A 1P CRM Onboarding With a 31% Match Rate and Runaway Frequency
An auto advertiser onboards a 1.8 million-record CRM list of hashed emails through a data partner for a conquesting campaign. The trader expects a 55% match rate based on the partner's benchmark but sees 31% in platform. The campaign also layers a third-party in-market auto-intender segment of 4 million users, and the two are combined in a single ad group with "Include" logic on both, plus a 1% lookalike built from the CRM list. After three weeks, reach is 1.1 million uniques, frequency is 11, and CPA is $96 against a $60 goal. The client says the CRM data "doesn't work" and wants to drop it.
1. Diagnose the match-rate gap. What would you check on the data itself, and what would you tell the client about whether 31% is actually bad?
Read the worked answer
A benchmark is only meaningful if the list resembles the benchmark's list. The 55% figure almost certainly comes from a different list profile, geography, or identity environment. Before accepting or disputing the 31%:
- Record quality: list age (email decay is material beyond 12–18 months), share of role/B2B addresses, domains with low match propensity, typos, and duplicates. A conquesting list built from dealership inquiries often contains shared household emails and business addresses that match poorly.
- Format and hashing: SHA-256, lowercased, trimmed, no stray punctuation. A formatting error silently drops records before matching.
- Identity environment: is the match measured against UID2 space, cookie space, or a partner's graph? Match rates differ substantially by identifier. If the partner onboards to UID2 but the campaign's inventory is largely cookie-based, the effective match will look low.
- Job status: confirm all 1.8M records were accepted. Some records are typically rejected as invalid, so the denominator may already be smaller than the client thinks.
- Geography: match rates fall sharply if the list spans markets where the identity graph has thin coverage.
What to tell the client: 31% on a large, aged CRM list is plausible and not by itself evidence that the data is broken. What matters is whether the matched portion is large enough to target (roughly 558,000 users - workable) and whether it performs. The real problem in this campaign is not the match rate; it is the targeting structure built on top of it.
2. Explain why the campaign's reach, frequency, and CPA look the way they do. What is the targeting logic doing?
Read the worked answer
The targeting setup is manufacturing the poor results.
- "Include" on both segments means AND, not OR. The ad group is only eligible to bid when a user is in both the CRM-matched audience and the third-party in-market segment. That intersection is a small slice of either audience, which explains why reach stalled at 1.1M and why the campaign is not reaching the broader intender pool.
- The lookalike is built from a weak seed. A 1% lookalike sourced from the matched portion of an aged CRM list inherits whatever bias that seed has. If the seed skews toward existing customers or a narrow geography, the lookalike will too.
- Frequency of 11 is the predictable result of a narrow intersection plus no exclusions. With a small eligible pool and a campaign trying to spend, the same users get hit repeatedly. That drives up cost without adding reach.
- No exclusion of existing customers or recent leads. For a conquesting campaign, this is the most direct waste: paying to reach people who already own the vehicle or have already inquired.
- CPA of $96 against a $60 goal is therefore not a data-quality signal. It is what an over-narrowed, over-frequented, unexcluded audience produces.
Dropping the CRM data would remove the only audience with a genuine first-party signal while leaving the AND logic, the lookalike, and the missing exclusions untouched. The campaign would likely get cheaper per impression and no better per lead.
3. Propose a restructured targeting plan and describe the test that would prove whether the CRM data is adding value.
Read the worked answer
Restructure into separate ad groups so each data source can be measured and controlled.
- Ad group 1 - CRM conquesting: CRM-matched audience only, with existing customers and recent leads excluded. Budget ~40%.
- Ad group 2 - Third-party in-market: the 4M segment with the same exclusions, no CRM restriction. Budget ~35%.
- Ad group 3 - Lookalike: rebuilt from a cleaner seed (recent converters or high-value CRM records), 1–2% expansion. Budget ~15%.
- Ad group 4 - Test/prospecting: open in-market or a second data partner. Budget ~10%.
- Frequency cap: 3–4/day per ad group, with a campaign-level cap to prevent overlap stacking.
- Exclusions: existing customers, prior leads, and recent site visitors who already converted - applied across all ad groups.
Proving CRM value:
- Compare CPA and, more importantly, cost per qualified lead (test drive, dealer visit) between ad group 1 and ad group 2 over a fixed window with comparable budgets.
- Run a holdout within the CRM-matched audience - suppress a random 10% - to measure incremental lift rather than attributed conversions.
- Re-onboard a cleaned, more recent slice of the list, or test a second onboarding partner, and compare match rate and downstream CPA. If a higher match rate does not improve CPA, match rate was never the right KPI.
What would justify dropping the CRM data: ad group 1 matching or underperforming ad group 2 on cost per qualified lead after the restructure, with the holdout showing no incremental lift. That is a data conclusion, not an inference from a match-rate benchmark.