How the work originally happened
More than one hundred workers reported their production through WeChat groups. Two clerks then counted, checked and consolidated the messages line by line. The reporting already happened; the waste sat in rebuilding the result every day.
The company did not hand me a formal system requirement. I noticed the pattern in the field, concluded that production data was not entering a usable management loop, and proposed a workflow for parsing, cleaning, review and aggregation.
I did not replace WeChat first
Forcing every worker to learn a new entry point would have increased adoption cost. I kept the familiar WeChat flow. In the current version, an operator manually triggers the Codex workflow, after which parsing, cleaning, review and aggregation run automatically.
The back end needed to:
- identify and split reporting messages;
- organise them by product, worker and date;
- hold uncertain or abnormal entries for human review;
- aggregate results according to operations and finance rules.
The important choice was not the technology. It was placing change where it created the least disruption for the frontline.
The harder part was aligning rules
Operations, finance and engineering did not speak about the same data in the same way. Operations needed reporting and exception rules. Finance needed a reliable counting basis. Engineering needed explicit inputs and conditions that should not be guessed automatically.
I worked through those questions with each group and made the open decisions and responsibilities visible. Projects often stall not because the code is impossible, but because the problem remains vague or nobody keeps following up on unresolved details.
What the data eventually supported
The system first connected the daily report. The same data then supported:
- machine capacity;
- daily exceptions;
- finance-required reporting.
That moved operations, management and finance toward a shared counting basis instead of separate manual reconstructions.
What this case does not claim
I do not claim a percentage productivity gain or a precise labour saving because there is no complete public before-and-after measurement. Two clerks doing the original consolidation does not mean the system simply replaced two jobs.
The confirmed value is narrower and more useful: a repeated operational burden was identified, the familiar worker entry point remained, cross-functional rules were aligned, and the same data basis extended into daily output, capacity, exception and finance-required reporting.
Related: No one asked for a system and ways to work together.