Programmatic Attribution Models: Why They Matter More Than You Think
Attribution models might be the most underrated part of AdTech. Why? Because the way you decide to credit conversions can change what your DSP, your agency and your Media Traders try to optimize. 📏
We spend hours debating CPMs, audiences and placements. Yet a more fundamental question is often left unanswered: what exactly counts as a successful contribution to a conversion?
1. What is an attribution model in programmatic advertising?
Attribution is the method used to assign credit to advertising touchpoints along a customer's journey before a conversion. A touchpoint might be a display impression, a video exposure, a paid-search click or another measurable interaction.
The goal is not simply to count conversions. It is to decide which interactions deserve credit and how much. That choice shapes reported CPA, ROAS and the optimization decisions that follow (the exact definitions depend on the advertiser's measurement setup).
2. Post-click versus post-impression: what do attribution windows really mean?
In programmatic advertising, a common setup combines post-click (PC) and post-impression or post-view (PI/PV) windows. For example, a campaign might use 7-day post-click and 1-day post-impression attribution.
| Method | What it typically credits | Important limitation |
|---|---|---|
| Post-click (7 days) | A qualifying conversion after an ad click within seven days | A click does not prove the ad caused the purchase |
| Post-impression (1 day) | A qualifying conversion after an ad exposure within one day, even without a click | The impression may not have been viewed or influential |
| Multi-touch attribution | Credit distributed across several recorded touchpoints | Still depends on observable data and modeling assumptions |
Post-impression measurement raises three questions I think every Media Trader should ask: Was the ad actually visible? Did the placement have a realistic opportunity to attract attention? And, most importantly, did the exposure change the user's behavior?
These questions are not interchangeable. As we explain in our article on the limitations of viewability optimization, a viewable impression is not automatically an attentive or effective one. And our analysis of attention on the Open Web explores why the quality of exposure matters.
3. How post-impression attribution can reward cookie dropping
Here is the uncomfortable part. A model that rewards any qualifying impression shortly before a conversion may create an incentive to maximize the number of people tagged at the lowest possible cost.
That can encourage a practice often called cookie dropping: serving very cheap impressions, sometimes with aggressive reach and frequency strategies, primarily to create an attribution opportunity rather than meaningful advertising impact.
Imagine two placements. Placement A delivers a genuinely engaging creative in a prominent position. Placement B delivers a barely noticed impression at a fraction of the CPM. If the reporting model awards the same post-view conversion credit to both, the cheaper placement can look disproportionately attractive (even when its actual incremental contribution is weaker).
This does not mean every post-view conversion is worthless or that all low-CPM inventory is ineffective. It means that the attribution rule can reward exposure volume rather than true influence.
4. Shapley-value attribution: what can team sports teach AdTech?
There are more sophisticated approaches. Some attribution providers, including Wizaly in its methodological positioning, have used approaches based on Shapley values, a concept from cooperative game theory developed by mathematician Lloyd Shapley.
Think of a team sport. How do you determine how much each player contributed to a victory? One intuition is to compare the team's results across different possible combinations of players. Shapley values formalize the idea of assigning each participant an average marginal contribution across possible coalitions.
In advertising, the players become touchpoints: a video impression, a display exposure, a search interaction or a direct visit. A Shapley-inspired model distributes credit by examining how the presence of touchpoints changes the modeled value of different journeys.
The attraction: instead of giving all the credit to the last recorded exposure, the model can recognize that several interactions may have contributed. Cheap impressions added just before a purchase may then receive less credit if they contribute little explanatory value.
(Real-world implementations generally rely on statistical estimation, sampled or modeled paths and available identity signals; they do not literally observe every possible journey or every conversion, including untracked ones.)
5. Important distinction: attribution is not incrementality
Even a sophisticated multi-touch model is not a causal experiment. Attribution estimates how credit should be allocated among observed interactions; incrementality asks how many conversions would not have happened without the advertising.
For example, a loyal customer might buy tomorrow whether or not they see today's banner. An attribution system can assign that banner some credit, but that alone does not establish that the banner created an additional sale.
Where feasible, use controlled lift studies, holdouts or other appropriate experimental designs to test incremental outcomes. Use attribution to guide operational diagnosis, not as the sole proof of advertising effectiveness.
That is also why I question the assumption that DSP optimization always means true business optimization: the algorithm can only optimize against the signal it receives.
6. Choose KPIs that match the audience's stage in the funnel
At GreyFox Ads, my recommendation is to align your campaign's measurement framework with the conversion funnel, rather than forcing every impression to compete on the same last-touch CPA.
| Audience situation | Primary objective | Useful KPI examples |
|---|---|---|
| Matches the buyer persona but shows little immediate purchase intent | Branding | Qualified reach, attention, brand lift |
| Shows category intent but does not yet know the advertiser's website | Consideration | Qualified visits, engaged sessions, incremental site traffic |
| Shows intent and has already interacted with the advertiser | Conversion | Incremental conversions, qualified leads, CPA or ROAS with suitable validation |
The logic is simple: do not judge a first introduction to a brand as though it were a retargeting impression aimed at someone ready to purchase. Likewise, do not reward retargeting solely because it touches customers who were already likely to convert.
7. A practical attribution checklist for Media Traders
- Write down the attribution rules: click/view windows, lookback periods, deduplication and conversion definitions.
- Check exposure quality: viewability, attention indicators, placements and frequency (where measurement is available).
- Separate prospecting from retargeting: they have different roles and baseline conversion probabilities.
- Compare models: see whether budget winners change under post-click, post-view and multi-touch views.
- Test incrementality: when budgets and measurement capabilities permit, validate whether the campaign produces additional business outcomes.
- Document the trade-offs: no attribution model perfectly captures privacy-restricted or cross-device journeys.
For ongoing analysis, a granular DSP-to-Excel scorecard can help you spot how different placements, devices and line items respond to changes in measurement and optimization.
8. GreyFox Ads: optimize for meaningful Open Web outcomes
Here in France, we believe the quality Open Web can deliver real value to advertisers—provided the measurement framework rewards effective, high-quality exposures rather than just cheap attributed conversions.
GreyFox Ads helps international Media Traders reach the French market through our Master Curated Deal: curated supply, market knowledge and simplified activation in their existing DSP. The aim is to make buying more effective, simpler and more transparent.
My takeaway? The best attribution model is not the one that reports the most conversions. It is the one that helps you make better decisions about the conversions your advertising actually influences. 😉
Want to improve your programmatic performance in France?
Work with GreyFox Ads to activate high-quality French Open Web inventory through a simple, transparent curated deal.
Discover the Master DealContact GreyFox AdsOriginal LinkedIn post: Programmatic: Attribution Model, the most underrated element of AdTech.