Your marketing director watches the MDM dashboard report 97 percent match rate six weeks after go-live. The figure comes from the initial batch load of legacy CRM and billing tables, not the daily stream of new accounts created inside the mobile app.
The matching engine was configured with exact name-plus-postcode rules and a handful of fuzzy thresholds that worked for the migration weekend. App registrations carry device IDs, partial addresses and promotional codes instead, so the system treats every signup as unique.
Segmentation models downstream now split the same household across four records because the app never writes back a stable identifier. Campaign spend rises while response rates drop, yet the MDM team keeps reporting the original match percentage as the success metric.
Finance starts questioning why the same customer receives three different renewal offers in one quarter. The root cause sits in the ingestion layer that never enforced a probabilistic match on the new data source.
Resetting the rules mid-stream requires a change request that the vendor prices as a new implementation phase. The marketing team therefore keeps feeding the same broken graph into every AI experiment that follows.
The pattern repeats whenever any operational team introduces a fresh front-end without updating the matching contract. One frozen configuration becomes the permanent ceiling on data quality and model performance.