A 2025 attribution benchmark reported a median person-to-account identity resolution match rate of 72% at production launch. The top quartile hit 88%. That gap is the difference between a pipeline report your CFO trusts and one that gets picked apart in the next QBR.
What most demand gen teams miss: that match rate degrades. Roughly 25–30% of B2B contact data goes stale every year. People change jobs. Corporate IPs get reassigned. Hashed emails go cold. The identity graph you validated in Q1 is a different animal by Q3, and your campaign measurement inherits every bit of that rot.
Identity Is Not Attribution
An identity graph answers one question: "Who is this person, and which account do they belong to?" It does not answer the question you need for measurement: "Which touchpoint caused this account to move into pipeline?" Treating a strong identity graph as proof of attribution is how teams end up defending numbers that don't survive scrutiny.
Only 31% of companies track multi-touch attribution beyond first and last touch, according to a 2024 benchmark. That means 69% are compressing a 6-to-10-stakeholder buying process into one or two moments. Layer a decaying identity graph underneath that already-thin model, and you're measuring ghosts. Identity resolution is infrastructure for measurement, not measurement itself. The graph gives you the plumbing. The experiment gives you the signal.
Graph Health as an Operational KPI
If 42% of B2B pipeline comes from marketing-influenced touchpoints (a benchmark cited in 2024 research), then the accuracy of your identity graph directly determines whether you can claim credit for that pipeline. A stale graph misattributes touches. Misattributed touches inflate some channels and deflate others. Budget follows the wrong signal. The cycle compounds.
Graph health metrics worth tracking quarterly:
- Match rate: what percentage of campaign impressions resolve to a known account? Practitioner guidance puts realistic B2B person-level resolution at 5–20% of traffic, with stronger graphs reaching 30–40%+ depending on coverage and freshness.
- Confidence threshold: what's your cutoff for probabilistic matches? A 60% confidence match inflates reach but introduces noise.
- Conflict/merge rate: how often does the graph produce conflicting account assignments for the same identifier? Rising conflict rates signal decay.
- Data freshness: what's the median age of your identity links? If the median timestamp is older than 90 days, you're measuring with last quarter's map.
None of these show up in your DSP dashboard. You have to instrument them yourself or demand them from your identity vendor. Vendor-published match rates are directional, not independently audited. Validate against your own CRM truth data.
The Two-Graph Trap
You build an audience using one vendor's identity graph. You measure campaign delivery using a different vendor's graph. The same impression resolves to "Boeing" in system A and "Walmart" in system B. Targeting blames measurement, measurement blames targeting, and nobody can arbitrate because there's no shared truth set. This isn't a hypothetical edge case. It's structural whenever activation and measurement run on different identity foundations.
The fix in principle: target and measure on the same graph. In practice, it means auditing which identity sources feed your DSP, your measurement tool, and your CRM, then reconciling them. Walled gardens complicate this further. LinkedIn's professional identity graph, for example, powers its Sponsored Content products effectively within the platform, but cross-channel measurement requires reconciling LinkedIn's in-platform reporting with your first-party and CRM-based data. That reconciliation is where graph fragmentation bites hardest.
Run It This Week
The hypothesis (make it falsifiable): if we audit our identity graph's freshness and match rate against CRM-verified records, then we'll find at least 15% of attributed pipeline touches are misassigned, because our graph hasn't been validated against truth data in over six months.
Setup: Pull 500–1,000 CRM records where you know the IP, cookie, or hashed email and the verified employer. Strip the employer field. Send the blind list to your identity vendor and score their responses against your truth set.
Success = documented match accuracy for your specific traffic mix (not the vendor's published benchmark). Guardrails = flag any identifier category where accuracy falls below 70%. Stop-loss = if overall accuracy is below 60%, pause attribution reporting that depends on that graph until you've addressed the decay.
What to measure: Match rate by identifier type (IP vs. HEM vs. cookie). Don't over-interpret aggregate match rate alone. A graph can score 80% overall but fall apart on remote/hybrid workers where IP-to-company mapping is weakest.
The trade-off you're accepting: This audit takes ops cycles and may surface uncomfortable truths about your current attribution numbers. Pipeline credit may shift between channels once you clean the graph. That's the point.
The measurement variable most teams ignore isn't a new metric or a fancier model. It's whether the identity layer underneath all their reporting is still telling the truth. By Q3, a graph validated in Q1 has lost a quarter of its accuracy. That's not a data hygiene problem. That's a measurement integrity problem, and it's sitting under every number you present to the board.
