AI in SME operations: what to automate (and why to start small)

Operations are, in many manufacturing and distribution SMEs, the department where a delay is paid in cash: an empty shelf is a lost sale, an unplanned machine stoppage is a halted line, a late purchase order is a penalty. It's also the department that lives on monitoring loops — stock, logistics exceptions, maintenance signals — and it's precisely there that AI, in 2026, promises the most. But it's the department where AI projects fail more often than elsewhere, and for reasons that aren't technical. The serious question, as always, isn't "does the agent exist?" but how small it's worth starting so that the project actually reaches production.

We've gathered the 2026 field evidence and try to answer in plain language: where AI in operations delivers a real return, why the adoption numbers must be read with more caution than in sales or marketing, and what the scoping discipline is that separates the few projects that reach production from the large majority that stall.

The pattern of the cases that work: AI wins on monitoring loops

In operations too, the winning use cases share a trait: they are high-frequency, data-dense loops — watch a state, catch the exception, act before it becomes a problem — not judgement decisions that require the context of a supplier relationship or a contract. In practice:

• Stock replenishment and rebalancing — in place of manual review of the reorder point. Agents trigger transfers between warehouses before the stockout, not after. It's among the highest-return categories cited for 2026.
• Logistics exception handling — recalculating routes on disruptions in place of the manual rerouting decision. Cited alongside replenishment as the year's highest-ROI deployment.
• Procurement intake and sourcing automation — assisted review of requests and supplier comparison, in place of the manual RFQ cycle.
• Predictive maintenance — intervention on degradation signals in place of calendar-based or reactive maintenance. A mature use case, with an already-established tooling market.
• Demand planning — forecasts in place of manual spreadsheets; cited alongside replenishment among the highest-return categories.
• Pick-route optimisation — real-time resequencing on the actual order mix, in place of the static picking list.

The aggregate numbers, and why here they must be read with more caution

On the positive side the evidence is there: the 2026 analyses place the return of AI deployments in supply chain around +19% over traditional automation, with adopters reporting improvements in the order of +34% in production and supply-chain efficiency, and a very large share of manufacturing executives declaring their intent to invest in agentic AI this year. Intent, as in sales and marketing, runs well ahead of real deployment.

And this is where honesty is needed, because in operations that gap is wider than elsewhere. When you move from assistants to agents — systems that act on their own — the failure data worsen sharply: an analysis of hundreds of deployments found that about three projects in four did not reach their objective, and multiple sources indicate that the vast majority of AI agents never reach production. Of a thousand initiatives launched as pilots, only a single-digit-percentage minority reach production, and fewer still hit the expected return target.

The part that matters: the causes of these failures aren't technical. They're the ill-posed objective, treating the initiative as an IT project rather than a process change, unrealistic timelines and data quality. It's not that AI doesn't work — it's the way the project is set up. Anyone who sells you "the autonomous agent for the supply chain" as a first step is steering you towards the wrong statistic.

The discipline that separates those who reach production: starting small

It's the point where operations differ from sales and marketing, and the most useful lesson in all the evidence. Precisely because agentic projects fail often and for organisational reasons, the recommendation for an SME is clear-cut and counterintuitive relative to the demos: don't sell (or buy) a complete rollout of supply-chain agents as a first engagement. You choose one monitoring loop — stock, or maintenance signals — you take it to production, you measure it, and only then do you extend it.

This tight scoping isn't timidity: it's the direct countermeasure to the first two causes of failure cited above — the ill-posed objective and the project treated as IT rather than as process change. A single loop has a clear objective, an owner in the department and a number that says whether it's working. It's exactly what the stalling projects lack.

Two tool families, and in what order to use them

Without drawing up a shopping list, the operations market reads by tiers. There's the horizontal automation layer — the plumbing that connects the systems (event → action across applications): it's not operations-specific, but it's the realistic and cheapest entry point to test a workflow. Then there's the vertical layer built for SMEs, the AI that genuinely automates the purchasing or supply-chain process, where it makes sense to move up once the workflow is validated. And there's the enterprise source-to-pay tier, powerful and complete but with big-enterprise costs and timelines: a reference ceiling, not a starting point for an SME.

There's a particularity worth knowing: unlike sales and marketing, where the tools' price lists are fairly transparent, in operations and procurement the price of the verticals is almost always on request, opaque below the enterprise tier. The practical reading follows directly: test the workflow on the horizontal layer, where the cost is low and clear, before opening a quote conversation with a vertical. Negotiate the vertical tool only when the specific loop — stock, maintenance, procurement intake — is validated and its return case is concrete enough to justify that quote.

And compliance? Procurement raises the bar

An agent that touches supplier data and spend decisions raises due-diligence questions specific to procurement — where the supplier's data lives, who answers for a decision an AI takes on a contract — on top of the common base of data processing for AI: impact assessment (DPIA), minimisation, checking where the vendor processes and stores the data. The scope of use itself, for an SME, is rarely "high risk" under the EU AI Act, but the supplier risk assessment must be done with the same rigour as any software choice. This is exactly what our compliance overlay wires to every workflow we design.

Where to start, in practice

If operations are the department where you want to begin, the sensible path is short and ordered:

• Choose a single monitoring loop — stock or maintenance signals deliver the fastest return and the lowest risk. Not a rollout, a loop.
• Treat it as process change, not an IT project — with an owner in the operations department, not just in the technical office. It's the first cause of failure you avoid.
• Test on the horizontal layer before the vertical — connecting automation costs little and clarifies the workflow; the vertical on request comes later, on a validated case.
• Fix the data before the tool — clean item master data, stock levels and postings are half the work; data quality is among the first obstacles cited, not a detail.
• Define the success criterion before the tool — a number that says whether it's working (stockouts avoided, downtime reduced). Without one, the project dies from confused objectives, not from the limits of AI.

Even before choosing the department, 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 loop. And if the theme is the compliance of what touches suppliers and spend, 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 ROI, efficiency and adoption figures cited come from market analyses and from self-reported industry sources, not independently verified: they should be read as indications of direction and not as guarantees of results. Every tool choice and every automation that touches suppliers, spend or production must be assessed against the data, the controls and the context of the individual company.