Two vendor pilots stopped and one use case put into production
A family-owned Portuguese manufacturer of about 150 people worked for six months with a part-time head of AI alongside its board.
At a glance
- Sector
- Industrial manufacturing, Portugal
- Size
- About 150 people, family-owned
- What we did
- A part-time head of AI, working with the board
- Length
- Six months, two to four days a month
- Result
- Two pilots stopped, €120k recovered, one use case in production
The problem
The company is more than 100 years old, supplies several listed companies and operates across Iberia. In the previous eighteen months the CEO and the board had received seven pitches from AI vendors, for predictive maintenance, visual quality control, customer service automation, ERP integration and document processing.
Two pilots had been signed and were spending budget. Neither had a clear measure of success. The board wanted an AI strategy, and nobody inside the company had the method or the standing to assess vendors, set a threshold for return or question a weak pilot without harming the relationship behind it.
What we did
A seat at the board
For six months, two to four days a month, we took part in the board's work. We turned each vendor pitch into a decision scored for return and risk, prepared each meeting and followed up on the actions agreed. We also set up the quarterly AI dashboard that the board now reviews on its own.
A grid for judging vendors
We built an evaluation grid that scores technical fit, exposure under the EU AI Act, total cost of ownership and the cost of leaving. With it, the board stopped the two pilots that had no clear return criteria. The €120k they were using went to one use case with a clear return.
Handing over
A senior IT manager took on a part-time role as AI operations lead. We handed over the grid, the dashboard and the routine for reviewing vendors, so the work continues without us. The handover point was set in the contract from the first day.
Vendor evaluation grid
| Scored on | Pilot 1 | Pilot 2 | Chosen use case |
|---|---|---|---|
| Technical fit | Scored | Scored | Scored |
| Exposure under the EU AI Act | Scored | Scored | Scored |
| Total cost of ownership | Scored | Scored | Scored |
| Cost of leaving | Scored | Scored | Scored |
| Clear return criteria | No | No | Yes |
| Outcome | Stopped | Stopped | In production |
The €120k the two pilots were using went to the chosen use case.
The results
| Figure | What it measures |
|---|---|
| 2 | vendor pilots without clear success criteria stopped |
| 1 | use case put into production |
| €340k | annualised value of that use case |
| 6 months | length of the engagement, after which the work moved in-house |
The board now reviews the AI dashboard each quarter, and the next vendor pitch will be judged by the company's own team with the same grid.
In the client's words
“We didn't need someone selling us AI. We needed someone helping us decide which AI to buy and which to walk away from. That's what we got.”
Delivered as Arqmetrica, the name we worked under before Ontia Labs.
Other work
Regulated financial services, EU
Client onboarding checks cut from five days to 1.2
Read the caseTransport and logistics, Iberia
AI training and shared rules across the group in eight months
Read the case
Start with a 30-minute call
Bring one task that takes your team too long, ideally one that comes round every week.
Book a 30-minute call