74,000 customers up to date every morning

Context

Several business applications read the same customer base. Every morning around 5:37 a.m., a full extract from the management software lands in a mailbox: about 74,000 rows.

The problem

The reference base, however, had been frozen since April 2026: 26,303 rows. The teams were working with outdated customer data.

What we built

Three automations:

• the first receives the morning file;
• the second compares it with the current base and writes only the differences (additions, changes, deletions), in batches, with safeguards: expected headers, row count, deletion rate;
• the third keeps watch at 8 a.m.: if nothing has run, it raises the alarm.

Every morning, a summary goes out by email: the additions, the changes with the columns that changed, the deletions. Or “No changes”.

New customers are enriched from the public registers: legal form, head office, registry city, share capital. The existing base was enriched in a few days: 24,615 distinct company numbers.

Result

First load on 21 October 2026: 74,276 rows, zero errors. First real morning on 22 October 2026: 74,465 rows read, 6 additions, 11 changes, 1 deletion, in 3 min 33 s.

The applications are switching to this base one by one: the equipment request form from 23 October.