Programmatic Advertising Demand: What a Managed Service Actually Has to Do

Demand from advertisers for programmatic is not in doubt. What is usually in doubt is what a managed service is actually doing with the money. Five capabilities appear in almost every pitch — data, audiences, creative, platform, strategy. Below is what each one looks like when it is real, and the evidence an operator can ask for in each case.

1. Measurement first, or the other four are guesses

A service that cannot show how conversions are counted cannot honestly claim to have improved them. Click attribution and view attribution count two different populations. Across six independent measurements the post-view share landed between 89% and 96.2% — so an operator reading clicks alone sees roughly one conversion in ten. That is not a rounding error; it is the difference between defunding a channel and scaling it. The full breakdown is here: click and view attribution in programmatic advertising.

Ask for: a conversion path export, and the two models placed side by side, before the first insertion order — not after the first disappointing month.

2. Audience segmentation you can inspect

Segmentation is only a capability if the segments are yours and their construction is visible. The standard worth holding to: behavioural segments built on your own signals, tested against a control, and documented well enough that a new analyst could rebuild them from the written taxonomy. A segment named on a slide and never exported is a claim, not an asset.

Ask for: the segment taxonomy in writing, and one worked example of a segment that was tested and abandoned.

3. Creative decisions made format by format

Connected TV, audio, digital out-of-home, video, display and native are not interchangeable inventory with different price tags. Choosing between them is a media decision with a measurement consequence attached: some formats are click-poor by construction, and buying them against a click-based target guarantees they will look like failures regardless of what they did. The format decision and the attribution decision have to be made in the same conversation.

Ask for: which formats the desk would refuse to run against your current KPI, and why.

4. Platform work with a named person attached

ThumbAd runs The Trade Desk as a fully managed service, and the platform work is concrete: seat configuration, supply-path decisions, brand safety, and the fee construction underneath both. What matters commercially is not the certification but the accountability — who is named on the account, at what seniority, and whether their reporting explains the decisions taken or only restates the numbers.

Ask for: the names and seniority of the people who will touch the account, every fee written down before signature, and a sample report you can read without a translator.

5. Strategy that survives past the quarter

Strategy in programmatic is mostly compounding: a taxonomy that persists, learning that is versioned, and verdicts recorded even when they turned out to be wrong. Twenty-one months of one US casino resort operator's first-deposit acquisition are published, anonymised: cost per depositor fell from €390 to €47, on 10.4× the volume for 2.3× the budget. The reason for publishing the whole run rather than a quarter of it is that the curve is only visible over that length of time — the case study is here, and the same 21 months are runnable as an interactive model — no email required.

The shortest way to test all five

ThumbAd published twenty questions every iGaming operator should ask a programmatic partner, scored against six criteria, and then answered all twenty on the record. The scoring runs entirely in your browser: no email, no account, nothing sent anywhere. Run it on your incumbent, run it on whoever is pitching you, and keep both scores.

Where the line between managed service and in-house should actually sit is a separate question, and it is mapped function by function in managed service or in-house? — including the functions it costs us revenue to concede.

Score your desk — 20 questions