Menu

How to Improve Programmatic Campaigns with Reverse Analytics

By Max Legrand · GreyFox Ads · Insights from the French programmatic market

Reverse analytics in programmatic advertising: comparing bid request volumes before and after targeting filters

A Media Trader once told me: “The problem with semantic and brand safety segments on marketplaces is that they’re a bit of a black box.” And they had a point. Testing and auditing a targeting segment can be surprisingly time-consuming. But what if we could reverse the question? 🔍

In brief: Reverse analytics compares inventory data before and after a pre-bid filter is applied. By examining where bid request volumes change, Media Traders can estimate a filter’s impact, spot unexpected behaviour and make better-informed targeting and data-fee decisions (provided both datasets are genuinely comparable).
In this article

1. What is reverse analytics in programmatic advertising?

Reverse analytics is an analytical approach inspired by reverse engineering: rather than accepting a targeting provider’s description at face value, you study the changes produced by its filter. You compare an a priori dataset (before activation) with an a posteriori dataset (after activation).

The objective is twofold: quantify how much eligible inventory the filter removes, and infer where and how it seems to operate. It is not a magic window into a vendor’s algorithm; it is a practical way to make an opaque result more observable.

In the French Open Web, this is especially interesting when Media Traders want to understand the real impact of contextual, brand safety, audience or sustainability-related filters across different publishers.

2. How to audit a pre-bid filter in five steps

  1. Define a fixed inventory perimeter. For example: France, display, web and mobile web, with the same dates, supply paths, devices and other targeting settings.
  2. Export the baseline. From a DSP, SSP or Curator planner, retrieve eligible bid request volumes by site or domain before the filter is applied.
  3. Activate only the filter you want to test. Keep the remaining configuration unchanged.
  4. Export the filtered volumes. Use the same reporting dimensions and, ideally, the same underlying planning window.
  5. Join and compare the files. Match domains using Excel VLOOKUP/XLOOKUP, Power Query or SQL, then create a Pivot Table to examine the differences.

Simple enough to try in a spreadsheet. And far more informative than relying on a marketplace segment name alone. 🧪

3. Example: estimate the filter’s impact by publisher

Imagine a contextual segment tested against three publisher domains. Here is a purely illustrative dataset:

PublisherBaseline bid requestsAfter filterEstimated reduction
Publisher A1,000,000850,00015%
Publisher B600,000180,00070%
Publisher C400,000392,0002%

Estimated reduction (%) = (baseline volume − filtered volume) / baseline volume × 100.

Publisher B is much more heavily affected than Publisher C. That observation is worth investigating: perhaps the segment classifies their content differently, or perhaps another characteristic of the inventory explains the discrepancy.

Crucially, this is an estimated reduction in eligible bid requests, not necessarily a verified block rate of individual impressions (planners may use sampling, modelling or different counting rules).

4. Four ways to use reverse analytics

Brand safety: where is your filter doing the work?

Map where a brand safety filter materially reduces eligible inventory. If some environments show almost no change, you can question the incremental value of paying for that particular filter there—but a low reduction is not proof that the filter is useless (the baseline inventory may already be safe).

For more on the trade-off between protection and media efficiency, see our analysis of brand safety protection and costs.

Contextual targeting: does the segment make sense?

Imagine a “baby” contextual segment allowing substantial volumes on a webmail domain. Is that really the intended context? It might be a signal of broad classification, page-level variation or measurement limitations. Reverse analytics gives you a concrete anomaly to investigate rather than a vague suspicion.

Audience targeting: find high-affinity supply

Compare the distribution of eligible requests for a demographic or intent-based segment across sites. Publishers with consistently high relative inclusion may be useful candidates for direct tests or curated supply strategies (high inclusion alone does not establish that actual users have the desired characteristics).

This complements our article on incremental targeting strategies and the role of programmatic curation.

Carbon and sustainability filters: explain exclusions

When a carbon-related pre-bid solution excludes a publisher, compare the before-and-after inventory distribution to identify the affected domains. Then ask the provider which signals or thresholds are driving the decision (an observed exclusion does not reveal the tool’s actual carbon calculation).

5. Five methodological safeguards before drawing conclusions

Where possible, use a controlled A/B setup or repeated matched exports to separate the filter effect from ordinary supply fluctuations. (And be careful not to confuse a segment’s apparent selectivity with its actual accuracy.)

6. Turn the audit into smarter programmatic decisions

Once you have the site-level comparison, sort publishers by absolute request reduction, relative reduction and remaining scale. Review outliers with your provider, check whether the filter delivers meaningful incremental value, and test alternatives before changing your activation strategy.

You can also use this work to question unnecessary data fees and opaque supply-chain decisions. As discussed in our article on financial transparency in programmatic, knowing what you pay for matters just as much as knowing what you buy.

For Media Traders working on France from international agency teams, a curated approach can simplify this analysis by combining local publisher knowledge with a more transparent activation framework.

7. The takeaway: don't just buy a segment—interrogate it

Reverse analytics is not about reverse-engineering a proprietary algorithm line by line. It is about turning a black box into a set of measurable questions: What changed? Where? By how much? And is that change useful?

Sometimes the best optimisation isn't another algorithm. It's a spreadsheet, two exports and a little curiosity. 📊

Original LinkedIn post: Read the original discussion by Max Legrand.

About the author

Max Legrand is the founder of GreyFox Ads, a French programmatic curation specialist focused on making the quality Open Web easier to activate, measure and understand.

Want to activate quality French Open Web inventory?

GreyFox Ads helps international Media Traders reach the French market through one Master Curated Deal, with a focus on efficiency, simplicity and transparency.

Discover our Master DealContact GreyFox Ads