Audience overlap diagnosis
Are two ad sets eating each other's new-customer reach? The most common deep dive within an NRoAS drop.
Planned
The question: "When two ads target overlapping interests, we reach the same customers from both and NRoAS drops. Which of our ad sets overlap?"
When NRoAS drops without an obvious creative or landing-page reason, the next suspect is audience overlap: two ad sets targeting overlapping audiences, competing for the same new customers, with both claiming credit.
This is the second branch of the NRoAS drop RCA logic tree.
Who asks
- The performance analyst, in the deep dive after the weekly RCA
- The performance lead, deciding which ad sets to consolidate
Frequency
When the NRoAS RCA points at overlap, often weekly.
Data you need
adsandad_sets, with creative and interest tagsorderswith customer ID and ad attributioncustomerswith first-purchase date (for new vs repeat)- Ad-set targeting config: the interests, behaviours and lookalikes per ad set
Common gap: ad-set targeting is often not exported to the warehouse. Teams fall back on ad-set names that encode the audience, which is incomplete. Without it, this diagnosis rests on a proxy; see below.
How to ask it
- "Which ad sets have the highest audience overlap, based on who they acquired?"
- "For SKU-A, are any ad sets reaching customers who already bought from another SKU-A ad set in the last 30 days?"
- "Find ad-set pairs where more than 40% of new customers from ad set A had already been reached by ad set B in the prior 14 days."
What you'll get back
A pairwise overlap table for the product or campaign you asked about (figures made up):
| Ad set A | Ad set B | Overlap % | Cost of overlap | Suggested action |
|---|---|---|---|---|
| SKU-A_Prospecting | SKU-A_Interest_30plus | 42% | 1,800 / week | Pause Prospecting (lower NRoAS) |
| SKU-A_Festive | SKU-A_Always_on | 18% | 400 / week | Consolidate audiences |
Open the Trail to see the query behind each overlap figure.
Measuring overlap without targeting data
Without the targeting config, ask for a behavioural proxy:
- For each ad set, list the customers acquired in the last 30 days.
- For each pair of ad sets, count customers that appear in both, and rank by overlap %.
- Where impression data exists, check how many customers were shown ads from both.
When the targeting config reaches your warehouse, ask your data analyst to describe it, and ask the question again against it.
The logic tree
NRoAS dropped on a product and overlap is the suspect?
├── Compute pairwise overlap % across the product's ad sets
├── For each pair above 30%:
│ ├── Which ad set has the higher NRoAS?
│ │ └── Keep that one, pause the other
│ └── Which has fresher creative?
│ └── Tie-breaker if NRoAS is similar
├── Are more than two ad sets converging on the same audience?
│ └── Consolidate, don't just pause
└── Record the call (for example Save as report) so future RCAs can refer to itHow to make it recurring
Don't schedule this on its own; it produces false alarms when nothing is wrong. Ask it when the NRoAS drop RCA tracker flags a drop and the tree points at overlap. Re-check monthly even when NRoAS looks fine, because audiences drift.
Pitfalls
- Mistaking the funnel for overlap. A customer who sees a top-of-funnel ad and then a bottom-of-funnel ad for the same product is the funnel working.
- Treating the proxy as proof. Without targeting data you're inferring. Treat the result as a hypothesis until you can check it.
- Pausing too aggressively. Consolidating usually beats pausing: a paused ad set loses its reach; a consolidated one keeps it.
Improving accuracy
- Export ad-set targeting config to BigQuery.
- Add exposure history per customer over the lookback period.
- Add exposure from other channels (Google, marketplaces) to measure overlap across channels.