Programmatic Advertising Case Studies: How to Read the Numbers

Programmatic case studies are marketing documents. That does not make them useless — it means they have to be read the way you would read a company’s own quarterly numbers: looking for what is missing rather than admiring what is there.

This piece does two things. It sets out what a case study has to contain before its numbers mean anything, and it gives a benchmark to read them against: what traffic actually costs and how much of it exists, from our own platform, with the date it was measured. Without that anchor, “+381% ROAS” is a number without a scale.

Why most published case studies cannot be checked

The pattern repeats across vendor blogs, and it is worth naming because it is what makes the genre unreliable rather than merely optimistic.

• Ratios without absolutes. A 300% lift is arithmetic, not evidence. From a base low enough, any competent campaign produces it, and the base is usually the one number omitted.
• No spend disclosed. Results at test volume and at scale come from different problems. Without spend you cannot tell which one was solved.
• A window chosen after the fact. Any campaign has a best fortnight. Case studies tend to describe it.
• No baseline. “Improved conversion rate” implies a measured before. Frequently there wasn’t one, and the comparison is against an estimate.
• The publisher sold the media. Not disqualifying — we are an ad network writing this — but it means the sample is the best client, not the median one.

None of this means the campaigns did not work. It means the document cannot tell you whether the approach would work for you, which is the only reason to read it.

The six questions worth asking

Run any case study — ours included — through this before taking a number seriously.

What normal looks like: our own numbers

The reason percentages travel so far is that almost nobody publishes the denominators. So here are ours. Figures are from the Youtarget traffic showcase on 30 July 2026, for pop inventory, and they move with demand — treat them as an order of magnitude rather than a rate card.

Three things fall out of this table that matter more than any single case study.

• Traffic price varies about tenfold across GEOs. India at $0.17 and the United Kingdom at $1.68 are not the same business. A case study that does not name its GEO has withheld the most important variable.
• Inventory is overwhelmingly mobile. 95% of available pop impressions in India and Indonesia, 74% in the United States. A campaign left on default targeting is a mobile campaign whether that was the plan or not.
• Volume and price move together, but not proportionally. Indonesia offers 302 M impressions at $0.73 while Germany offers 43 M at $1.30 — seven times the volume at roughly half the price, on completely different audiences.

Video inventory follows a different map again: 301 M impressions available in India, 91 M in the United States, 71 M in Germany and 70 M in France. Why that differs from pop is covered in programmatic video advertising.

What $500 actually buys

Here is the arithmetic a case study usually skips. This is not a result we are claiming — it is the published CPM above multiplied out, with every assumption named, so you can see how much of any “success story” is decided before the campaign starts.

The click-through and conversion rates here are assumptions, deliberately round and deliberately stated: 0.1% and 2% are ordinary for pop traffic on a simple offer, and your own numbers will differ. The point is the shape. The same $500 buys nine times more leads in India than in the United States, which means an offer paying $10 works in one column and cannot work in the other — before anyone touches a creative.

This is also why a case study without a GEO is unreadable. A campaign reporting a $9 cost per lead has told you almost nothing: in India that is roughly break-even against these rates, and in the United States it would be a remarkable result.

How to run a case study you can actually trust

The version worth doing is unglamorous and takes a fortnight. It also produces numbers you can defend to a client or a partner.

Change one variable

One GEO, one format, one offer, one creative concept. Two changes at once and the result explains nothing, however good it looks.

Write the baseline down before you start

Current cost per conversion, current conversion rate, current volume. Written down, with the date. A baseline reconstructed afterwards is not a baseline.

Decide the window in advance

Pick the period and the minimum volume before launch — enough impressions per zone to be meaningful, not enough days to get bored. Then report that window whether or not it flatters you.

Read performance per zone, not per campaign

A campaign averaging a loss almost always contains profitable placements and catastrophic ones. The campaign-level number is the one that hides the finding.

Report spend, absolutes and the failures

What it cost, what it returned in units rather than ratios, and which segments lost money. The last part is what makes the rest credible.

What a credible case study contains

If you publish one, this is the minimum. It is also the checklist to apply to anyone else’s.

• Objective stated before the result, not derived from it.
• GEO, format, device split and vertical.
• Spend and duration, with exact dates.
• Baseline and final figures in absolute units, with the ratio as a secondary detail.
• What was changed, and what was deliberately held constant.
• What did not work — the zones cut, the creatives that failed, the GEO abandoned.
• Who paid for the media and who wrote the document.

Where to go from here

For the mechanics behind these numbers — which metrics decide profit and which merely look good — see programmatic advertising analytics. For what the cheapest inventory actually costs and how it behaves, pop traffic. And for the economics of buying and reselling attention in general, traffic arbitrage.