Services

Every service mapped to a Halfords P&L line.

We don't sell models. We ship the working component of a Halfords next best action system — Autocentres, Motoring Club, cycling, retail, EV.

Autocentres

Problem at Halfords: The MOT diary is the single most valuable predictable event in the Halfords customer file — and today most reminders go out on a calendar rule, not a probability.
Approach: Time-to-event model on registration data, service history, garage capacity and postcode competition. Gradient-boosted propensity with a survival head predicts the day booking probability peaks.
Output: Per-vehicle daily booking probability and the optimal contact window per channel.
Decision use: Feeds the MOT reminder engine — right customer, right week, at the Autocentre with capacity to fulfil.

Motoring Club

Problem at Halfords: Motoring Club is the retention flywheel — but conversion from free identified customers to paid tiers is still driven by generic promo windows.
Approach: Multi-task model predicting free→paid propensity and 24-month CLV conditional on tier. Pareto/NBD-style value curve blended with an ML head on visit and service frequency.
Output: Per-customer conversion probability, tier-level expected value, and the offer with the highest expected margin.
Decision use: Personalises Motoring Club nudges in the app, on receipts, and at checkout in stores and Autocentres.

Retention

Problem at Halfords: Subscription churn is opaque until the renewal fails. Discounting every at-risk member destroys the P&L; discounting none loses them.
Approach: Survival churn model with SHAP reason codes (unused benefit, service friction, competitor exposure). Uplift model matches each at-risk member to the save action with the highest net expected value.
Output: Churn probability, expected time-to-churn, top 3 drivers, and a matched save offer.
Decision use: Triggers a retention playbook — sometimes a bill credit, often a service benefit reminder, sometimes nothing at all.

Cycling

Problem at Halfords: Halfords sells a bike; the customer disappears. The 6- and 12-month service is the single biggest missed revenue moment in the cycling business.
Approach: Post-purchase propensity on bike SKU, ride season, weather and prior visits. E-bike battery-health model for the £2k+ segment.
Output: Service booking probability by month and the specific service (safety check, gears, brakes, battery diagnostic) most likely to convert.
Decision use: Drives outbound to the customer's local store with pre-slotted Bike Care appointments.

Cross-sell

Problem at Halfords: Winter tyres, wipers, bulbs, roof boxes, dashcams — high-margin retail attach is timed by campaigns, not by the individual customer.
Approach: Product-affinity embeddings over 3 years of transaction history, seasonal decomposition, and weather-conditioned demand.
Output: Per-customer per-product buy probability and the optimal week to surface it.
Decision use: Personalises email, app, in-store handovers, and the Autocentre inspection upsell script.

Problem at Halfords: The EV base is small but strategically decisive. Identifying ICE customers about to switch is worth more than any single retail transaction.
Approach: Composite signal: vehicle age, service spend trajectory, postcode EV adoption, financial signals from Motoring Club, and behavioural signals from EV content engagement.
Output: Probability of EV purchase in the next 12 months, and the pre-transition service basket (home charger, tyres, service plan).
Decision use: Cross-sell charger installation and lock in the after-sales relationship before the vehicle changes hands.

Causal

Problem at Halfords: Halfords runs many concurrent campaigns; separating what the campaign caused from what would have happened anyway is hard.
Approach: Two-model X-learner and causal forests over randomized holdouts. Incrementality reporting on every deployed model, not just at pilot.
Output: Individual treatment effect — the lift caused by the offer, per customer per action.
Decision use: Suppress contact to sure-things and lost causes; concentrate spend on true persuadables.

Decisioning

Problem at Halfords: Halfords has a data platform, models, and a CRM — but the CRM still sends by campaign, not by customer.
Approach: Real-time scoring service + constraint solver (Autocentre capacity per postcode, contact cap per week, channel affinity, margin weighting).
Output: A single ranked action per customer per day with full audit trail and override hooks.
Decision use: Direct integration into the Halfords app, email, contact centre, Autocentre booking flow and in-store colleague tooling.

Where would we start at Halfords?

A two-week diagnostic across MOT reactivation, Motoring Club conversion, and bike servicing — sized by expected uplift.