## Black-Box and Smartphone Telematics: Two Data Architectures, Different ML Problems
UK motor telematics insurance splits into two distinct data architectures with meaningfully different ML characteristics. Black-box (OBD-plugin or professionally fitted) devices provide high-fidelity, tamper-resistant accelerometer, GPS and CAN-bus data at consistent sampling rates, making them well suited to detailed driving-event classification (harsh braking, cornering g-force, acceleration profiles) with relatively low noise. Smartphone-app-based telematics, now the dominant model for UK new-business quotes given lower customer friction and no hardware/fitting cost, introduces materially harder ML problems: distinguishing driver from passenger phone usage, filtering GPS drift and accelerometer noise from a device not rigidly mounted to the vehicle, and — critically for fair pricing — reliably detecting whether the policyholder was actually driving versus a passenger in someone else's car, since misattributing a passenger's smooth ride or a taxi's harsh driving to the policyholder directly undermines pricing fairness. UK insurers running smartphone telematics report materially higher false-attribution rates than black-box programmes, which has driven continued ML investment specifically in driver-detection classifiers as a precursor to any behaviour-scoring model.
## Driving-Behaviour Scoring Models and Real-Time Pricing Adjustment
Core behaviour-scoring models in UK telematics motor insurance typically combine harsh-braking and harsh-acceleration event frequency, cornering g-force, speed relative to posted limits (increasingly cross-referenced against UK speed limit map data rather than assumed road-type defaults), time-of-day and day-of-week exposure (night driving carrying materially higher risk weighting given UK road casualty statistics skew heavily toward night-time hours), and mileage/exposure normalisation, since raw event counts without exposure normalisation systematically penalise higher-mileage drivers regardless of actual skill. The move from static annual telematics scoring toward real-time or monthly pricing adjustment — increasingly common among UK usage-based insurance (UBI) providers — raises the stakes on model calibration: a policyholder whose price adjusts monthly based on a rolling behaviour score needs that score to be genuinely stable and explainable, not an artefact of noisy month-to-month event counts, since sharp unexplained premium swings are precisely the kind of customer outcome the FCA's Consumer Duty was designed to catch.
## FCA Consumer Duty: The Fair-Pricing Lens on Telematics Models
The FCA's Consumer Duty, in force since July 2023, requires firms to demonstrate that pricing outcomes are fair and that customers are not paying disproportionately for equivalent risk or receiving genuinely poor value — a standard that applies directly to telematics-based pricing models. In practice, this means UK motor insurers deploying behaviour-scoring ML must be able to show: that the model does not produce disparate pricing outcomes correlated with protected characteristics under the Equality Act 2010 even where those characteristics are not direct model inputs (age is a particular focus area, since young driver telematics programmes are the single largest UBI segment in the UK market and age-correlated behavioural proxies require careful fairness testing); that customers understand in plain language what driving behaviours affect their price and can access their own behaviour data; and that a genuinely poor telematics score does not push a customer into non-renewal or a large premium increase without adequate notice and a route to query or appeal the scoring. The FCA has signalled continued supervisory attention to algorithmic pricing more broadly under Consumer Duty's price and value outcome, and telematics-scored motor insurance is a natural area of focus given its scale and the sensitivity of driving-behaviour data.
## UK GDPR and Driver Privacy: Continuous Location Tracking as Special-Category-Adjacent Risk
Continuous GPS location tracking inherent to telematics insurance sits at an uncomfortable intersection of UK GDPR obligations even though driving location data is not itself a special category under Article 9. The core obligations insurers must satisfy: a clear lawful basis (typically contractual necessity or legitimate interest, with the latter requiring a documented legitimate interests assessment given the privacy-invasive nature of continuous tracking); data minimisation (Article 5(1)(c)) — retaining raw GPS trace data only as long as necessary for pricing and claims purposes rather than indefinitely, since aggregated behaviour scores, not raw trip-by-trip location history, are what pricing models actually require; transparency about exactly what is tracked, for how long, and who can access it, particularly relevant where family members share a vehicle and one household member's driving is scored under another's policy; and a defined data subject rights process, since telematics customers frequently exercise subject access requests specifically to see their own scored driving data, which insurers must be able to produce in an intelligible form rather than raw sensor logs. The ICO has previously flagged continuous location tracking generally as warranting a higher bar of transparency and necessity justification than typical customer data processing, a standard telematics insurers should treat as the operative benchmark even absent insurance-specific ICO guidance.
## Building a Defensible Telematics ML Programme
A UK motor insurer or MGA building or refreshing a telematics scoring model should treat fairness testing, explainability and data minimisation as first-class model requirements alongside predictive accuracy, not as a compliance layer bolted on after the model is built — retrofitting fairness testing onto an already-deployed pricing model is materially more disruptive than designing for it from the outset, particularly once real customer pricing history depends on the existing score. Codeworth's machine learning engagements with UK motor insurers and MGAs in this space typically combine driver-detection and behaviour-scoring model development with a parallel fairness and Consumer Duty documentation workstream, ensuring the pricing model that reaches production already carries the disparate-impact testing, explainability documentation and data minimisation design the FCA and ICO would expect to see — distinct from the broader underwriting risk model work Codeworth supports across general insurance lines, since telematics scoring carries its own specific data architecture, real-time pricing cadence and continuous-tracking privacy considerations that a general underwriting model does not.
Related Codeworth Services
Need Help with Implementation?
The Codeworth team is ready to implement these ideas in your project. Free consultation.
Discuss the ProjectCodeworth team specialist.