Re-screen the customer base on a list change
Onboarding screens new customers; re-screening protects the ones already there. A designation is retroactive to the relationship, so a changed list creates an obligation on a book of business that a point-in-time check never touches.
Procedure
- Take the delta from the list refresh, not the whole list, and keep the row so the trigger is reproducible:
diff <(sort prev.csv) <(sort sdn.csv) | grep '^>' | cut -d, -f1-2 > delta.csv
wc -l delta.csv
Run the delta against the customer index within the internal SLA of the list landing. Whole- file re-screens are for the periodic cycle; delta screens are for timeliness.
Match on address for wallet-linked customers and on identity fields for others. For addresses, exact match only. For names, the identity fields from KYC are what disambiguate:
python3 - <<'PY' import csv delta = {r[1].lower().lstrip("0x") for r in csv.reader(open("delta.csv")) if len(r) > 1} cust = {r[0].lower().lstrip("0x"): r[1] for r in csv.reader(open("cust_addrs.csv"))} hits = [(a, cust[a]) for a in delta & cust.keys()] print(hits or "no exact address hits") PYRun the periodic full re-screen on its own cadence (commonly daily to monthly by risk tier) and record the coverage: how many customers were in scope, how many screened, how many skipped for missing data.
Route hits to the escalation queue as new alerts, not as updates to a closed case. A designation on an existing customer is a new event with its own clock.
Freeze or restrict per policy on a true match, and record who made the call and when.
Pitfalls
- Screening only new signups and calling the programme current. Regulators test exactly this: whether the book, not just the funnel, is covered.
- Re-running the whole list nightly and reporting a huge alert count, which buries the one fresh designation in old false positives. Delta-first keeps the signal.
- Skipping customers whose stored identifier is a name only, which is the segment most likely to be missed and therefore the boundary you must report.
- Failing to record coverage numbers, so no one can tell whether the re-screen ran or ran partially.
- Closing a reopened case by editing it; version the case so the timeline of the list change and the decision is legible.
Verification
awk 'END{print NR-1" customers screened"}' cust_addrs.csv
grep -c "^" delta.csv
Compare the delta row count to the alert count produced; a delta with rows and zero alerts means the join key is wrong, not that nothing matched.
Report the delta size, in-scope and screened counts, skipped customers with the reason, and hits raised. Determining what a hit obliges you to do is a compliance and legal decision for a qualified officer.