Marketing Sciences · MMM
Know which dollar actually moved revenue.
Kleyr Modeler is a marketing mix modelling workbench for analysts and operators. Traditional MMM today, Bayesian Structural and a skill-based AI agent next, with the same data spine, the same outputs schema, and a public API to your warehouse.
PyMC-Marketing backbone·Adstock + Hill saturation·Stepwise OLS / Ridge·Reproducible runs
What's in the box
A complete MMM workflow, web-first.
01
Data View
Drop a CSV. We infer column types, surface per-column stats, and let you tag KPI / media / control roles in one click. Connectors for Shopify, Klaviyo, Meta, Google, TikTok, GA4 land post-v1.
02
Transforms
Geometric and Weibull adstock for carry-over. Hill, Michaelis-Menten, and S-curve saturation for diminishing returns. Live preview chart so you see the shape before you fit.
03
Stepwise regression
OLS or Ridge backbone with forward / backward / bidirectional selection on AIC, BIC, or p-value criteria. Robust HC3 standard errors. Full coefficient table with t-stats and p-values.
04
Diagnostics suite
Durbin-Watson, Breusch-Pagan, Ljung-Box, Shapiro-Wilk, VIF, residual plots, ACF, normal Q-Q. The diagnostics aren't an afterthought — they're shown by default.
05
Channel decomposition
Per-period contribution per channel. Aggregated totals, share-of-spend vs share-of-effect, response curves. CSV bundle that matches the analyst-grade output schema downstream tools already consume.
06
Calibrated by experiments
One click attaches a geo lift estimate from the Experiments module as a Bayesian prior on the corresponding channel. The MMM treats it as a calibration target — exactly the loop that makes enterprise MMM credible.
Methodology
Built on the canon, not on a vibe.
Every method we ship traces back to a paper. Read the references; we'll meet you where the literature is.
PyMC-Marketing
open-source · pymc-labs/pymc-marketing
Bayesian MMM, BTYD CLV, customer choice. Active OSS by the PyMC team.
Google Meridian
open-source · 2024 launch · google/meridian
Hierarchical Bayesian MMM with geo-level pooling. Reference for multi-geo customers.
scikit-learn / statsmodels
standard libraries
OLS / Ridge regression, robust SEs, full diagnostic test suite, stepwise selection.
Foundational paper
Jin et al. (Google, 2017) · Bayesian Methods for Media Mix Modeling
The reference Bayesian formulation. Chan & Perry (2017) covers the failure modes.
Use cases
Three jobs, one model.
Reallocate the media budget
- ✦Find wasted spend in saturated channels
- ✦Project the upside of moving £1 between channels
- ✦Bring it to the CFO with credible intervals
Plan next quarter
- ✦Forecast revenue under each plan
- ✦Stress-test against do-nothing
- ✦Commit a number you can defend
Settle internal debates
- ✦Brand vs performance
- ✦Long tail vs the big four
- ✦Promo cannibalisation
FAQ
MMM, in detail.
Traditional MMM (OLS / Ridge with stepwise selection) is v1 — same shape as the established analyst workflow but on a web platform. Bayesian Structural MMM via PyMC-Marketing lands in v1.5. Both share the same input shape, run lifecycle, and output schema, so switching is a config change, not a re-tooling.
✦ Get started
Bring your spend, leave with a model you can defend.
Book a 30-minute working session with the Kleyr team. We'll model your last 12 weeks of marketing spend on the call — no slides, no abstract demo.