The Maze: One advertiser's measurement systems tell opposite budget stories. Platform attribution reports £8.20 back for every £1 spent on Google Ads, more than double its £3.40 MMM estimate. Meta, TikTok and Snap move the other way: every social channel looks better under marketing mix modeling. The useful conclusion is not “cut search.” It is that conversion credit and incremental contribution answer different questions—and a budget built on only one can starve the channels that created the demand Google later captures.
Google's apparent lead shrinks by £4.80 per £1 spent. In Nick Handley's source case, Google Ads moves from £8.20 under platform attribution to £3.40 under MMM. Platform attribution is therefore 2.41 times the MMM estimate. Search can sit close to the conversion and inherit credit for demand formed earlier, especially through brand queries. The result does not make Google unprofitable. It makes the size of its lead uncertain.
Every social channel reverses direction. Meta rises from £2.80 to £4.20, TikTok from £1.90 to £2.80 and Snap from £1.80 to £2.60. That is a 50% lift for Meta and roughly 47% and 44% for TikTok and Snap. Under platform attribution, all three sit below £3 and Google dominates. Under MMM, Meta ranks first while TikTok and Snap close much of the gap. The measurement method does not merely change decimal places. It changes the order in which a finance team may fund channels.
Attribution allocates credit; incrementality estimates causation. Google's Meridian documentation shows why modern MMM is broader than a historical regression exercise: it can calibrate with incrementality experiments, model reach and frequency, and use search-query volume to control for organic demand. Meta's GeoLift methodology frames the same problem as observed results versus the counterfactual without advertising. Attribution remains useful for operational reporting. It is just not a substitute for a causal baseline.
A better stack triangulates, rather than crowns one model. MMM can estimate cross-channel and offline effects, but it is only as good as its inputs, model design and calibration. Geo holdouts or randomized lift studies can test specific causal claims; TikTok's Conversion Lift Study, for example, uses randomized test and control groups. Attribution can still help teams inspect journeys and execute campaigns quickly. Agreement across methods raises confidence. Disagreement is a signal to investigate, not permission to choose the friendliest number.
The case is a warning, not a benchmark. The source does not disclose the advertiser, market, period, spend, model specification, confidence intervals or whether the methods use identical revenue definitions. Channel returns also depend on scale: a heavily funded search program and a smaller social program may have different marginal economics. The eight values are useful because they expose a governance risk. They cannot prescribe another company's budget split.
Why it matters: Retail and ecommerce teams increasingly buy across search, social, marketplaces, retail media and creator channels, while customers move between them without respecting reporting boundaries. If the final observable click becomes the budget's single source of truth, demand capture will systematically look safer than demand creation. The management fix is not another dashboard. It is a measurement portfolio: attribution for speed, MMM for the whole system, and controlled tests for causal checks. Budgets should follow the overlap, not the loudest platform report.
