MAID Graph Economics: Why Identity Resolution Costs What It Costs

Identity-graph pricing looks opaque until you model the inputs: seed freshness, signal density per MAID, match-rate math, cohort decay, and regulatory ring-fencing. A buyer requesting MAID-to-HEM quotes on one million seeds and receiving three prices that differ 3x is seeing those inputs diverge, not random markup. This guide explains why MAID Feed and Core Email File economics behave as they do. Pair with identity graphs 101, device graph decay, and 5 questions before licensing a MAID feed.

Picking lowest unit price without seed match testing usually produces renewal disappointment. The vendor with the highest headline graph may carry the weakest observable cohort for your audience. Flip the denominator to matched, reachable keys and rankings change.

Match-Rate Math and Seed Quality

Match rate equals overlap between your seed and the vendor's observable graph, after normalization and suppression. Fresh CRM emails match better than aged lists; consumer seeds match differently than B2B domains. Vendors price on successful matches, attempted matches, or graph access fees: three incompatible models. Normalize quotes to cost per matched HEM or MAID at your seed's freshness tier.

Signal Density and Graph Depth

Graphs linking MAID to HEM to household to CTV ID, the identifier surfaces covered by IAB Tech Lab addressability standards, carry more compliance and storage cost than MAID-only files. Each additional edge requires consent scope, retention policy, and deletion propagation. Deep graphs command premium when match quality is documented; they are overpriced when edges are probabilistic junk.

Cohort Decay and Renewal Repricing

Graphs lose three to seven percent of MAIDs monthly under Apple App Tracking Transparency, and eight to fifteen percent of active HEMs annually. See device graph decay. Vendors that quote historical cohort size hide effective price inflation. Contract for observable-cohort reporting and renewal true-ups tied to documented composition, not surprise uplifts.

Regulatory Envelope as Cost Driver

FTC consent orders against X-Mode and Outlogic, alongside the FTC order against InMarket, forced sensitive-category scrubbing, consent-chain documentation, and deletion infrastructure: real vendor cost passed through pricing. Cheap graphs without those controls carry downstream risk, not savings. Diligence on data brokers post-FTC orders belongs in identity procurement.

Identity Graph Procurement Framework

Run seed match tests on identical files across finalists. Record match rate, matched-key cost, refresh cadence, deterministic versus probabilistic tier pricing, and deletion SLAs benchmarked against the IAPP US State Privacy Legislation Tracker. Model three-year TCO with decay assumptions, not year-one license alone. Identity economics become predictable once buyers stop comparing incompatible denominators.

Finance and ad ops should share one scorecard: observable graph size, match rate on standard seed, cost per matched key, refresh SLA, and compliance pass/fail. Vendors that cannot populate the scorecard are selling opacity, not identity resolution.

A Match Rate Is a Fraction, and Vendors Choose the Denominator

Match rate is the headline number in identity pricing and the least standardized. The same underlying performance can be quoted very differently depending on what sits below the line: every record you submitted, only the records that passed validation, only those in the vendor’s addressable geography, or only those the vendor considers resolvable at all. Each exclusion raises the quoted rate without changing what you receive.

Fix the denominator before collecting quotes and state it in the request, so every vendor answers the same question. The useful denominator is almost always the full file you intend to send, including the records you expect to fail, because that is what determines your cost per usable output. Then compare vendors on that figure alone and treat every other rate as marketing. This is also the fastest way to expose a quote that looked far better than the rest.

Not Every Match Is Worth the Same

Match counts treat every link as equivalent when they are not. A link built from several corroborating observations over months is a different asset from one inferred from a single weak signal, and both are usually delivered as a match. The consequence is that two vendors quoting the same rate can deliver materially different accuracy, and the difference only shows up downstream as wasted spend or misdirected messaging. Confidence tiers in a MAID-to-HEM identity graph make that distinction explicit, but the price comparison still depends on how those tiers perform against the buyer's seed.

Ask for confidence to be delivered with each link, and ask what evidence sits behind the tiers. Then test whether precision at the top tier justifies its price, because for many use cases a smaller high-confidence set outperforms a larger mixed one, especially where a wrong match carries a cost beyond an ignored impression. Where confidence tiers are not available at all, treat the quoted rate as an upper bound on usable matches rather than a description of them.

The Unit Price Is Rarely the Cost

Identity resolution is priced per match, per record, or per seat, and the invoice is usually a minority of the real spend. The rest sits in the work required to make the output usable: preparing and validating input files, building and maintaining the join, absorbing schema changes, re-running matches as the graph refreshes, and storing the results. None of it appears in a rate card, and all of it recurs.

Model the first year including that work, and model the second separately, since ongoing cost differs sharply from setup. Then examine the refresh terms carefully, because a graph that requires re-matching to stay current turns a one-off purchase into a subscription whether or not the contract calls it one. Ask what the second-year cost looks like at steady state, and compare vendors on that number rather than on the first invoice.

Where the Meter Actually Runs on a Match Contract

Normalizing three incompatible pricing models into one cost per matched key requires knowing what the meter counts, and that definition sits in the order form rather than the rate card. Ask whether a record that matches this month and matches again next month on a refreshed pull bills once or twice. Ask whether a key you suppress after match still counts as a match. Ask whether re-running an identical seed during a pilot is billable. Vendors quoting the same headline rate can diverge by a multiple on those answers alone, and none of it is visible on a data sheet.

Commit structure is the second place the meter moves, and it interacts directly with the decay described above. A commit denominated in matched keys keeps buying the same nominal quantity from a thinning observable cohort, so its real value erodes across the term. A commit denominated in dollars at a fixed unit rate carries the opposite exposure, since underconsumption forfeits spend. Neither is wrong, but they fail in opposite directions. Say out loud which one you are signing, and model both against your own cohort assumptions before the renewal conversation rather than during it.

Finally, pin the definition of a match to an artifact you hold. Require delivery of the matched key list itself, not only the count, so your own recomputation becomes the invoice check. Where hashing is involved, put the normalization spec into a contract exhibit covering case folding, whitespace handling, plus-addressing, and the hash algorithm. A preparation mismatch produces a low match rate that looks exactly like a thin graph and gets argued as one for weeks before anyone checks the inputs, which is why the prep spec belongs in the seed match test too. The scorecard both teams share is only defensible when either side can recompute the same number from the same file.