Customer value

Customer lifetime value, churn and product recommendations

Models that decide who to acquire, who to keep and what to show them: predicted value used for activation, churn scored where an action still exists, recommendations that stay auditable even when an LLM is involved.

What this usually looks like

  • Acquisition is optimized on cost per install, and the installs do not become customers.
  • The CLV model has a good R² and the CRM team cannot use it for anything.
  • Churn is predicted a week before it happens, when nothing can change it.
  • The recommender is a black box, and merchandising has stopped trusting what it ranks.

What you get

Value scores built for the decision

Directional where ranking is what matters, calibrated where a euro figure is. Wired to the channels that act on them: audiences, CRM, budget.

Churn and uplift, not just propensity

Who to contact because contact changes the outcome, with holdouts that prove it.

Recommendations under control

Business rules first, model second, LLM only where it earns its place. Every ranking explainable to the person who owns the catalogue.

The handover

The scoring pipeline, monitoring on the metrics that move, and documentation of every threshold.

Depth

Jun 18, 2026

For Activation, a Directional LTV Model Beats an Accurate One

A staff-level field guide to predicted LTV for marketing activation: why ranking beats accuracy, why a directional model can beat an accurate one for seeding audiences, and where the whole thing quietly goes wrong.

Aug 15, 2026

Pourquoi un bon modèle pLTV peut avoir un mauvais R²

Pourquoi le R² évalue souvent mal un modèle de predicted Lifetime Value destiné à l'activation marketing, et quelles métriques utiliser pour piloter les segments en production.

Mar 2, 2026

From Propensity Scores to Learning Systems: What Most CRM Teams Get Wrong

Most CRM organizations don't have a feedback loop. They have a broadcasting system. The shift from propensity to uplift modeling changes everything.

Apr 12, 2026

Building a recommendation engine that doesn't trust the LLM

This is the engineering companion to the production architecture piece. Instead of re-arguing why open-ended agents are risky in commerce, it walks through the actual implementation choices in `ai-florist`: FastAPI boundaries, LangGraph orchestration, pgvector retrieval, learned scoring weights, deterministic fallbacks, and runtime observability.

Aug 15, 2026

Building a Customer-Brand Affinity Engine for Physical Retail

A technical architecture for customer-brand affinity in physical retail: customer windows, lift and PMI, Item2Vec embeddings, graph construction, constrained layout optimization, and causal validation.

Jul 28, 2025

Double Jeopardy in marketing: Why market leaders win twice

Understanding the mathematical law that explains why large brands not only have more buyers but also enjoy higher loyalty—and how it reshapes marketing strategy

How it starts

Diagnostic first, build second.

01 · Decision System Diagnostic · From €12,000

Three months, one critical decision. We establish which customer decision the model serves, measure the baseline, test what the data can support, and design the operating cadence. You get evidence, a target architecture, and a costed roadmap.

02 · Design and build

Scoring models, the activation logic, the holdout design, and the pipeline that runs them. Scoped from the roadmap, priced on the engagement.

03 · Operate and hand over

Monitoring, drift controls, documentation, and the training to run it without me.