Industrial AI “with boots on the factory floor”: how SMEs can gain efficiency with minimal data

If you ask a plant manager about their priority, they will rarely say “a multi-billion-parameter LLM”. The urgent stuff is on the shop floor: avoiding stoppages, cutting scrap, and using fewer kilowatt-hours per part. That is why AI for industrial innovation sounds different: so-called edge AI, deployed directly on machinery or local devices, uses compact models that learn from small datasets and integrate without tearing down what already works. It is a pragmatic, affordable path for SMEs that want to solve age-old problems with new tools.

Unlike cloud AI, which centralizes analysis of large volumes of data, edge AI runs in situ, on the plant equipment itself. This brings clear advantages for an SME: low latency, greater data security (since data never leaves the factory), and tighter costs. The typical setup combines low-cost sensors, the PLC (Programmable Logic Controller) already running the line, and a gateway where the model executes, learning gradually from “few and noisy” data. The context favors this shift. According to the IndesIA alliance, only 2.9% of Spanish industrial SMEs use AI, despite 36.2% annual growth. In general, Spanish industrial firms prioritize AI across three very concrete fronts: energy management, predictive maintenance, and quality control.

From the lab to the factory

Unplanned downtime is a black hole for any production line, and predictive maintenance aims to anticipate it by analyzing vibrations, temperatures, or noises. Today, thanks to lightweight models, it is within reach for SMEs. One example is Fracttal, a Madrid-based startup offering AI-driven maintenance platforms. Their system moves companies from rigid calendars to alerts based on real data, reducing mean time between failures (MTBF) and optimizing spare-parts inventory. In small industries, where a single stalled motor can block all production, the difference translates directly into euros.

The second front is quality. For decades, part inspection has been manual or rule-based: measure dimensions, spot cracks. But the speed of lines and customer demands require something else: more precision and more speed, made possible by embedded computer vision with AI. Here SMEs face a challenge: how do you train an algorithm with few examples of defects?

The answer lies in techniques such as transfer learning (using pre-trained networks) and data augmentation (simulating variations or defects). Technology centers like TECNALIA in the Basque Country support companies with pilot projects aimed at reducing rejected parts and improving consistency in the final product. The workflow is straightforward: a camera detects micro-defects invisible to the naked eye, and the system learns to distinguish what truly matters from environmental “noise”.

Finally, if there is one area where edge AI shows immediate potential, it is energy. Compressed air, for example, can account for up to 20% of a plant’s electricity use. Detecting leaks and modulating compressor usage with intelligent control algorithms enables substantial savings.

The same goes for industrial furnaces. CELSA Group, one of Spain’s largest steelmakers, has digitized its steel mills with sensors and advanced analytics to stabilize consumption and reduce peaks. That same logic, applied at small scale with embedded models, can help an SME trim its monthly bill significantly. Ultimately, it is about turning data into kilowatt-hours saved per part.

Challenges, opportunities, and the SME’s role

The potential is clear, but so is the risk of getting stuck in pilot purgatory. In fact, many SMEs lack hybrid professional profiles, operators who understand data or data scientists who know the line. This is where programs like Cre100do, from Fundación Innovación Bankinter, play a key role: helping startups cross the dreaded “valley of death” in tech transfer and, above all, bringing SMEs closer to the so-called middle market, where they can consolidate and expand.

In parallel, operators like Telefónica are participating in the European IPCEI on microelectronics and secure communication, which includes workstreams dedicated to edge AI in industrial environments. These are signs that the infrastructure which a decade ago seemed exclusive to large multinationals is spreading as the backbone of SME 4.0.

Edge AI still requires upfront investment and runs into technical limits: the need for quality data, interoperability between systems, and a lack of standards. Risks also exist: excluding very small SMEs or creating dependency on external vendors. Even so, the success stories repeat a common recipe: 90-day pilots with a well-defined business KPI, measured and tracked. Sometimes that is all you need: cutting rejects by 10%, or energy use by 5%, can justify the investment on its own.

In the industrial world there is plenty of talk about “conscious factories” and “cognitive systems”. But the reality that matters to an SME is different: a pump that does not fail in the middle of the night shift, a camera that flags a defective part in time, a compressor that is not running idle. In this sense, the industrial AI that works does not live in the cloud, it lives on the factory floor. It learns from imperfect data, runs in gray boxes bolted to the wall, and reports its results in euros to management. For many SMEs, that is the shortest path to steadier, more efficient, more competitive production.