Clubcard data-quality lesson — multiple users on one card produced false positives in mining (the "garbage in" warning from the best-documented winner)

Claim. Even in the best-documented loyalty-card win, data-quality failures bit the analytics: multiple distinct shoppers using one Clubcard produced false positives in segmentation and propensity mining.

Source. mugleston.co.uk summary of Humby et al., Scoring Points (2004), accessed 2026-06-21.

Confidence. Verified — documented in the participant-authored book itself; a candid acknowledgement, not an external criticism.

Caveats. Documented as anecdote rather than measured error-rate; the precise size of the bias is not stated.

Implication / use. Concrete instantiation of the "garbage in" principle for any information-edge argument. If it happened to Tesco at Clubcard's analytical sophistication, expect it for any SMB. Supports Rule: three thresholds before claiming an information edge — volume to clear noise, clean data, an actual decision someone will act on (and timeliness before commoditisation) (clean data is the second of three thresholds).