A CRM for the sales team of a food manufacturer
Before, the sales team kept customers and orders in Excel, typed by hand. Here is what we built, how each part works and what we learned.
The starting point.
The business lives on repeat orders from shops, restaurants and distributors.
The sales reps visit customers in person and work from their phones.
Many orders come in as a photo of a list or a WhatsApp screenshot.
Customers and orders were kept in Excel, typed by hand. The company's system could only export reports.
Every morning, a list of who to call, in the right order.
Each customer has a buying rhythm: how often they usually order. When they pass it, they move up the list, and the longer they wait, the more the call is worth.
Move the days forward and watch the list change.
The list puts first the calls most likely to bring an order today.
"Worth" is the customer's usual order, counted higher the later they are (up to three times).
An AI assistant that looks things up, records sales and asks before anything important, like marking a customer as lost.
The rep types, talks or sends a photo. The assistant uses the same steps as the screen, and every number it gives comes from the CRM, not from the AI.
It asks, the rep decides.
The system refuses, whoever asks: the rep, the screen or the AI.
Look up and record sales, visits, notes and sales stages.
A photo of an order becomes an order to check.
The rep attaches a photo, a WhatsApp screenshot or a PDF. The AI reads it next to the product list. What it can't read stays blank, prices are never guessed, and the rep checks before saving.
Order · Thursday
Product A — 12 units
Product B — 6 units
Prod∿∿∿ — ∿ units
Product C — 8 units
Deliver Monday
- #04Product A12 units
- #17Product B6 units
- —Not readable, left blank?
- #09Product C8 units
An old system with no connection: every total is checked before saving.
The company's system could only export reports. The CRM reads each one, adds up every line and compares the sum with the report's total. If they don't match, it stops and saves nothing.
The totals match, so everything is saved.
Diagram with example figures: this step runs in the background and has no screen.
The project in numbers.
Source: the project's database, 3 July 2026.
What we learned.
From day one we plan three things: bringing in the data, the first weeks of follow-up and an owner on the client's side.
In our next projects, the first week records calls per week, quiet accounts and hours of retyping, so there is a clear before and after.