Expert perspectives on marketing measurement, MMM, attribution, and building in-house capabilities.
A £1 pricing boundary hid a commercial opportunity that the national model could not see. A real project from Joe's career, reconstructed with synthetic figures to protect confidentiality.
Read the project story →The lesson
The modelling dataset is not the data.
Look at what you are about to aggregate away before you build the model.
A real activation question, explored with simulated data. A 12% store-level lift becomes a 3.25% national pulse over four weeks, below this example's 5.7% detection threshold. What to test instead, and when the evidence is inconclusive.
Read article →An illustrative research-buying decision: compare six supplier offers against what better information could change. Explore overlapping evidence, late results and why a good purchase alone can be a poor addition to a portfolio.
Read article →A simulated budget decision: why might a £3.1m recommendation become a £1m move? Explore how incentives and transition costs change the answer, then ask which constraints explain the compromise.
Read article →A Bayesian MMM gives you a distribution over response curves and the optimiser hands back a single split. What "maximise the mean" quietly assumes, why the recommended plan is often a coin flip against last year, and how to tell whether your risk dial is connected to anything.
Read article →Equalising marginal returns is the easy case. What happens when the curves have an S-bend, why "drop this channel entirely" is usually a corner solution, and why half of every flighting recommendation is an artefact of the horizon.
Read article →Would knowing the number change what you do? EVPI, EVPPI and EVSI put a pound figure on a study before you run it, a dataset before you buy it, and a model before you build it.
Read article →Share of visible inventory counts your facings and divides by everyone else's. Why a share cannot be a treatment, how the retailer's own planogram inflates the estimate five-fold, and what to count instead.
Read article →Diminishing returns in media are mostly a counting problem, not a psychological one. A synthetic single-source panel, the coupon collector, unduplicated reach across channels, and how to stop an MMM predicting that you reached 172% of the population.
Read article →San Francisco sells LLM orchestration on the side of a bus. Whether it pays back is one of the hardest questions in media: the data that actually exists, the causal graph underneath it, and the ladder from lab experiment to geo test, synthetic control and MMM.
Read article →A promotion lifts sales 62%. The number is real and on its own it tells you almost nothing. Six places that lift can come from, what each one does to the P&L, and how to stress test a model that cannot tell them apart.
Read article →The IPA has named Tyedal's founder as one of the technical judges for its 2026 Effectiveness Awards.
Read the announcement →Separating what your sponsorship did from what the tournament did. The estimand trap, where the identifying variation comes from, and an interactive Bayesian PITA tool for sizing FMCG impact with honest uncertainty.
Read article →Why association is not enough for budget decisions. A clear intro to counterfactuals, confounding, and how causal thinking sharpens lift tests, MMM, and everyday measurement debates.
Read article →Lift tests are the gold standard for proving marketing incrementality. Learn how to design geo-based and audience-based experiments, avoid common pitfalls, and analyse results using Bayesian methods.
Read article →Move beyond last-click. Understand how algorithmic MTA maps the full customer journey, from Shapley values to Markov chains, and how to implement it in a privacy-first world.
Read article →A deep dive into Bayesian MMM: the core regression framework, adstock, saturation curves, budget optimisation, and why open-source tools have made in-house MMM accessible to every team.
Read article →