The Model Is Not the Problem. The Data Is.

The Model Is Not the Problem. The Data Is.

Lujane Brinkman · Technique Works · June 2026

AI high performers (McKinsey, Nov 2025): 6% · the remaining 94% are using AI without seeing the EBIT impact that defines a high performer.

We have spent the last couple of months watching the same pattern repeat across boardrooms in Amsterdam, Rotterdam, Dubai, Riyadh, and beyond.

A leadership team invests in an AI strategy tool. A consulting firm deploys a model. The outputs are rapid, the dashboards are clean, and the presentations are compelling. Three months later, the leadership team is making the same decisions they were making before, except now they are making them with more confidence and less scrutiny.

The model did not fail. The fed data did.

The question nobody is asking before they build

The AI consulting market in 2026 is moving at a speed that rewards confidence and punishes honesty. Every major firm now has an AI-powered strategy offering. The pitch is consistent: faster synthesis, pattern recognition across larger datasets, and scenario modeling at a scale no human team can match.

None of that is wrong. The processing capability is real. The pattern recognition is real. The speed is real.

In most industrial organisations, the data that feeds those models is not real.

Publicis Sapient's 2026 Guide to Next, drawing on a survey of more than 500 industry leaders, identified the central finding of the year plainly: organisations are failing at AI not because their algorithms are flawed, but because the data feeding them is inconsistent, fragmented, and ungoverned. "AI won't fail for lack of models. It will fail for lack of data discipline." (Publicis Sapient, 2025)

We have a different way of describing the same problem. The data is fragmented because the operational reality it is supposed to represent has never been accurately captured. You cannot govern data that was never trustworthy to begin with. And you cannot build a strategy on analysis that has been fed a version of the truth.

What the numbers actually show

Many boards are approving AI strategy investments on the assumption that the technology will surface insights their teams have missed. In fact, the technology surfaces patterns in the data it is given, and if that data reflects filtered, compressed, or politically softened reporting rather than operational reality, the model will produce filtered, compressed, or politically softened insights. Faster, and with more decimal places.

MIT Project NANDA, published in July 2025, found that 95% of organisations deploying generative AI saw zero measurable return. Not a low return. Zero. The failure, consistently, was not the model. It was data readiness, workflow integration, and the absence of a defined outcome before the build started. (MIT Project NANDA, 2025)

McKinsey's research, cited in Dataversity's 2026 data management analysis, found that nearly two-thirds of firms have not yet begun scaling their AI projects. (McKinsey, cited in Dataversity, 2026)

Only 6% of organisations in McKinsey's November 2025 research qualified as AI high performers, defined as organisations attributing an EBIT impact of 5% or more to AI use and reporting significant value from it. (McKinsey, 2025)

The 94% who are not in that category are not failing because they chose the wrong model. They are failing because they brought AI into organisations where the information infrastructure was not ready to support it, and where nobody had the honesty or the access to say so before the investment was approved.

A faster model fed bad data produces faster bad decisions

In our HSEQ work, we have always understood that a lagging indicator measured accurately is more useful than a leading indicator measured poorly. The same logic applies here.

An AI tool that synthesises six months of operational reporting at speed is only as useful as the accuracy of that reporting. In most industrial organisations, the reporting that reaches the strategic level has already been through four layers of summarisation, each of which has removed the friction, the ambiguity, and the inconvenient detail that makes operational data useful for actual decision-making.

The AI model does not recover what those four layers removed. It accelerates the remaining signal, which is a cleaner, more confident, faster version of the same incomplete picture the board was already receiving.

This is the specific risk we are seeing in 2026. Not that AI will replace strategic thinking. That AI will give boards more confidence in conclusions they should be questioning more rigorously.

What changes and what does not

The consulting industry is beginning to reckon with this. McKinsey disclosed that roughly a quarter of its global client fees now come from outcome-based contracts, a structural shift that signals clients are no longer willing to pay for process without demonstrated results. (McKinsey, reported via Business Insider, Nov 2025)

Separately, Deltek's 2025 Professional Services Roundtable found that 73% of clients now expect real-time visibility into project status and performance, and that transparency has moved from differentiator to baseline requirement. (Deltek, 2026)

Both shifts point in the same direction. The value of a consulting engagement is no longer the framework or the model. It is the outcome that the framework or model produces in the specific operational context of the client.

That requires something AI cannot do on its own: it requires someone who has been in the facilities, who understands how information actually travels and degrades between the shop floor and the boardroom, and who can establish whether the data being fed into any model reflects operational reality or a managed version of it.

The technology does not change the diagnostic requirement. It raises the cost of getting the diagnosis wrong.

Where we start every engagement

Before we recommend any tool, any model, or any analytical framework, we establish one thing: Is the information we are working with accurate?

That question is not technical. It is operational. It requires access to the people who actually run the systems, not the reports those systems produce. It requires the credibility to ask uncomfortable questions and the experience to recognise when an answer has been softened.

Across 47 facilities in petrochemicals, manufacturing, logistics, and life sciences, we have built the capability to establish that foundation and to identify, before any model is deployed, whether the data it will be fed is capable of producing insight or only capable of producing confidence.

There is a significant difference between the two. In 2026, the gap between them is costing industrial organisations more than any AI investment is returning.

The question to ask before the next AI strategy briefing

Is the data your model will be trained on an accurate representation of your operational reality, or is it a representation of what your operational reporting has been comfortable saying?

If you cannot answer that question with certainty, the model is not your next step.

The data audit is.