The story
Last month I sat down with a finance manager at a mid-sized trading company. Her desk was spotless — Scandinavian minimalism, zen garden vibes.
One laptop and two screens. And on them? Chaos. Three systems side by side: ERP on the left, an accounting package that looked like it had last been updated during Bill Clinton heyday, and a cloud CRM that was modern enough to be smug but not too competent. And none of them agreed.
Her job? To make them agree. Which is corporate code for _“fight entropy with Excel.”_
Every invoice had to be checked against payments in the ERP, then matched with entries in the accounting system, and finally compared with orders in the CRM. Each time the numbers didn’t line up — which was often — she’d dive into email chains, call suppliers, or trawl through exported spreadsheets formatted like ransom notes.
Those were three machines speaking different dialects of the same truth. In binary, in bureaucrat, in chaos theory. And the interpreter was her. Hours a day. Weeks a month.
The solution
- Decide which data source is the golden source. It can be one of the sources or the golden source could be decided dynamically based on criteria like last updated, most complete source and so on.
- Do the matching between records based on the relevant criteria like amounts, customer name, document number, and attempt to fill the missing fields inferring info from previous records and previous matches.
- Dive into the documentation trail by summarizing emails, and reading knowledge bases and procedures.
- When in doubt, the machine should show the reasoning to the user, assemble the relevant data and ask for guidance.
- Add the corrections as journals to show when and by whom the corrections were made. If the database schema or the end-user app allows it add a column with the correction reasoning, if not, keep the whole journal with the adjustments history for audit and future inferences.
- Always start small, and build trust on your way up!