Case study · Engineering & technical studies · Switzerland
A real engagement, client anonymised. Figures measured with the client and rounded; not audited by a third party.
A 180-person engineering firm puts a figure on what AI brings in: CHF 180,000–220,000 a year
After 18 months of AI tools nobody had measured: five processes tracked, a decision on each of seven tools (2 stopped, 3 adapted, 2 extended) and savings validated by finance.
Scope and limitations
Rounded figures. A real engagement: the client’s name and any detail that would identify it have been removed.
- Company
- Swiss engineering and technical studies group, around 180 people, 4 offices
- AI tools in place
- Office copilots and an internal document assistant, in use for 18 months
- Amount committed
- Around CHF 300,000 over 18 months (licences and projects), with no measure of results
- Starting point
- No measurement before the tools, no indicators, usage statistics scattered across vendors
Starting situation
Eighteen months after rolling out office copilots and then an internal document assistant, the group had committed around CHF 300,000 in licences and projects without a single results indicator. Feedback was anecdotal: some staff said they saved hours, others had quietly stopped using the tool. When the board asked what AI was bringing in, management had nothing to show.
Two camps had formed. Finance, seeing no evidence of return, wanted to cut the budgets; the teams, convinced by their own experience, wanted to extend the tools. Neither side could prove its case: nothing had been measured before. Usage statistics sat with the vendors, and billable time, the one measure everyone understood, was too sensitive to handle carelessly.
The question asked
For each tool: extend, adapt or stop. But on what evidence?
Constraints
No before-measurement
Nothing had been measured before the tools arrived. Handling times from that period existed nowhere: they had to be reconstructed without being invented.
Scattered usage statistics
Usage figures sat with the vendors, in poor, inconsistent and sometimes missing exports.
Measure without monitoring people
The system had to measure processes, never individuals. That limit was set on day one.
Billable time is sensitive
In a firm that lives from billed hours, any measure touching time is delicate. Every indicator was designed with that in mind.
What we put in place
We added no tool: we built the measurement that was missing. For five important processes, we linked AI use to a verifiable effect, reconstructed the situation before, and set in advance (before seeing any figures) what would justify extending, adapting or stopping.
The building blocks
Stages of the engagement
Analysis and choice of processes 3 weeks
Inventory of the seven AI tools and projects under way, choice of 5 important processes, use → result links built with business leads, first decision criteria agreed with the managing and finance directors.
Before-situation and measurement 5 weeks
Earlier handling times reconstructed from samples and historical data, staff representatives informed, per-process recording put in place, decision rules set with management.
Two months of measurement 8 weeks
Two full monthly cycles: recording, dashboard, reviews with management, decisions prepared and documented for all 7 tools, including two deliberate stops.
Handover 4 weeks
The client’s analyst takes over, the monthly review joins the existing management meeting, the calculation method and decision rules are documented.
- Measurement per process, never per person
- Staff representatives informed before set-up
- A single method for counting gains, validated by finance
- Decision rules set before the first measurement
- Every decision to extend, adapt or stop documented with its evidence
- Monthly review within the existing meeting, not another committee
Measured results
The first result is not a figure: it is being able to answer the board with evidence. The indicators below compare the situation at the start and at the end of the five-month engagement.
Rounded figures. A real engagement: the client’s name and any detail that would identify it have been removed.
What we would do differently
Every study also says what could have been done better.
- 01
Negotiate access to usage statistics when buying the licences. Vendor exports were poor, and we lost three weeks reconstructing what a contract clause would have provided.
- 02
Involve staff representatives from week one, not just inform them before measuring. Two weeks of explaining what the system measured, and did not measure, would have been saved.
- 03
Make the decision rules non-negotiable before the first measurement, not just written down. The temptation to move the target once the figures are known is real: we experienced it.