Across banking, insurance, government, and retail, the pattern is consistent enough that I have stopped being surprised by it. An organisation has invested heavily in AI. The data science team is capable. The infrastructure is modern. The models are technically sound. And the programme is not delivering. When I trace it back, the answer is almost always the same. The business does not own it.
By ownership I do not mean who signs the budget. I mean who is accountable for the outcome AI is supposed to produce. In most enterprises that accountability sits with technology. The Chief Data Officer, the AI platform team, the cloud architecture function. These are the people who built the programme, who report on it, and who define what success looks like. The business leaders whose decisions AI is supposed to improve are typically consumers of the output, occasionally sponsors of the initiative, and rarely if ever held accountable for whether the AI investment changed anything that matters commercially.
That structure is not an accident. It is the predictable consequence of how the AI market has developed.
Technology vendors sold AI as infrastructure
The organisations that shaped enterprise AI adoption are, with few exceptions, technology companies. They built platforms, sold compute, and packaged AI capability as a layer of the technology stack. The pitch was capability. Here is what the system can do. Here is the accuracy it achieves. Here are the benchmarks. The business case was constructed in technical terms because the people constructing it were technical people selling to technical buyers.
The result is that most enterprise AI programmes began with a technology question. What can we build with what we have? That is the wrong question. The right question is which decisions are we making badly, and what would it be worth to make them better. Those two questions produce completely different programmes. The first produces solutions in search of problems. The second produces targeted interventions in specific decision processes with measurable business outcomes attached.
Vendors do not ask the second question because it requires domain knowledge they do not have and commercial accountability they are not prepared to accept. A technology vendor can demonstrate model accuracy. They cannot sit across from a Chief Risk Officer and explain which credit decisions are losing the most money and why. That conversation requires someone who understands the business, not someone who understands the model.
What technology ownership looks like in practice
The symptoms are recognisable. Success is defined as model performance rather than business outcome. No economic baseline is established before the programme starts, which means there is no credible way to answer the question of what it actually delivered. The business sponsor is identified after the technology has been selected rather than before. Proofs of concept prove the model works without proving the business case. And when the programme stalls, the conversation turns to data quality, infrastructure limitations, or model complexity rather than to the more uncomfortable question of whether the right decision was targeted in the first place.
The Data and AI team fiefdom compounds this. In most large enterprises, the data science function has built its identity, tooling, and institutional standing around a specific set of platforms and methodologies. That is not inherently problematic. What is problematic is when that functional ownership becomes a barrier to asking the business questions that would redirect the programme toward higher value. I have seen organisations where the data science team is genuinely excellent and genuinely isolated, producing technically impressive work on decisions that do not move the commercial needle, because nobody with business accountability was ever in the room to redirect them.
Business ownership is not a structural change. It is an accountability change.
The organisations getting this right have not reorganised their technology functions or moved data scientists into business units. They have changed where accountability sits. A named executive who is responsible for the quality of a specific set of business decisions owns the AI programme targeting those decisions. They define what success looks like in business terms before a model is built. They are present at reviews. They are the person who answers the board’s question about what the AI investment produced.
Technology remains in its proper role, building and operating the systems that deliver the capability the business has specified. That is an important and skilled role. It is not the ownership role. When technology is in the ownership role, AI programmes optimise for the wrong thing. They produce sophisticated systems that measure their own performance against internal benchmarks and cannot answer the one question that justifies the investment. Did the decision get better?
The uncomfortable truth for the technology vendors and consultants who have built the current AI market is that the capability they are selling is necessary but not sufficient. The domain knowledge required to ask the right business question, target the right decision, and hold the programme accountable to a business outcome is not in the platform. It is not in the model. It is in the business. And until the business owns AI rather than consuming it, the gap between AI investment and AI value will remain exactly where it has been for the past several years. Wide, expensive, and largely unexamined.
