Behavioural analytics · Customer Knowledge

The people behind the numbers.

Customer lifetime value, segmentation, churn, propensity, and forecasting — in one model that knows which customers your media is actually reaching. Reconciled to MMM elasticities so customer-level forecasts and channel-level decompositions tell a coherent story.

BTYD CLVHDBSCAN clusteringCalibrated churnFast Forwards forecaster
What's in the box

One graph, four modelling layers.

01
01
Customer lifetime value

BTYD-family models — Pareto/NBD, BG/NBD with Gamma-Gamma — implemented via PyMC-Marketing. Bayesian uncertainty, hierarchical pooling across segments, the canonical method since 1987.

02
02
Behavioural segmentation

HDBSCAN as default — density-based, doesn't require choosing k, robust to non-spherical clusters. RFM as the interpretable floor. Longitudinal segmentation tracks customers across cohorts.

03
03
Calibrated churn

LightGBM workhorse classifiers with isotonic / Platt calibration. Reliability diagrams shown by default — the calibration matters more than the AUC. Survival analysis when time-to-churn matters.

04
04
Per-customer explanation

Why is this customer at risk? Surface the specific behaviours and product relationships that drove the score. Not a black-box prediction — an auditable inference.

05
05
Fast Forwards forecaster

Forward-looking scenario simulation reconciled to MMM elasticities via MinT (Hyndman et al. 2011). When you change the media plan, the customer-level forecast updates coherently.

06
06
Segment exports

Send a segment to MMM as a covariate, to Experiments as an audience definition, or out via the public API for activation. Same segment object, every direction.

Methodology

Canonical methods, current implementations.

The methods are mostly four decades old. The implementations are not.

Pareto/NBD CLV
Schmittlein, Morrison & Colombo · 1987
Original BTYD model. Hardest to fit, most flexible.
BG/NBD
Fader, Hardie & Lee · 2005
Easier to fit, near-identical accuracy. The practical default.
Gamma-Gamma monetary
Fader & Hardie · 2013
Pairs with BG/NBD for revenue forecasts.
HDBSCAN
Campello, Moulavi & Sander · 2013
Density-based clustering. Robust to non-spherical clusters and noise.
Calibration
Niculescu-Mizil & Caruana · 2005
Reliability diagrams as the trust signal — not just an AUC number.
Hierarchical reconciliation
Hyndman et al. · 2011
MinT — Fast Forwards reconciles customer- and channel-level forecasts.
FAQ

Customer Knowledge, in detail.

No. Customer Data Platforms (Segment, mParticle, Hightouch) ingest, resolve, and activate customer data. Kleyr Customer Knowledge consumes resolved customer data and gives you the analytics layer on top — CLV, segmentation, churn, forecasting. We accept resolved IDs from CDPs as input.
Deterministic-first for v1 — email normalisation, customer-id matching, household roll-up. Probabilistic / cross-source resolution comes in v2 once we have customer signal that requires it. Most DTC customers solve this with their CDP or e-commerce platform.
Fast Forwards is a customer-level forecast that pulls MMM channel elasticities as constraints. When you simulate a media plan change in MMM, the customer-level forecast updates coherently — same elasticity drives both. Reconciliation via MinT.
Conceptually yes — customers, behaviours, products, and segments are nodes; relationships are edges. Implementation in v1 is relational tables in Postgres + NetworkX in-process for graph queries, not a separate graph database. Honest framing: most of the value is from CLV / clustering / churn over relational data; the graph framing matters for retrieval-augmented agent reasoning, which lands in v2.
✦ Get started

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