The Identity Graph You Use for Planning and the One Your DSP Uses for Delivery Are Rarely the Same Graph

Media planning and campaign delivery feel like a single workflow. You size your audience, set your budget, launch, and measure. The hand-off between those stages looks administrative. It is not. At the seam between planning and delivery, your audience passes through at least two identity graphs that almost certainly do not share the same resolution logic, the same data inputs, or the same coverage profile. The population you planned against and the population your DSP bids against are structurally different groups, and almost no campaign budget accounts for that gap.

Where the Split Happens

When a planning team estimates audience size, they are typically working inside a data platform, a clean room, or a CRM onboarding interface. Each of those environments resolves identity through whatever graph the platform vendor licenses or builds. When that same audience moves into a DSP for activation, the DSP resolves identity again, through its own graph or a graph it has contracted with. Those two graphs are almost never identical.

The difference is not a bug in either system. Each vendor has made different sourcing choices: different publisher login data, different device signal partnerships, different probabilistic linking rules, different refresh cadences. Two graphs can look similar at an aggregate level and still diverge sharply on the specific customer segments your campaign needs. A planner who sizes an audience at, say, four million addressable households is measuring graph A's coverage of that segment. The DSP bidding against that plan is working from graph B's coverage. The actual delivered audience reflects graph B, not the plan.

Why This Is Not Just a Reach Discrepancy

Buyers often notice a version of this problem as a simple reach shortfall: the plan projected X impressions, delivery came in under. That framing makes the issue sound like a scale problem, and vendors are good at offering scale explanations. The deeper issue is compositional, not volumetric.

Consider a hypothetical: your planning graph resolves your CRM file and identifies a high-value segment as skewing toward a specific geography or purchase behavior. Your DSP graph resolves the same file and finds a different subset of that segment addressable, because its coverage of that geography or that device type is structured differently. You do not just get fewer impressions. You get impressions against a population whose composition was shaped by the DSP graph's particular strengths and gaps, which were never part of your plan.

This matters most when your campaign goal depends on reaching a specific type of buyer, not just a volume of buyers. Frequency, sequencing, and message relevance are all calibrated to the planned audience. If the delivered audience is structurally different from the planned one, those calibrations are off before the first impression runs.

The Refresh Timing Problem Compounds This

Even if a planning graph and a DSP graph were well-matched at the moment you set up a campaign, they update on different schedules. Planning graphs used in data platforms may refresh monthly or quarterly. DSP graphs update continuously, incorporating new signal as it arrives. A campaign with a long flight window is being planned against a static or slow-moving population estimate while delivery resolves identity against a graph that has changed since planning concluded.

This is worth flagging to your planning team before setting expectations with stakeholders. The audience size number in your plan is a snapshot from a specific graph at a specific moment. Delivery will reflect a different snapshot from a different graph at a series of different moments across the campaign flight.

Practical Questions to Ask Before a Campaign Launches

You do not need to resolve every graph incompatibility to run a better campaign. You do need to know where the seams are so your interpretation of results is accurate.

First, ask your planning platform which identity graph powers its audience sizing. This is a reasonable vendor question, and a planning partner who cannot answer it is giving you audience estimates with no documented methodology.

Second, ask your DSP which identity graph or graphs it uses at bid time, and whether that varies by inventory source. Many DSPs swap graphs depending on whether inventory comes from an open exchange, a private marketplace, or a direct publisher deal. Your planned audience may resolve cleanly in one context and fragmentarily in another.

Third, ask whether the two graphs share a common resolution layer. Some identity partners offer graph interoperability that reduces, though rarely eliminates, the divergence between planning and delivery populations. If your planning platform and your DSP both connect through a shared persistent identifier standard, the translation loss is smaller. It is still not zero.

Fourth, build a post-campaign review that explicitly asks whether the delivered audience was consistent with the planned audience's key dimensions, not just whether reach and frequency targets were hit. This requires that you document your planned audience's defining characteristics before launch, which is a useful discipline regardless of the graph alignment question.

What Good Looks Like

A well-managed campaign does not assume graph alignment. It treats planning estimates as directional and builds a delivery review into the workflow that checks compositional consistency, not just volume delivery.

Some buyers are starting to request that planning and activation occur inside environments that share a single graph layer, or that explicitly map between known graph states. This is a reasonable ask, particularly for campaigns where audience composition is central to the strategy, such as CRM-based retention programs or high-value B2B account targeting.

Others are using post-campaign clean room analysis to reconstruct what the delivered audience actually looked like against their first-party data, rather than trusting platform-reported audience descriptions. That approach has its own graph-state limitations, but it does surface compositional gaps that pure delivery reporting hides.

The core discipline here is simple, even if the infrastructure behind it is not. Planning and delivery are two separate identity resolution events. Treating them as one continuous operation is the assumption that causes the most persistent and least visible measurement error in digital media buying.