Mid-caps: digitising without losing the operators' know-how

A share of the French industrial and services fabric rests on mid-caps that are fifteen, thirty, fifty years old. Their operational know-how — how a complex case is handled, how an edge case is diagnosed, how a tight-deadline trade-off is made between two suppliers — is rarely written down. It lives in the heads of shop-floor managers, branch directors, senior operators. It is precisely this know-how that makes the difference with a less experienced competitor.

Three moments make that knowledge vulnerable: a leadership succession, a takeover by a fund, or a mass retirement of a generation that didn't train the next one. When one of these three events arrives, the digitisation question becomes urgent — not to modernise for the sake of it, but to capture in a system what would otherwise leave with the people.

Why traditional ERPs and vertical software fail

The classic answer to the digitisation need has been, for thirty years, to install an ERP or a vertical-industry software. These tools capture flows — orders, invoices, stock — but not the operational know-how. An ERP knows that a case has been handled; not why the shop-floor manager chose to route it to one workstation rather than another. That logic, both banal and crucial, stays in the operator's head.

The result is well known in most mid-caps: a heavily invested ERP, a system that works for invoicing and accounting, but that doesn't change the operational dependency on key people. When the shop-floor manager leaves, the software stays, the knowledge goes.

Modelling instead of recording

The approach that changes things is to stop looking for software that records flows, and to look for software that models the concepts of the business. Concretely: the actors' roles, a case's states, the transition rules between those states, the invariants to respect, the recognised exceptions, the common trade-offs. This modelling, done with the operators themselves, produces a system that becomes the digital twin of the organisation.

The exercise is more demanding than an ERP install. It requires interviews with shop-floor managers, on-site observation, and a modelling discipline few traditional vendors offer. But the result is qualitatively different: the software doesn't just trace what happens, it knows why it happens.

When AI starts to be useful

Once the business is modelled in the software, the question of AI agents takes on new meaning. An agent operating on a well-laid domain model can assist the teams in place without replacing them: it proposes a routing, a prioritisation, a typical answer, but the operator keeps the final call. The agent learns from real human decisions, and after a few months it helps the next generation by suggesting what the previous generation would have done.

It is the opposite of the replacement fantasy. The AI agent, in a mid-cap, isn't a substitute for the shop-floor manager — it is a way for their successor to benefit from experience they haven't yet acquired. The know-how no longer leaves with the person, it diffuses into the system.

The right sequence

A digitisation that succeeds in a mid-cap follows a precise sequence. First, identify the two or three operational processes most dependent on key people — not all of them, otherwise the project never lands. Then model those processes with the operators concerned, over eight-to-twelve weeks. Then ship the software that materialises the model, with a progressive rollout that doesn't disrupt production. Finally, plug AI agents into the functions where the most critical know-how was identified.

The full timeline for a complete activity is eighteen-to-thirty-six months. But the first results — and the securing of critical know-how — are visible by month six. That is what allows continued investment in family-owned mid-caps where financial prudence is a rule.