules.
Read the guideTwo candidates for slot one on work email, taken apart on credits and coverage.
See the comparisonBook a fit check. We'll look at where your list sits, which field your campaigns are short of, and whether a second data provider is the right next spend at all.
Book a Fit CheckNo hard sell. No fake numbers. Real good work speaks for itself.