This is not a textbook case. It is a technical services SME in French-speaking Switzerland, with a sales team, a back-office team, technicians in the field and an ERP already in place. A business that is doing well, receives plenty of enquiries, and was starting to lose some of them along the way.
We have been working with them for several months. Here is what we built, in what order, and what the measurements show over a full month. The company's name and the details of its trade remain confidential.
The starting point
The initial diagnosis looked like that of many SMEs:
- Enquiries from everywhere: website form, advertising campaigns, phone, social media. Each channel landed in a different place.
- Handling times that were far too long. Between a lead arriving and a salesperson's first action, an average of 216 hours went by, or nine days.
- An ERP that was reliable for the back office but avoided by sales staff, who kept their follow-ups elsewhere.
- Alerts nobody read. The existing system sent a notification for every event; the actual response rate was below 7 %.
- No reliable view of how profitable the advertising channels were.
What we built
Step 1: a CRM connected to the ERP
Rather than replacing the ERP, we built a sales layer on top of it, synchronised both ways every hour: activity, signatures, amounts, stages, owner, invoices, lost deals. With strict safeguards: an amount validated in the ERP is never overwritten, and any bulk correction goes through a simulation before approval.
Leads from every channel now arrive in one place. They are cleaned and de-duplicated, then automatically assigned to the right salesperson by area, service and language, taking absences into account. Every morning, each salesperson sees their "Top 3 to call", based on urgency, value and likelihood of signing.
Step 2: a price estimator calibrated on real data
Prospects want a ballpark figure before booking a meeting. We built an instant estimator, with an interactive map, calibrated on 241 of the company's real quotes. Not a theoretical price grid: the prices the company actually charges.
Step 3: AI agents that prepare, humans who approve
Around forty AI agents run today, for roughly 64,000 automatic executions a month. The main ones:
- Inbox reading: when a client replies to a salesperson, the agent recognises the deal and updates the record within 3 to 14 minutes.
- Signature and payment reminders at fixed intervals, using templates approved by management.
- Automatic review requests at the end of a job, with spaced follow-ups.
- Monthly lost-deals report written in plain language.
- Regular CRM / ERP comparison to detect discrepancies before they pile up.
The dozens of daily alerts were replaced by a single, capped digest, with the few truly priority actions and a button to act directly from the email.
Step 4: a client portal and a field app
Clients have a personal portal with no password (magic link login): their project roadmap, documents to sign, invoices, messages. Technicians have a mobile app showing their next job, the route and photos, which are automatically filed in the right folder.
The results, measured over one month
We measured the effect over a full month of activity. Here are the figures, with no flattering rounding.
67 hours of work saved per month. Record updates, follow-ups, reports, lead sorting, document filing: tasks that required no human judgement and that nobody has to do any more. On top of that comes around 29 hours of new capacity: tasks that simply were not done before, for lack of time (systematic follow-ups, review requests, analysis reports). Supervising the whole system takes about 4 hours a month.
Time to first action on a lead: from 216 h to 95 h, a 56 % reduction. Sorting, automatic assignment and the daily "Top 3" did most of the work. The work continues: 95 hours is still too long, and it is the next project.
Website leads sign 2.3× more often than those from paid social media (16.9 % versus 7.2 %). This figure was invisible until every signature was linked back to the lead's source.
Around CHF 20,000 a year of unprofitable advertising budget identified and cut. A direct consequence of the previous point: some campaigns generated volume but almost no signatures.
What really made the difference
Three choices, with hindsight:
- Not replacing the ERP. The back office did not change tools. The CRM plugged into it. Adoption by the sales team was immediate because the tool saved them time from day one.
- Measuring response, not sending. The switch from alerts to a digest came from one simple measurement: a response rate below 7 %. Without it, we would have kept adding alerts.
- AI never talks to a client alone without a framework. Everything that goes out to a client follows an approved template; anything uncertain remains a proposal a human approves in one click.
And for your business?
Every SME is different, and these figures are not a promise. They show what you get when you start from how the business really works, connect to what already exists and measure honestly.
The details of this project, with interface mock-ups, are presented on the client case page. To explore all our work in business software and automation, visit AI & Software. And if you would like to know what could be automated in your company, contact us.