MMM & Attribution
Marketing-mix modeling (MMM) explains your sales using your spend: privacy-durable, no cookies, no user tracking.
A free 30-minute call to map your data, your decisions, and where measurement is letting you down.
A calibrated Bayesian marketing-mix model fit to your spend and sales, with uncertainty baked in.
Pressure-test against experiments, holdouts, or a known answer. Honest about what's solid and what isn't.
A plain-language readout plus a tool or model your team keeps using, not a black box.
After handoff, I stay on to recalibrate the model as spend, channels, and seasonality shift, so it keeps reflecting reality.
Typical first engagement: 4-6 weeks from kickoff to a model your finance team can trust.
Marketing-mix modeling rests on four ideas. Play with each one below; the data comes from a known "true" model, so you can watch the method recover the answer.
A burst of spend doesn't all convert the week you pay for it. Awareness lingers and decays. Adstock smears each week's spend forward: aₜ = xₜ + λ·aₜ₋₁. Slide the carryover λ and watch the bars (raw spend) turn into the curve (effective, remembered media).
Doubling spend never doubles sales. Each channel has a response curve that bends over as it saturates (a Hill curve: f(a) = aᵃ / (aᵃ + κᵃ)). The half-saturation point κ is where you've "used up" half the channel's potential. The dot marks where this channel currently operates.
Put adstock and saturation together for every channel, add a baseline (what you'd sell with no marketing), and you can decompose observed sales into its drivers: salesₜ = baseₜ + Σ βc·f(adstock(spend)). The shaded bands show each channel's contribution; the dots are actual sales. Toggle channels to see what each is really worth.
Because every channel saturates, the best plan equalizes the marginal return across channels, pour into channels with headroom, pull back from saturated ones. Below, the model reallocates your weekly budget to maximize sales. Slide the total budget and compare your current split (left bars) to the model's recommendation (right bars).