Consider a wholesale distributor: around 900 active accounts, 6,000 stock items, eight people in sales, three years of Dolibarr history. Revenue is stable. Nothing is obviously wrong. Management's instinct is that growth requires more marketing spend and another salesperson.
The assessment week. Rather than start building, we spend a week reading the data. Purchasing patterns, stock movement, quotation history, payment behaviour. The goal is to find the three questions worth answering — not to demonstrate technology.
What tends to surface. A cluster of accounts that reorder on a predictable cycle and have quietly passed it. A group of SKUs whose reorder points were set at launch and never revisited, now driving both overstock and stockouts on the same shelf. Quotations from good customers that were never followed up because nobody owned the list.
What gets built first. Not everything. One workflow — usually the lapsed-account list, because it is the fastest to verify and the easiest for a sales manager to act on. It runs against real data, and the sales team checks whether the accounts it flags are genuinely worth a call. If they are not, we have learned that in three weeks rather than six months.
What production looks like. The assistant is available to managers who never had direct ERP access. The daily brief lands each morning. The inventory module flags stock health continuously. None of it can write to the ERP, and all of it can run on the company's own server.
The point of this scenario is the sequence, not the numbers: read the data first, prove one workflow, then build. It is deliberately the opposite of buying a platform and hoping it finds something.