Methodology
How the HMRC enquiry-risk model works.
The enquiry-risk score is trained exclusively on publicly-available HMRC enforcement disclosures. It is designed to be reviewable — every score is accompanied by the drivers that produced it.
Data sources
The model draws on the HMRC published deliberate defaulter lists, enforcement notices, appeal outcomes, tribunal decisions and other statutory disclosures. No privileged data is used and no client-identifying data leaves the practice.
Score construction
Each filing receives a 0–100 enquiry-risk score alongside a confidence indicator. The score reflects the modelled probability of an HMRC enquiry given the filing's structural features, cross-filing consistency and comparability to enforcement precedent.
Explainability
Every filing detail view surfaces the top three drivers of the score with a plain-English tooltip. Partners can see what would need to change to reduce the score, and can dismiss a driver with a captured reason for the audit log.
Model governance
The model is retrained quarterly. A model card is published to this page with each retrain, describing input features, holdout performance and calibration.