Resource · AI in Industry
Industrial AI Checklist: 7 Essential Keys Before You Invest a Single Euro
The industrial AI checklist below is designed to save you money before you spend it. 95% of enterprise generative AI projects fail to deliver measurable returns, according to the MIT NANDA 2025 report, and Gartner forecasts that 60% of AI projects without prepared data will be abandoned by 2026. If an industrial SME starts a project without meeting these 7 keys, the odds of joining those statistics are high.
Why so many companies fail with AI: the 2025-2026 evidence
Four independent studies agree on the diagnosis.MIT NANDA documents that 95% of enterprise GenAI projects generate no measurable P&L return, despite an estimated global investment of $30-40 billion.Gartner warns that 60% of AI projects without "AI-ready data" are abandoned.BCG, in its September 2025 survey of 1,250 executives, found that only 5% of companies get material value from AI at scale. And the Gartner I&O April 2026 survey of 782 leaders found that only 28% of AI infrastructure use cases meet ROI targets, and 20% fail outright.
The causes are organisational, not technological: 57% attribute the failure to expectations that are too high, too fast, 38% to a lack of skills and 38% to data quality or availability, according to the same Gartner report. The 2025 CIO analysis adds that 43% of companies cite data quality as the main obstacle.
The 7 keys of the industrial AI checklist
1 · A measurable use case with a numerical KPI
Saying "we want to digitalise" isn't enough. You need a specific KPI that AI is going to move: cutting gas consumption by X%, detecting defects before Y, or planning inventory with error below Z%. If you can't write the objective in a single sentence with a number in it, you're not ready to start.
2 · Last year's data is accessible
No historical data, no AI. The good news: almost every industrial plant has SCADA, ERP or MES generating data without realising it. Before the pilot, verify that this data is accessible, complete and consistent. The McKinsey State of AI 2025 study is unequivocal: companies that scale AI invest first in data preparation, not in models.
3 · An assigned operational owner for the problem
Every use case needs a plant or operations owner to champion it. Without that owner, the tool doesn't get used and the project stays stuck in eternal pilot mode. The "learning gap " identified in the MIT NANDA report is exactly that: tools that exist but don't fit into the real workflow because no one champions them.
4 · A stable process to optimise
AI learns patterns. If your process changes every week (new products, new materials, constant reorganisations), the model will never become reliable. Stabilise first. This aligns with the acatech Industrie 4.0 Maturity Index maturity model, which places the "Visibility" phase before "Predictive Capacity": seeing and understanding what's happening comes before predicting it.
5 · A budget sized for 12-18 months
A well-executed pilot costs, in the Spanish market, between €8,000 and €25,000 depending on complexity, and moving it into production and maintaining it adds between 15% and 25% a year. But the ROI, when done right, is real: McKinsey documents that in manufacturing, AI cuts maintenance costs by between 25% and 40%, and that 78% of plants with AI reduce waste. The challenge is reaching production: BCG reports that "future-built companies" take 62% of their AI initiatives to production, compared with 12% for laggards.
6 · Identified public funding
For an industrial SME in Cantabria, there are four active levers today:
- Industria 4.0 Cantabria 2026: up to 75% grant funding, maximum €200,000 per company, total fund €1.5M (Cantabria portal).
- INNOVA Cantabria 2026: total fund €5M, variable funding rates depending on project and company size (innovation portal).
- Cheques Innovación Cantabria 2026: up to 50% of eligible costs, call open since 16 April 2026 (details).
- Kit Digital · AI category: up to €6,000 in the specific AI category, national funding available for all SME segments (Acelera Pyme).
7 · A workforce willing to collaborate
The operator who has run the same machine for 20 years knows things the data doesn't capture. If the culture is "they're here to replace us with AI", the project fails. If it's "let's give the system what we know so it can apply it at scale", it succeeds. This cultural variable is the difference between pilots that last a month and systems that have been in production for three years.
How to interpret your industrial AI checklist score
- 0-3 keys: it's not the time yet. Spend 3-6 months stabilising processes, collecting data and training your team.
- 4-5 keys: you're close. Identify the 2 you're missing and work on them before investing.
- 6-7 keys: you're ready. It's time to talk to a technical partner who understands industry.
Maturity references to assess your starting position
Two frameworks can help you find your bearings. The Gartner AI Maturity Model sets out five stages (Foundational, Emerging, Operational, Scaled, Transformational) across seven pillars (strategy, value, organisation, people, governance, engineering, data). Useful for a business-level view.
The acatech Industrie 4.0 Maturity Index is the European reference for industrial settings: six stages (Computerisation, Connectivity, Visibility, Transparency, Predictive Capacity, Adaptability) across four areas (resources, information systems, culture, organisational structure). It has an openly downloadable PDF.
Public reference cases to inspire your project
Siemens documents, at its Amberg plant, 75% automated processes and a 99.99885% quality index. It's not an SME, but it serves as a reference for how far applied discipline can go. For a European benchmark in industrial SMEs, the paper "AI implementation in manufacturing SMEs: A resource orchestration approach" offers peer-reviewed academic evidence with Swedish cases from the packaging, plastics and metal sectors. The EIT Manufacturing report aggregates European cases by sector.
The next step for your industrial SME
If you want to validate your industrial AI checklist score with someone who has done this dozens of times, at Deduce Data Solutions we offer a 30-minute exploratory meeting where we go through the checklist point by point with you. You can also read about how we apply AI across different industrial sectors and see the GASAI case with Reinosa Forgings & Castings where this checklist was applied before getting started.
Sources consulted
Why this industrial AI checklist saves you months of work
The most expensive mistake in industrial AI projects isn’t made during implementation: it’s made before, when a company decides to start without having validated the minimum conditions that make the investment viable. This industrial AI checklist exists precisely to avoid that mistake: 7 questions that any operations leadership team can answer in 30 minutes, and which, if they fail, indicate that the project isn’t mature enough to start.
The typical use of the checklist is as an internal validation session before contacting any AI provider. If your team can’t answer 5 of the 7 questions with concrete clarity, don’t hire AI yet: spend 6 weeks first preparing the ground. Cutting corners here gets expensive when this phase is skipped.
How to apply the industrial AI checklist in your organisation step by step
The industrial AI checklist works as a structured conversation between three roles: operations leadership, IT/OT lead and executive sponsor. These three profiles should sit at the same table for a couple of hours and answer the 7 questions honestly, not with the politically correct answer.
Session 1 (60 min) · Questions 1-3. Focus on the problem, the sponsor and available data. If any of these three fails, there’s no point continuing. The most costly mistake in the end is assuming these elements are in place when they aren’t.
Session 2 (60 min) · Questions 4-7. Focus on technical integration, explainability, model maintenance and governance. These four determine whether the project, if it starts, reaches production instead of staying a nice-looking pilot.
Outcome. After the two sessions, the team comes away with a clear traffic-light rating per question (green / amber / red). If there are more than two reds, the next quarter is spent preparing the ground, not hiring AI. If everything is green or manageable amber, the project can start with solid guarantees.
Common mistakes the industrial AI checklist catches
Of the industrial AI projects that fail, almost all share one of these anti-patterns that this checklist catches early:
Starting with the technology, not the problem. “We’ve seen an impressive machine vision demo, can we apply it to our plant?” is a conversation that signals the order is backwards. Technology should be the answer to a specific plant-floor pain point, not a solution looking for a problem.
A financial sponsor without an operational sponsor. The CFO signs off on the investment but the plant manager doesn’t back it. Result: the system gets installed but ignored. The checklist forces you to be explicit about who the operational owner is, with real authority to change routines.
Promised data vs. real data. During the sales phase, everyone insists “we have enough data”. During the technical phase, it turns out sub-metering is missing, there are months-long gaps in historical records, or the labels are inconsistent. The checklist forces specificity from the start.
A black box with no explainability. A model that predicts “imminent failure” with no justification doesn’t get used on the plant floor. Explainability isn’t optional, it’s a requirement for adoption. The checklist surfaces this point before choosing technology.
No model maintenance plan. On day 1, the model works. By month 6, the model has silently drifted out of tune. Without drift governance and retraining, every AI project quietly expires.
What to do if your industrial AI checklist comes back red
A red result on the checklist isn’t bad news. It’s valuable information: the project isn’t mature enough to start TODAY, but the path to getting there in 6-12 months is clear. Three typical actions depending on which question came back red:
If the specific problem fails (question 1): spend a quarter auditing where the biggest real operational losses are, in €/year. Without a problem with a clear economic dimension, AI has no return-on-investment argument.
If the operational sponsor fails (question 2): the right investment isn’t in AI, it’s in internal alignment. If operations leadership doesn’t see the problem as a priority, hiring AI will generate friction rather than value.
If the data fails (question 3): spend 6-12 weeks on a sub-metering and instrumentation project. It’s investing before investing, but without that step no model is going to learn what it needs to learn.
If integration fails (question 4): talk to your IT/OT team to map out the bridges that need to be built between SCADA, MES, ERP and the future AI layer. Before selecting a provider.
How we use the checklist with industrial clients
At DDS, when an industrial company contacts us interested in exploring AI, the first thing we do in the initial diagnostic session is go through this same industrial AI checklist together. We don’t sell software or consulting until it’s clear the client is in a position to extract value from the investment.
That exploratory meeting lasts 30 minutes. There’s no closed proposal at the end, no sales pitch. There’s an honest traffic-light view of where your organisation stands and what to do over the next 3-6 months, even if that recommendation is “not yet”. We’ve chosen to say “wait” in some conversations where an aggressive salesperson would have signed a contract, because we know a badly started project destroys trust for the next 5 years.
If you lead operations, digital transformation or innovation at a Spanish industrial company and want to use this industrial AI checklist as the starting point for a serious conversation, a 30-minute session will make clear where you stand. No obligation, no pitch.
