All case studies

Case study · Prepare · AI strategy & readiness

From 43 ideas to 3 funded initiatives: the AI portfolio of a Swiss industrial group

How a Swiss industrial group turned 43 AI ideas and three stalled pilots into a portfolio of three funded initiatives — method, controls, measured results.

Anonymised case — representative example from real Numezis engagements

The board asked the same question at every meeting: what are we doing about AI? The organisation had answered with volume — internal workshops had collected 43 ideas, three pilots had been launched in different corners of the company, and several vendors kept pushing their own POCs. None of the pilots had reached production.

The problem was not a shortage of ideas but the absence of a basis for deciding. Taken in isolation, every initiative looked defensible; none had a named owner, a measured baseline or a condition for stopping. The annual budget window was approaching: without an investment thesis, the topic would either slip by a year or be funded blind — and the executive committee had been debating it for eight months.

The decision on the table

Where can AI create defensible value first — and what must be true before we invest?
01

Trade secrets and export customer data

Technical and commercial data of export customers is covered by confidentiality obligations. Every initiative had to state where this data flows before it could even be assessed.

02

Limited change capacity in production

Production teams were running at full load. Any initiative claiming shop-floor time had to justify it — the portfolio had to respect this capacity rather than assume it away.

03

An IT team of 4

Internal IT already ran the ERP and the infrastructure of both sites. Each initiative was therefore assessed on what it costs to operate, not only on what it costs to build.

04

Annual budget window

The investment decision had to be ready for the budget cycle. An elegant analysis delivered too late would have been worth as much as no analysis.

What we built

The deliverable was not a list of use cases but a decision system: instruments that let management tie ideas to value, compare them on explicit criteria, check what has to be true on the data side, sequence the investment in conditional horizons — and repeat that arbitration every month, without us.

System components

Value tree per workflowTies every idea to a real workflow and its economic lever — instead of ranking technologies
Potential × executability gridScores each initiative on defensible value and on the organisation’s actual capacity to deliver it
Data & access readiness mapDocuments, per initiative, the state of the required data, its sensitivity and who can access it
Risk classificationQualifies each initiative — confidentiality, vendor dependency, production impact — before any funding
12–18 month roadmap in 3 horizonsSequences Enable (foundations), Prove (measured evidence), Scale (deployment) — each horizon conditional on the previous one
Monthly portfolio governanceThe body that arbitrates: continue, redirect or stop, based on gates and baselines
01

Immersion & field mapping 3 weeks

Workshops with the teams of both sites and direct observation on the shop floor. The 43 ideas are attached to real workflows; those without an identifiable economic lever are discarded early.

02

Scoring & feasibility 3 weeks

Potential × executability scoring of the remaining initiatives, interviews with prospective owners, verification of vendor dependencies and production constraints.

03

Data readiness & reference architecture 4 weeks

Assessment of the data required by the short-listed initiatives — quality, access, sensitivity — and a reference architecture compatible with the ERP, M365 and an IT team of 4.

04

Executive thesis & sequencing 2 weeks

Consolidation into an investment thesis: 3 initiatives to fund, Enable / Prove / Scale horizons, decision gates and governance. Presented to the executive committee, then to the board.

Measured results

A framing engagement is judged twice: on the decision it enables, then on what actually exists a few months later. The figures below compare the state of the portfolio before and after the engagement; production status is measured at 9 months.

BeforeAfter
Candidate ideas433 funded initiatives
Initiatives with owner and baseline0%100%
Pilots without exit criterion30
AI initiatives in production02at 9 months, after the engagement
Committee investment decision8 months of prior debatetaken in 2 sessions

Rounded figures, recorded at the end of the engagement and at 9 months for production status. Anonymised case: a representative example drawn from real Numezis engagements, not a nameable reference.

This section is part of our editorial standard: no case study without its lessons.

  1. 01

    Go to the shop floor in week one. The most material opportunity in the portfolio came from a conversation with a team leader at his station, not from the ideation workshops — field observation should have preceded the workshops, not followed them.

  2. 02

    Simplify the scoring grid. Management decided on three criteria while the grid carried nine: the other six documented the analysis but weighed down every collective discussion.

  3. 03

    Involve the CFO from the framing stage. The cost model had to be rebuilt to fit the budget format — a matter of form that belongs at the start of an engagement, not at the end.