Customer identity across loyalty, web, and till; loyalty data quality; basket analysis; and what has to work in the data layer before personalisation or recommendations projects.
Loyalty, CRM, e-commerce, and the till often hold different identifiers for the same person. Until those records resolve to one customer, segment counts, churn, and “omnichannel” revenue are not comparable — marketing and finance will argue about the same headline from different bases.
This is a linkage and ownership problem, not a dashboard shortage. Transaction lines from retail data management must be trustworthy before you attach them to a person.
One person, many keys — Card number, email, mobile, guest checkout. Without merge rules and an owner for duplicates, base size and campaign reach are inflated.
Loyalty built for redemption — Points tables are not designed for lifetime value or clean segmentation. Useful attributes need transactions joined on consistent product and customer keys.
Consent and suppression — Use cases that need opt-in must respect it in the systems that run campaigns, not only in policy PDFs.
Scale — A small shop may run on a single loyalty ID at the till; a chain has duplicate rates in the double digits and CRM, loyalty, and web owned by different teams.
Decide the primary key, who approves merges, and which attributes win when records collide. Deterministic matches (same loyalty ID + email) and fuzzy matches (phone + postcode) need governance — not only an algorithm.
Enrolment errors, cards never linked at sale, and duplicate enrolments are routine. Someone should own hygiene: how often duplicates are merged, and whether inactive cards stay in the denominator.
You will never identify every cash sale. You should still know what share of revenue is identifiable and whether that share is improving. Web and store need the same customer key where technically possible.
Affinity and category migration need line-level sales with consistent product coding. RFM or similar segments need agreed definitions of sale, return, and net revenue — the same definitions finance uses.
Recommendation or personalisation projects fed from unresolved identity — models interpolate across the wrong history.
Lift measurement broken by duplicate keys or overlapping campaigns — attribution becomes storytelling.
Different definitions of “customer” or “lifetime value” — dashboards disagree because nobody fixed the metric, not because the tool failed.
Personalisation and advanced analytics need a governed identity layer and clean transactions first; see AI Readiness for the executive view. Inventory and demand planning often use the same sales history — the same data-quality issues show up there.