Marketing science platform — MMM · Experiments · Customer knowledge

Know whichdollar didthe work.

One connected system for measurement: Bayesian-grade MMM, geo experiments that calibrate it, and a customer layer that explains the people behind the numbers. One source of truth, not five vendors arguing.

Book a working sessionTry the benchmark simulator
Experiments
Causal lift
producesBayesian priors
MMM
Marketing mix
producesChannel elasticities
Customer Knowledge
Behaviour model
producesSegment mix
Bayesian structural MMMGibbs samplergeometric adstockHill saturationaugmented synthetic controlplacebo permutationdifference-in-differencesstepwise AICridge regression95% credible intervalsHolt-Winters baselineposterior propagationk-means · k-NN segmentstemporal-holdout propensityexperiment-calibrated priors
The product · live in the browser

Not a dashboard.
A modelling workbench.

app.kleyr.co.uk/modeler/bayesian
Actual vs posterior fit95% credible · 95% predictiveBayes R² 0.943
MAPE
3.1%
draws kept
400
priors
1 experiment
Three modules · one platform

Measurement, experiments, customers — in conversation.

Geo lift becomes an MMM prior. MMM elasticities feed the customer forecaster. Segments flow back into mix decomposition. The answers compound.

01MMMMarketing sciencesWhere the money actually went. Stepwise OLS, ridge, and Bayesian structural regression with full posteriors — adstock, saturation, diagnostics, reproducible runs.
02ExperimentsCausal inferenceLift you can defend. Geo holdouts via augmented synthetic control with placebo inference and power — one click sends the result into MMM as a prior.
03Customer KnowledgeBehavioural analyticsThe people behind the numbers. Behavioural segments, calibrated churn propensity with per-customer explanations, and forecasts reconciled to MMM elasticities.
0
modules, one connected system
0+
statistical methods across the engines
0%
credible intervals on every estimate
0
click from experiment lift to MMM prior
How it works

Six steps to a defensible answer.

No slides, no black box. Every step is something you can open, inspect, and re-run — the workflow an in-house science team would build, already built.

01
Connect or upload

Drop in a CSV today; Shopify · Klaviyo · Meta · Google · TikTok · GA4 connectors land next.

02
Tag columns

Mark KPI, media, and controls. Adstock + Hill saturation are live-previewed before you fit.

03
Run a model

Stepwise OLS, ridge, or Bayesian structural regression with full posteriors and the complete diagnostic suite.

04
Triangulate with experiments

Geo lift via augmented synthetic control, classic SCM, or DiD. Send the lift into the next MMM as a Bayesian prior — one click.

05
Explain customers

Behavioural segments, calibrated churn propensity, per-customer explanations — not just scores.

06
Decide and ship

Reproducible runs, saved to your own database. Fast Forward turns the model into next quarter's plan with uncertainty bands.

FAQ

Frequently asked

Methodology that matches the enterprise vendors (PyMC-Marketing backbone, full diagnostic suite, reproducibility guarantees), priced for the Shopify-DTC mid-market, and packaged for analysts and operators on the same platform. Bayesian Structural and a skill-based AI agent ship next; v1 is Traditional MMM today.
One click. A geo lift estimate from the Experiments module attaches as a Bayesian prior on the corresponding channel coefficient in the next MMM run. The MMM engine treats it as a calibration target — exactly the loop sophisticated marketing-science teams build by hand today.
No. The Operator tier is a conversational AI agent that drives the Traditional and Bayesian MMM tools on your behalf — it loads the data, picks adstock and saturation defaults, runs the model, and explains the result in plain language. Honest about uncertainty: it'll tell you when your data can't support an answer and recommend a geo experiment.
For MMM: weekly or daily marketing spend by channel, conversions or revenue, plus controls (price, distribution, seasonality, macro). For Customer Knowledge: customer-level transaction history. CSV upload today; Shopify, Klaviyo, Meta Ads, Google Ads, TikTok Ads, GA4 connectors land post-v1.
Supabase EU region by default. Encrypted at rest, scoped per-workspace, never used to train shared models. DPA available before public launch.
Yes. Every run is exportable as a CSV bundle (matching the established analyst output schema) plus the model artifact, posterior samples (Bayesian runs), and a reproducibility lockfile. There's a public API on the Pro tier so you can pipe outputs into your warehouse, BI tool, or activation system.
Two seat tiers — Operator (cheaper, conversational-only) and Analyst (full modelling UI). Bundled monthly allowance of agent tasks and model fits per seat. API and integration capabilities sit as a separate add-on. No outcome-based pricing, no hidden usage bills. See the pricing page for ranges.
The methodology is. We use PyMC-Marketing for Bayesian, scikit-learn / statsmodels for Traditional, GeoLift for synthetic-control geo, EconML for HTE. All open-source, all referenceable in the literature. The Kleyr platform layer (data ingestion, run management, agent skills, app surfaces) is proprietary.
Book a working session

Bring your spend, leave with a model you can defend.

30 minutes with the Kleyr team. We model your last 12 weeks of marketing spend live on the call — no slides, no abstract demo.

  • Walkthrough of your decomposition, by channel.
  • Saturation curves you can pressure-test on the call.
  • Geo-lift design tailored to your top spend channels.
Prefer email? hello@kleyr.co.uk

We only use your details to set up the session. No marketing automation, no third-party tracking.

The answers
compound.

Start with a working session30 minutes · your data · no slides