AI in SME manufacturing: the incentive before the pitch (Transizione 5.0 and voucher)

For a manufacturing SME, the question about AI almost never starts with "which tool": it starts with "how much does it cost me and who pays for it". And this is exactly where manufacturing is different from every other sector — and the only one where, in 2026, the answer begins with the State. Even before choosing the use case, a company that produces can fund most of the adoption with public incentives, starting with software that makes a machine it already owns "intelligent". The right reading, for anyone who manufactures, is counterintuitive relative to every demo: the incentive comes before the pitch.

We've gathered the 2026 evidence and calls for tender, and we try to answer in plain language: how the State co-funds AI adoption on the shop floor, which use case is worth starting from so that the return is measurable and low-risk, and why the real obstacle — cost — is exactly the one the incentives address.

The incentive before the pitch: how the State co-funds adoption

Manufacturing is the only sector with a public-subsidy path tied directly to AI spend. Two instruments, in 2026, are worth the conversation before any software tool:

• Transizione 5.0 (Legge di Bilancio 2026) puts AI at the centre of the incentive, funding technology, consulting and training within defined thresholds. The point that changes everything for an SME: the iperammortamento (hyper-depreciation) at 180% also covers the AI software that makes an existing machine "intelligent" — not just new hardware. In practice, the company can often fund the adoption work itself with the incentive.
• The Voucher Doppia Transizione 2026 (Unioncamere) brings 150 million euro for the 2026–2029 three-year period and covers up to 70% of costs for AI, cybersecurity, cloud, big data and analytics, IoT and collaborative robotics. Pre-filing of applications has been open since 8 July 2026 on the ReStart/InfoCamere platform: a tender that has just started, not a window already closed.

The practical consequence is simple, and it holds as the opening of every project on the shop floor: you start from the incentive, not from the tool. Working out which spend falls within the voucher or the iperammortamento — and designing the intervention so that it does — is the first piece of value, even before the use case. It's also why "we implement it" counts double here: the intervention has to be designed to be both effective and fundable.

Where the return starts: predictive maintenance and quality control

Once the intervention is funded, the serious question remains: which use case to start from so that the return is concrete and the risk low? The field evidence points to two natural candidates for a producing SME, both data-dense loops where AI monitors and acts before the problem:

• Predictive maintenance — intervention on degradation signals in place of calendar-based or reactive maintenance. It's the most mature use case and the one with the most legible return: it cuts unplanned machine downtime by about 45% and maintenance costs by about 25%. A stoppage avoided is a number the shop floor understands at once.
• Quality control and inspection — automated analysis in place of manual visual sampling. It's cited among the three main use cases for smaller manufacturers, alongside predictive maintenance and supply-chain optimisation.
• Consumption and energy optimisation — automated monitoring in place of manual. It's indicated as an entry, low-friction use case for those who want to start small before touching the line.

The selection criterion isn't "which is the most advanced" but "which has the sharpest before/after": for a first adoption, predictive maintenance or quality control give the clearest metric with which to build the case — and with which, later, to justify the extension.

The numbers, read honestly

The positive side of the evidence is there, and it's concrete once the project reaches production: about 44% of AI projects that reach production achieve positive ROI within 12 months, and production deployments report on average a return of 5.8 times within 14 months. These are numbers that justify the attention — on one condition.

The condition is the part no demo shows: the majority of AI initiatives in manufacturing stall at proof-of-concept and never reach scale. The gap isn't between those who have AI and those who don't, but between those who take one case to production and measure it, and those who go on collecting pilots. And this is where the incentive also helps on the right front: 43.0% of Italian companies point to high cost as a barrier to adoption. The voucher and the iperammortamento don't address a hypothetical obstacle — they address that one, the most cited.

Starting small (and often without custom): the Brescia lesson

There's one last idea that lowers the entry barrier more than the technology does: to begin, you almost never need a bespoke system. A small Brescia-based producer of flexible packaging improved its production-capacity analysis and streamlined its flows using generic, off-the-shelf tools — ChatGPT, Gemini, Copilot — without building anything custom to start. It's the same lesson that holds in every department: you begin with the cheapest, fastest layer to test the flow, and you move up to the vertical only once the case is validated and its return justifies the larger spend.

For a factory this means: the predictive-maintenance or quality-control pilot can be set up in a lightweight, measurable and funded way — and only afterwards, with the number in hand, do you decide whether and where to move up. Human oversight stays on the process: AI flags and proposes, the decision that stops or restarts a line stays with the people who know it.

And compliance? Within limits, but with method

A maintenance or quality use case in an SME, in itself, rarely falls into "high risk" under the EU AI Act: for the most part these are systems that watch machine data, not decisions about people. But production data, supply data and — if the AI touches workplace safety — the implications for staff must be treated with the same rigour as any software choice: an impact assessment where needed, minimisation, clarity on where the vendor processes and stores the data, and a documented human oversight. It's exactly what our compliance overlay wires to every workflow we design — so that the funded case is also defensible.

Where to start, in practice

If you produce and AI is in your sights, the sensible path is short, ordered and — almost always — co-funded:

• Start from the incentive — check which spend falls within the Voucher Doppia Transizione (up to 70%) or the iperammortamento 180%, and design the intervention so that it does. It's the first piece of value, before the use case.
• Choose a single pilot with a sharp before/after — predictive maintenance or quality control give the clearest metric and the lowest risk. Not a rollout, one case.
• Test with ready-made tools before custom — the generic layer costs little and clarifies the flow; the bespoke system comes later, once the case is validated.
• Measure, then extend — a number the production department understands (downtime avoided, rejects reduced). Without it, the project dies at PoC like the majority, not because of the limits of AI.

Even before choosing the case, though, it's worth knowing where you are: our AI-readiness assessment helps you understand where to start with more return and less friction, and which controls to put around the first pilot. If the theme is the compliance of what touches production, suppliers or staff, our compliance overlay explains how we wire the controls to every design.

We've turned the first step into a self-serve, free assessment: a few questions and an indication of where to start, with what controls around it. Take the AI-readiness assessment — then, if it makes sense, let's talk.

This article is for orientation. The return, efficiency and adoption figures and the details of the cited tenders (Transizione 5.0 and iperammortamento in the Legge di Bilancio 2026; Voucher Doppia Transizione 2026 by Unioncamere, with pre-filing from 8 July 2026 on the InfoCamere platform) come from industry sources and from communications on the incentives, and should be read as indications of direction, not as guarantees of results nor as tax advice: amounts, thresholds and eligibility requirements must be verified against the official tender texts and with your own advisor before any spending decision. Every tool choice and every automation that touches production, suppliers or safety must be assessed against the data, the controls and the context of the individual company.