What we mean when we say data strategy

The phrase ‘data strategy’ has come to mean almost anything. A warehouse migration, a governance committee, a Power BI rollout, a cloud platform decision or a roadmap document that lives in a shared drive and gets dusted off once a year. When all those things can sit under the same heading of ‘strategy’, that framing has stopped being useful.

This matters more now than it used to. AI investment has moved from experiment to expectation, and the question of whether AI is delivering has become a real one in most boardrooms. The honest answer, in most cases, lies in the data underneath it, not in the model, not in the vendor, not in the team running the implementation. Which means the question of what data strategy means, properly, has become quite an important one to be able to answer.

Forrester's most recent State of AI research, reported into 2026, found that only 15% of AI decision-makers had seen an EBITDA lift for their organisation in the past 12 months, and fewer than one in three could tie the value of AI to changes in their P&L. An uncomfortable statistic, given how much of the conversation in boardrooms over the last two years has been about exactly that.

Five things, working together

When data strategy works, it tends to mean five things rather than one: quality, integration, foundations, literacy and business alignment. And the thing that distinguishes organisations getting real value from AI from those still stuck in pilots is rarely depth in any one of these but instead it’s the way the five interact.

Quality is whether the data the business runs on is accurate, current, and trustworthy enough to act on. Strong quality on its own gets you accurate numbers nobody's connecting to the right questions.

Integration is whether data can move between the systems and people that need it, when they need it. IDC found that 81% of IT leaders cite data silos as a major barrier to digital transformation, and the typical enterprise now runs close to 900 applications with under a third of them properly connected. Strong integration without quality underneath it just scales the wrong answer faster.

Foundations is the piece most often filed under "governance," which is where the conversation tends to lose people. We don't mean a compliance framework, and we don't mean a committee, we mean the agreement underneath the data; who owns what, what it means, how it's used, and whether the outputs can be trusted enough to put in front of a customer. This is the piece that determines whether AI initiatives stay inward-facing experiments or become something the business is willing to expose externally. It's also the piece that, when it isn't clear, tends to surface uncomfortably late in the project.

Literacy is whether the people across the business can interpret, challenge, and use data well enough to make decisions with it. Strong literacy without the other four gets you a confident workforce making sharp decisions on shaky inputs.

Business alignment is whether all of the above is tied to the questions the business is actually trying to answer. This is the one most often missing in practice, and the one that most distinguishes a data strategy from a data programme. You can be mature on the first four and still be pointing in a direction the business isn't going.

The honest part

Most organisations are reasonably strong on one or two of these. Very few are strong on all five. And it's only when all five are working together that AI investments can consistently deliver at scale.

That's not a criticism; it's a reflection of how data work has tended to evolve inside most organisations. Quality became someone's project, integration became someone else's, governance got owned in a different part of the business again, and the threads were rarely pulled together because there was rarely a single moment that demanded it. AI is that moment, AI compounds whatever balance you already have. Strong on quality but weak on alignment, and you'll automate the wrong decisions confidently. Strong on integration but weak on literacy, and you'll have beautifully connected data nobody can use. Strong on foundations but weak on business alignment, and you'll have a perfectly governed data estate that doesn't earn its keep.

This is part of why so many pilots that demo well never make it into production. McKinsey's research suggests that of organisations running AI initiatives, around a quarter are actually succeeding at scale. The rest tend to start the same way; somebody decides to try out a new tool on a small subset of data and on a small piece of infrastructure, and it works. Then comes the point of trying to scale it, and that's where the loose ends surface. Quality issues, ownership questions, trust gaps in the output, none of which were the model's fault.

There's a useful analogy here; putting a strong AI model on weak data is a bit like putting a Ferrari engine into a car with flat tyres on a road it was never meant to run on. There's nothing wrong with the engine. It's just not in the right place.

A more useful question

The question worth asking probably isn't “which AI tool should we buy next”? It's something more like “what would we need to believe about our data for AI to do what we're expecting of it”?

That question changes the conversation. It moves things from procurement to readiness, from vendor selection to honest self-assessment, from what AI could do for the business to what would have to be true about the data for it to. And once that question is on the table, the five tend to follow naturally.

One last thing worth saying about the foundations underneath all of this; they tend to be a once-and-done piece of work, in the sense that the technology layer above them keeps changing; new models, new tools, new architectures arriving by the quarter, and getting the foundations right means each new thing on top costs less to experiment with. That's the bit that quietly de-risks the next investment, and the one after that. Done properly, it's not a project that ever really finishes, but it's one you only must start once.

We've written this up at more length; what data strategy looks like across the five, where mid-market organisations tend to sit on each, and how to tell the difference between a real foundation and a documented one.

Access the whitepaper >