The Enrichment Cascade: Ordering Sources for Match Rate and Cost

You have a list of contacts and companies to enrich, and a stack of data sources you already pay for. This guide is the ordered method for deciding which source fires first, where verification sits, and how to prove the sequence beats any single source on your own data. It is written for revenue and recruiting operations - the people answerable for how the data was gathered - and it is vendor-neutral, so you can order whatever sources you have rather than being steered to one tool.

The job has three hard parts that vendor posts skip. First, the ordering arithmetic: cost per verified record and blended match rate, not just "put the cheap one first." Second, the own-sample benchmark that exposes headline claims. Third, the per-field provenance and consent trail that keeps a multi-source pipeline lawful. This guide covers all three, in order, with what a good result looks like at each stage.

What an enrichment cascade is and why order matters

An enrichment cascade, also called waterfall data enrichment, is a chain of data sources queried in sequence for each record, where the first source that returns a verified result wins and stops the chain. You pay only the winning provider per record. That single billing rule is what lets a cascade beat any one source on both coverage and cost at the same time.

The gain is real and it is bounded. Published benchmarks converge: a single source typically verifies 40 to 70 percent of a B2B list, while a three to four source waterfall consistently reaches 80 to 95 percent verified match. One vendor reports match improving from roughly 60 percent single-source to above 85 percent across three to four sources. Bounce rates fall in step, from 8 to 15 percent on single-source data to below 3 percent on waterfall-verified data.

But the curve bends like every diminishing-returns curve, because each added source only fills the residual gap the prior sources left. An independent benchmark measured the steps precisely: the best single vendor alone hit 87 percent raw phone coverage but only a 51 percent verified right-person match; a second provider lifted the combined right-person match to roughly 60 percent; a third added about 3 points, a fourth about 2, and by the fifth source the improvement barely registered. The practical ceiling is three to five sources.

That gap between 87 percent coverage and 51 percent verified match is the whole reason this job needs a method. A source can list a field for nearly everyone and still be wrong about who that field belongs to. Order the cascade against the usable number, never the headline.

Match rate by architecture: what the benchmarks actually show

Single-source enrichment verifies 40 to 70 percent of a typical list; a well-ordered three to four source cascade reaches 80 to 95 percent. Below that headline sit stricter independent numbers you should plan against, because they measure right-person verified match rather than raw coverage.

The benchmark rows are "right-person" verified, a stricter test than the vendor coverage rows above them. Treat the 80 to 95 percent band as a best case for clean, well-matched segments and the 51-to-65 percent progression as what a cautious plan should expect on real, aged data. The difference between those two views is not noise; it is the difference between headline coverage and the records you can actually send to.

Read the funnel as a warning against over-stacking. The step from one source to two is worth building for. The step from four to five is orchestration cost and duplicate spend for a gain you cannot measure on any reasonable sample.

The ordering rule: strength first, or cost per valid result

Order the cascade by data strength for the segment you are enriching, not by which provider is cheapest per credit - unless a blended cost-per-valid-result calculation on your own data tells you otherwise. Those are the two defensible positions, and both beat ordering by brand habit.

The strength-first camp reasons that the first provider to return a valid result wins the record, so leading with a weak source means you either never reach the good one or you pay twice for a cheap miss. The cost-first camp reasons that teams often place a premium provider first out of habit, paying top rates on records a cheaper source would have matched anyway, and that you should reorder by blended cost per valid result rather than reputation. Both are right about the mistake they name. The tie-breaker is arithmetic on your sample, not a rule of thumb.

Cost per verified record has no single published closed-form standard. Build it yourself: total provider spend plus verification spend, divided by the number of records that passed your verification threshold. Because you pay only the winning provider per record, a good order raises blended match rate and lowers blended cost against any single source at once. That joint improvement is the deliverable of this whole method.

Cheap misses are the real cost of a badly ordered cascade, not the price on the credit sheet.

Cost and decay by field type

Enrichment cost spans roughly a 10x range by data type, and each field type decays at its own rate. Order and refresh both have to respect this, because the expensive fields are also the ones that rot fastest.

Two decisions fall out of this table. First, resolve cheap, slow-decaying firmographics early and lean on them for the ICP gate before you spend on contact data. Second, treat direct dials as the premium, perishable field they are: they cost about 10x email, cover fewer contacts, and shift roughly 18 percent a year. The payoff for getting them verified is large, because main business lines return only 3 to 5 percent decision-maker connect against 12 to 18 percent for verified decision-maker mobiles.

Verification: the step that earns its own slot

Verification is a stage in the cascade, not an afterthought, and the cascade must stop only when a result passes verification, not when any result is returned. A provider returning a syntactically valid email that bounces is not a successful result; it is a liability that burns your sender reputation.

The consensus verification stack runs four checks and returns a 0 to 100 confidence score rather than a binary pass or fail: syntax, DNS/MX, SMTP handshake, and catch-all detection. The score lets you set your own risk threshold. One verifier bills only for results at 90 percent confidence and above and guarantees no more than 5 percent of billed addresses bounce. Pick a floor like that and gate both billing and CRM writes on it.

Catch-all domains are the main failure zone. An accept-all mail server returns 250 OK for addresses that do not exist, so a naive SMTP check counts them as valid and inflates your match rate. Advanced tools resolve catch-alls at 85 to 95 percent against 40 to 60 percent for standard SMTP. The rule for a catch-all is quarantine, not send and not silently drop.

This is why verification, not the provider stack, protects the domain. An unverified valid-syntax email that bounces damages deliverability more than a contact you never found. The verification layer earns its place in the order as much as any source does.

Build and benchmark the cascade, step by step

Run this procedure end to end for one segment before you scale it. The ordering only holds within a segment, so a UK mobile segment and a US email segment get their own bake-offs and their own chains.

Sizing the bake-off so the ranking holds

The bake-off is where headline claims meet your data, so it has to be powered to rank sources apart. Standard proportion math applies, and with no prior estimate you use the planning value of 50 percent, which is the worst case. For a 95 percent confidence interval of plus or minus 4 points you need about 600 records; for plus or minus 3 points, round up to 1,068. A 100-record test gives roughly plus or minus 10 points at a 50 percent rate - too wide to tell an 80 percent source from a 72 percent one with any confidence. As match rates climb, intervals tighten, so a strong segment can rank on fewer records than a weak one.

Pull the sample the same way for every source: the same 500 to 1,000 ICP-matching records, run through each candidate, scored the same way. Measure email validity, direct-dial connect, field completeness, and how recently each record was updated. That last field matters because a source can be accurate and stale at once.

Refolk removes the friction in the first step of this procedure, the ICP pre-filter, by letting you assemble a clean, in-ICP list in plain English before a single paid enrichment call fires - so you are never spending credits to discover a row was out of segment.

Provenance and consent: keeping a multi-source pipeline lawful

Log which provider resolved each field, how many providers were queried before a result was found, the per-record cost, and the lawful basis and opt-out status. A multi-source pipeline is only auditable if every field is attributable to a source and a basis. Without that trail you cannot answer a deletion request, honor an opt-out at the right source, or defend the pipeline in a review.

B2B enrichment does not sit outside data protection law. Work emails, titles, and direct dials are all personal data under GDPR. The regulation does not prohibit data-driven sales; it requires a clearly documented lawful basis - usually legitimate interest, sometimes consent - that data subjects are informed, and that their rights are respected. The "B2B data is exempt" assumption is the real compliance risk, not the enrichment itself.

Vet provenance at the source, not just per record. Ask where each source's data actually comes from: web scraping, cooperative publisher partnerships, public records, or user submissions. A provider that cannot clearly explain its provenance is a red flag on its own, and it is exactly the source you do not want winning records in an audited chain.

record_id
field: email :: winning_source: <source_id> :: confidence: <0-100> :: providers_queried: <n> :: per_record_cost: <amount>
field: direct_dial :: winning_source: <source_id> :: confidence: <0-100> :: providers_queried: <n> :: per_record_cost: <amount>
lawful_basis: <legitimate_interest | consent>
opt_out_status: <none | received_and_honored>
opt_out_source_notified: <source_id | n/a>
last_verified: <YYYY-MM-DD>

Geography sharpens this. As of the dossier's June 2026 reading, 19 US states have comprehensive privacy laws in effect, and EMEA carries the stricter GDPR burden. In Refolk's index of professional profiles there are only about 38 Revenue Operations title-holders in the United Kingdom against roughly 400 in the United States - about 10.5x more in the US. EMEA cascade design is far less staffed relative to its heavier compliance load, so if you own an EMEA pipeline, the provenance work is both more important and more likely to be sitting on your desk alone.

Refresh cadence: the silent tax on match rate

Refresh cadence, not just initial match rate, sets your steady-state accuracy, because contact data decays continuously between enrichment and first touch. Calibrate the cadence to each field, since the fields decay at very different rates.

The most-cited anchor is 2.1 percent per month, compounding to 22.5 percent a year, from a database decay simulation based on MarketingSherpa research. At that rate, a 10,000-contact list enriched and then sequenced three to six months later has already lost roughly 2,250 contacts before the first message. A weekly re-verification study of 5,000 contacts found real churn far higher, at 2.1 percent per week, compounding to about 67 percent a year. Field-level rates differ again: email decays 23 to 30 percent annually and phone numbers around 18 percent.

The cadence rule follows directly: a quarterly refresh on industry classification is acceptable; a quarterly refresh on direct dials is not. Refreshing every field on one schedule either over-spends on stable firmographics or leaves your most perishable, most expensive fields to rot.

How this goes wrong: failure modes and false positives

Most broken cascades fail in one of eight ways, and every one has a check you can run. This is the section to keep open while you build.

Two of these deserve extra weight. The first is believing a headline match rate over your own data. A vendor's 95 percent is measured on their sample, their ICP, and often their definition of "valid." The only figure that governs your ordering is the one from your bake-off. The second is stopping the cascade on return rather than on verification. A miss costs you a record; an unverified send that bounces costs you deliverability across the whole domain, which is far harder to recover.

Keeping the cascade current

A cascade is not a build-once artifact. Coverage shifts, prices change, and decay quietly erodes the match rate you measured last quarter, so re-run the bake-off and re-evaluate the order every quarter. The order that won on this quarter's sample is a hypothesis about next quarter, not a fact.

Watch three signals between full re-tests. First, blended cost per verified record drifting up means a source is weakening or a price changed - reorder before it compounds. Second, bounce rate creeping past your target means verification is leaking, most often through catch-alls counted as valid. Third, a widening gap between headline coverage and verified match on any source means that source is degrading on right-person accuracy even while its coverage holds.

The audience for this work is small and concentrated. In Refolk's index there are roughly 400 US Revenue Operations title-holders and about 1,129 US Sales Operations title-holders - the people who own cascade design - against only about 38 RevOps title-holders in the UK. If you are one of them, the method above is the difference between a pipeline that beats any single source on your own numbers and one that runs on a vendor's headline and hopes.