Why this matters
Each platform reports its own reach as if the others did not exist, and each cap counts only its own identifiers. The person who sees an ad everywhere is invisible to all of them. The cross-platform reach and frequency guide explains why summed reach overstates people and understates exposure, and the viewability and attention guide explains why an impression nobody could see should not count towards either. Here you apply both to one plan.
The brand and its media
Slumberfield is an invented online mattress brand. Its last flight ran display and YouTube through DV360, a Meta campaign bought separately and a direct video sponsorship served through CM360 tags. The creative was a sequence of messages, written to be seen a handful of times. The brand suspects the same people saw it far too often.
Slumberfield: last flight by platform (four weeks)
| Platform | Spend | Impressions | Reported reach | Average frequency | Share of impressions to identifiers above 12 exposures | New reach in final week |
|---|---|---|---|---|---|---|
| DV360 (display and YouTube) | £48,000 | 7,200,000 | 1,800,000 | 4.0 | 14% | 210,000 |
| Meta | £52,800 | 9,600,000 | 1,500,000 | 6.4 | 38% | 60,000 |
| Video publisher (CM360) | £18,000 | 1,200,000 | 400,000 | 3.0 | 2% | 45,000 |
Planned exposure: around 8 per person across the flight, and none above about 14.
Overlap assumption from a measurement partner's category panel: about 50% of Meta's audience was also reached by DV360, and about 40% of the publisher's audience by DV360 or Meta.
Caps last flight: DV360 line item caps only, no campaign cap; Meta none; publisher none.
Viewability (Active View for DV360, the vendor for the publisher):
| Source | Measurable rate | Viewable rate on measurable |
|---|---|---|
| DV360 display | 78% | 64% |
| DV360 YouTube | 97% | 88% |
| Publisher video | 95% | 80% |
Of DV360 display's unmeasurable impressions, most come from one group of app placements without a measurement SDK.
Reading reach and frequency
Start by refusing to add. Summed reach is an upper bound; the overlap assumption, written down, turns it into an estimate that can be challenged. Then read where frequency lives: the average hides the tail, and the share of impressions served above the intended ceiling is the figure that says whether money is buying people or repetition. New reach in the final week shows which platform still has people to find and which has saturated its audience.
Finally, check the base. Impressions in the unmeasurable bucket cannot be shown to have been seen, so a platform whose reach leans on them is weaker than its headline figure suggests.
Task
Answer before reading on.
- Calculate summed reach and a deduplicated estimate, stating the assumption.
- Estimate frequency for people reached on both DV360 and Meta, and say why the true figure is probably higher.
- Identify which platform is buying repetition rather than people, with the evidence.
- Set caps for the next flight in each system so a person reached everywhere stays near the ceiling.
- Rebalance the same total budget towards unique reach, and check each platform can absorb its new spend under its cap.
- Say what you would do about the unmeasurable display inventory, and name the test that would confirm the shift.
Worked answer
Slumberfield: deduplication, caps and rebalance
Summed reach: 1,800,000 + 1,500,000 + 400,000 = 3,700,000. Overlap: 50% × 1,500,000 = 750,000; 40% × 400,000 = 160,000. Deduplicated estimate: 3,700,000 − 750,000 − 160,000 = 2,790,000.
People on both DV360 and Meta: roughly 4.0 + 6.4 = 10.4 exposures, before the publisher, if they received each platform's average. Heavy users of both are more likely to be in the overlap, so the real figure is higher.
Next flight caps: DV360 campaign cap 6 per flight; Meta 1 impression per 7 days (at most 4 in the flight); publisher ad-level cap in CM360 of 3 per flight. Sum for a person on all three: 13.
| Platform | New spend | Impressions | People needed at the cap |
|---|---|---|---|
| Meta | £30,000 | about 5,455,000 | at least 1,364,000, within the 1,500,000 reached last time |
| DV360 | £70,800 | about 10,620,000 | at least 1,770,000; new reach still 210,000 a week |
| Publisher | £18,000 | 1,200,000 | at least 400,000, as last flight |
Total unchanged: £118,800.
Where the waste is. Meta carries the highest average, more than a third of its impressions above the ceiling and almost no new reach at the end of the flight: its audience was saturated and extra spend bought repetition. DV360 was still finding people every week, and the publisher reached a largely separate audience.
Caps. Each system can only cap its own identifiers, so the caps are set so their sum sits near the ceiling, with a campaign-level cap in DV360 rather than line item caps alone. They remain ceilings per identifier, not per person: people without usable identifiers will still exceed them.
Budget. Money moves from Meta to DV360, where the audience was not exhausted, and the arithmetic confirms each platform can spend its new budget under its cap. The publisher stays as it is.
Viewability. Exclude the unmeasurable app placements from DV360 display, or require the SDK, so the reach being paid for can be shown to have had a chance to be seen.
The test. The shift is a hypothesis. A geo test, holding Meta at its old level in some regions and at the new level in matched ones and comparing total mattress orders, shows whether the cut cost sales. The incrementality testing guide covers the design; its IncrementalityConversions caused by marketing. result, not a reach figure, should decide the following flight.
Common mistakes
- Presenting summed reach as people reached. Publish the overlap assumption with the estimate.
- Reading the average and not the tail. The share above the ceiling is the signal.
- Caps in one system only. Each platform caps its own identifiers; the person sees them all.
- Moving budget to a platform that cannot absorb it. Check audience size against the cap first.
- Counting unmeasurable impressions as reach. They may never have been seen.
- Treating the rebalance as proven. Only a test shows whether the cut lost sales.
Next, bring the whole level together in the programmatic audit capstone or return to the intermediate programmatic level.
