Executive summary
A multi-category e-commerce retailer is developing a measurement capability it can own and operate. Its capable internal data science team owns delivery; Tyedal works alongside it as an embedded adviser. The programme connects three workstreams: campaign-level multi-touch attribution (MTA), incrementality experiments and an early Marketing Mix Model (MMM) proof of concept.
The aim is not to crown one model as the answer. It is to give each method a clear job: MTA for tactical reporting, experiments for incremental effects, and MMM for broader planning. Progress is real, but production automation, final experimental conclusions and a planning-ready MMM remain ahead.
The challenge
The retailer already has analytical capability. The challenge is turning separate analyses into a dependable decision process without confusing attributed conversions with sales caused by marketing. A manual campaign attribution pipeline needs to move towards automated production and dashboard reporting, while leadership needs evidence that can support decisions beyond observed customer journeys.
Across multiple categories, plausible explanations also need testing. Differences in demand or customer behaviour may matter, but category-level hypotheses are not confirmed findings. The programme must preserve that distinction while making useful progress.
Tyedal’s role
Tyedal provides embedded advice alongside the internal team, helping scrutinise methods, assumptions and the interpretation of results. The team retains ownership of implementation and delivery. The adviser’s job is to challenge critical assumptions, suggest the next analysis and help connect the evidence to a decision, while building the team's ability to do that work independently.
Three complementary workstreams
1. MTA for tactical reporting
Campaign MTA is moving from a manual pipeline towards automated production and dashboards. The work includes Markov and Shapley approaches, alongside predictive research. These approaches can organise observed journeys and inform tactical reporting, but predictive accuracy does not establish causal attribution.
The practical next step is to put stable, interpretable outputs in front of marketers and learn whether they support real decisions. Assigning credit to a campaign does not, by itself, establish what would happen if its spend changed; that question needs additional causal evidence.
2. Experiments for incrementality
A geographic paid-search holdout analysis has been strengthened by freezing donor weights using the pre-period, rather than allowing the comparison to adapt to outcomes after treatment. The analysis has also expanded beyond revenue to orders and customers, broadening the view of the response.
Early findings suggest an incremental role for paid search alongside partial substitution into organic activity. The test is still underway, so the magnitude and uncertainty need to be assessed in the completed analysis. Differences between revenue, orders and customers have also prompted questions about basket and category mix, providing a focused agenda for further investigation rather than settled conclusions.
3. MMM for broader planning
MMM is an early proof of concept. Demand controls, event effects and model specification are still under review. The intended role is broader planning, complementing journey-level reporting and experimental evidence rather than overriding their limitations.
Experimental calibration is a future step, not something already completed. For now, the team is reviewing inputs, assumptions and model structure so that a good predictive fit is not mistaken for credible attribution. Completed experimental evidence can then help challenge the model, provided the intervention and outcome being compared are aligned.
Progress and next steps
Progress so far is qualitative: clearer methodological boundaries, a stronger holdout comparison, a wider set of outcomes under examination and leadership support for more experimentation. The conversation is moving beyond a single return-on-ad-spend figure towards what to test next, which outcomes matter and how the evidence should inform the plan.
Next steps are to advance MTA automation and reporting, complete the holdout analysis, test remaining hypotheses and continue reviewing the MMM specification. Over time, experimental evidence can inform calibration and planning. The goal remains a complementary system whose assumptions the internal team understands and whose delivery it owns.
Building your own measurement capability?
Discuss where embedded advice could help your team connect reporting, experiments and planning while keeping ownership in-house.