Check these programmatic advertising measurement mistakes

What to take away

1. One attribution window everywhere

If a demand-side platform reports on a seven-day click window and analytics uses a thirty-day view window, totals will never agree. Pick one window per objective and apply it across every report you send.

Where a client insists on a second view, label it clearly rather than blending the two into one figure. The wider choices are in our guide to programmatic advertising attribution methods.

Record the chosen window in the campaign brief before the first line runs. A planner who inherits the account should not have to guess it.

2. Blended channel totals

A single line called digital spend hides which channel earned the conversion, so split paid search, social and display before calculating cost per acquisition. Where a channel cannot be separated, say so in commentary instead of guessing a share.

Channel separation exposes overlap, where the same conversion is counted twice; flag the duplication so the client sees it.

3. Audience definitions left stale

Targeting built on last year's segment will report against the wrong people. LinkedIn's documentation on ad targeting options shows how many layers one audience can carry, so a definition needs an owner and a review date.

Re-check the segment whenever the brief changes, and keep one master list noting each segment's platform, because duplicated definitions drift apart within a quarter.

4. Consent treated as a technical detail

Consent mode affects what your tags can record, so a drop in recorded conversions may be a data gap rather than weaker performance. Compare consented and modelled totals separately. Report the gap to the client rather than smoothing it away. Record the consent rate next to the conversion rate, so a fall in one can be read against the other.

5. Crawler access blocking your own checks

Verification vendors and ad servers fetch pages to confirm placement and creative; rules that block them produce empty verification rows.

Google's robots.txt specification explains how crawlers interpret allow and disallow lines, so read it before blaming a vendor for missing data. Ask each vendor which user agent it crawls with, then confirm that name appears in your rules file.

6. Vanity metrics in the headline row

Impressions and reach belong in an appendix. The headline should carry the measure tied to the objective, whether that is a cost per qualified lead or a completed view rate. Our pillar on programmatic advertising measurement and reporting sets out how to structure that reporting pack.

7. No baseline before launch

Without a pre-campaign figure you cannot show movement. Record the starting cost per acquisition, conversion rate and audience size before the first impression serves. A team spending £12,000 a month, for example, still needs that baseline line.

8. Invented benchmarks

Quoting an industry average you cannot source weakens the whole report. Use your own account history, or cite a published figure with its date. The ONS bulletin on e-commerce and ICT activity is one UK source for context on business digital adoption, but it is context, not a campaign target.

9. Reports nobody can audit

If a client cannot trace a number back to its source, trust erodes. Keep a short method note with each report: window, channel split, consent treatment and the date the audience was last reviewed. Name the person who owns the note and when it was last updated, because an unsigned method note is hard to defend six months later.

Common questions

Which mistake costs the most?

Mismatched attribution windows, because every downstream total inherits the error. Fix the window first, then rebuild the report.

Should I include platform-reported and analytics-reported figures?

Yes, side by side and clearly labelled. The gap between them is useful information for the client.

How often should audience definitions be reviewed?

At every brief change and at least once a quarter. Assign one named owner per segment.

Does crawler access really affect measurement?

It can. If verification crawlers cannot fetch a page, placement checks return incomplete data, which then looks like poor delivery.