Data Strategy for AI: What Most Marketers Get Wrong Before They Start

Data strategy for AI is not a technical problem. It is a business clarity problem. Most organisations that struggle to extract value from AI tools are not failing because of infrastructure or compute costs. They are failing because they never defined what they wanted the AI to do with their data in the first place.

The organisations getting this right share one habit: they treat data strategy as a precondition for AI investment, not a consequence of it. That distinction changes everything about how you prioritise, what you build, and what you stop wasting money on.

Why Most AI Deployments Underperform

I have been in rooms where the conversation goes something like this: the business has bought an AI tool, the tech team has connected it to a data warehouse, and now everyone is waiting for the insight to arrive. It rarely does. Not because the tool is bad, but because nobody has agreed on what question they are trying to answer.

This is not a new problem dressed in new language. When I was running agencies, the same thing happened with analytics platforms. Clients would invest in a dashboarding tool, pull in every available data source, and produce a report that told them everything except what to do next. The issue was never the platform. The issue was that nobody had defined what a good outcome looked like before they started measuring.

AI amplifies this problem. It processes more data, faster, and with more apparent authority. Which means a poorly framed question produces a confidently presented wrong answer at scale. That is a more dangerous failure mode than a bad spreadsheet.

If you are thinking about how data strategy connects to broader go-to-market planning, the Go-To-Market and Growth Strategy hub on The Marketing Juice covers the commercial infrastructure that makes these decisions stick.

What a Data Strategy for AI Actually Means

A data strategy for AI is a documented set of decisions about what data you collect, how it is structured, what it is used for, who owns it, and how it connects to business outcomes. That last part is the one most organisations skip.

The strategy should answer five questions before any AI tool is selected or deployed:

• What commercial decisions do we need AI to improve?
• What data do those decisions require?
• Is that data currently available, clean, and accessible?
• Who is accountable for its quality and governance?
• How will we know if the AI is making those decisions better?

Most organisations can answer the first question vaguely and the rest not at all. That gap is where AI budgets go to die.

The First-Party Data Imperative

There is a version of this conversation that has been happening in marketing for about a decade. Third-party cookies were going away. Signal loss was coming. First-party data was the answer. And yet, when you look at how most marketing organisations actually operate, first-party data collection is still an afterthought bolted onto campaigns rather than a designed system.

For AI, this matters more than it ever has. The models that will give you a genuine competitive edge are the ones trained or fine-tuned on your data: your customers, your products, your conversion patterns, your category dynamics. Generic AI built on generic data produces generic outputs. That is not a strategic asset.

Building a first-party data foundation means designing every customer touchpoint with data collection in mind. Not in a surveillance sense, but in a value exchange sense. What are you offering customers in return for meaningful data? Personalisation, better recommendations, faster service, relevant content? The exchange has to be genuine or the data you collect will be shallow and useless.

Earlier in my career I spent a lot of time in the performance marketing trenches, optimising lower-funnel signals that felt precise and actionable. Over time I came to believe that much of what performance was being credited for was going to happen anyway. The customer was already in market, already close to a decision. We were capturing intent, not creating it. The same logic applies to first-party data: if you only collect data at the point of purchase, you are capturing the tail end of a experience you had no part in shaping. The richer, more useful data lives earlier in the relationship.

The Data Quality Problem Nobody Wants to Talk About

Here is something I have observed consistently across large organisations: the data exists, but it is a mess. Duplicate records, inconsistent naming conventions, fields populated incorrectly, CRM data that has not been touched in three years, attribution models that were set up by someone who left the business. The data warehouse is full. The data is not usable.

AI does not fix dirty data. It operationalises it. Feed a model inconsistent or mislabelled data and it will learn the wrong patterns with impressive efficiency. Garbage in, confident garbage out.

A data audit is not glamorous work. Nobody wants to spend three months cleaning CRM records when there are AI tools to deploy and board presentations to prepare. But it is the work that determines whether the AI investment pays off. I have seen organisations skip this step and then spend twice as long trying to understand why their AI-powered personalisation is producing recommendations that make no commercial sense.

The audit should cover four areas: completeness (what data is missing), consistency (is the same thing described the same way everywhere), accuracy (does the data reflect reality), and timeliness (how current is it). Each of those dimensions affects how an AI model will perform.

Governance Is Not a Bureaucratic Exercise

Data governance gets a bad reputation because it is often implemented as a compliance exercise rather than a commercial one. The result is a governance framework that slows things down without making the data more useful. That is the wrong version.

Good data governance for AI has three practical components. First, ownership: every data set needs a named owner who is accountable for its quality. Not a team. A person. Second, standards: consistent definitions, naming conventions, and update protocols that apply across the organisation. Third, access: clear rules about who can use which data for which purposes, with AI use cases explicitly included.

The governance conversation also needs to include your legal and privacy teams earlier than feels comfortable. AI use of customer data raises questions that many marketing teams are not equipped to answer alone. What data can be used to train a model? What consent did customers give? How do you handle data that is accurate but sensitive? These are not hypothetical questions. They are live risks that have caught organisations off guard.

Organisations that have scaled well, including those that have managed market penetration at speed, tend to have governance structures that enable speed rather than prevent it. The goal is not to create a committee for every decision. It is to make the rules clear enough that decisions can be made quickly and consistently.

Connecting Data Strategy to Commercial Outcomes

The test of any data strategy is whether it changes decisions that affect revenue. Not whether it produces interesting reports. Not whether it impresses the board with visualisations. Whether it changes decisions.

I spent years judging the Effie Awards, which are specifically about marketing effectiveness. One pattern that separated the winners from the shortlisted-but-didn’t-win entries was the directness of the link between what they measured and what they changed. The best entries could show a clear chain from data to decision to outcome. The weaker ones had sophisticated measurement but no clear line to commercial impact.

For AI, this means being specific about the decisions you want AI to inform. Not “better customer understanding” but “reducing churn in the first 90 days by identifying the behavioural signals that precede cancellation.” Not “smarter media buying” but “improving cost per acquisition in the 25-44 segment by adjusting bid strategies based on session depth signals.” The more specific the commercial question, the more useful the data strategy that supports it.

This specificity also makes it easier to evaluate AI tools. Instead of asking “is this AI tool good?”, you can ask “does this AI tool answer the specific question we have defined, using the data we have, in a way that changes a decision that affects revenue?” That is a much more useful procurement question.

The Segmentation Trap

One of the most common misuses of AI in marketing is using it to produce ever-finer customer segmentation without a clear plan for what to do with those segments. You end up with 47 microsegments, each with a detailed profile, and no capacity to activate against more than three of them. The AI has done its job. The marketing strategy has not.

Segmentation is only useful if it changes what you say, to whom, through which channel, and when. If your segmentation output does not connect directly to activation decisions, it is an interesting research exercise, not a commercial tool.

The better framing is to start with the activation constraint. What segments can you actually treat differently given your current creative, media, and CRM capabilities? Build your AI-driven segmentation around that constraint, not around the theoretical maximum the model can produce. A strategy that produces three actionable segments is worth more than one that produces forty unusable ones.

This connects to a broader point about AI in go-to-market strategy. The tools can outpace the organisation’s ability to act on what they produce. That is not a technology failure. It is a planning failure. BCG’s work on go-to-market strategy has consistently shown that execution capability is as important as strategic intent. AI does not change that. It accelerates the gap between what you know and what you can act on if you are not deliberate about closing it.

Building the Data Infrastructure Without Over-Engineering It

There is a version of data infrastructure for AI that costs millions, takes two years to build, and is out of date by the time it launches. Most marketing organisations do not need that version.

What most organisations need is a customer data platform that consolidates first-party data from key touchpoints, a clean and consistently maintained CRM, event tracking that captures meaningful behavioural signals (not just page views), and a measurement framework that connects media and content inputs to commercial outputs. That is not a small undertaking, but it is a manageable one.

The over-engineering trap is real. I have watched organisations spend 18 months building a data lake that nobody uses because the analysts who were supposed to query it left the business and the replacement team preferred a different tool. Infrastructure has to be built for the organisation that exists, not the one you aspire to become.

Start with the use case. What is the first AI application you want to deploy? What data does it need? Is that data available and clean? If yes, build the minimum infrastructure to connect those two things. If no, fix the data first. Repeat. This iterative approach is less impressive in a board presentation than a grand unified data architecture, but it produces working AI applications faster and with less waste.

Tools like Hotjar can contribute useful behavioural signal data at the session level, but they are inputs to a strategy, not a strategy in themselves. The same is true of most point solutions. They are useful when they connect to something larger. They are noise when they do not.

AI and the Measurement Problem

Measuring the impact of AI on marketing performance is genuinely hard. Not because the tools are opaque (though some are), but because AI is rarely the only variable that changes. You deploy an AI-powered personalisation engine at the same time as you refresh your creative, shift your media mix, and launch a new product. Which of those things drove the improvement in conversion rate? You do not know, and the AI cannot tell you.

The honest answer is that you need to isolate AI interventions where possible, run controlled tests, and accept that some of the measurement will be approximate. Analytics tools are a perspective on reality, not reality itself. That is true of all measurement, and it is especially true when you are measuring something as contextually dependent as AI-driven personalisation.

What you can measure more reliably is process efficiency. How long does it take to produce a campaign brief? How many creative variants can you test per quarter? How quickly can you identify underperforming segments and reallocate budget? These are operational metrics, but they are real, and they compound over time into commercial advantage.

Organisations that have thought seriously about growth at scale understand that measurement discipline is not about perfect attribution. It is about making better decisions more consistently than your competitors. AI can help with that. But only if the data strategy underneath it is sound.

The Organisational Side of Data Strategy

Data strategy for AI is not just a technical or commercial question. It is an organisational one. Who owns the data? Who owns the AI outputs? Who is accountable when the AI makes a bad recommendation? These questions do not have obvious answers in most marketing organisations, and the absence of clear answers creates the kind of diffused accountability that produces diffused results.

When I grew an agency from around 20 people to over 100, one of the clearest lessons was that accountability does not scale automatically. You can add people and tools indefinitely, but if ownership is unclear, quality degrades. Data strategy has the same property. As the data environment grows more complex, the accountability structure has to grow with it, or the complexity becomes unmanageable.

This means marketing leaders need to be active participants in data governance conversations, not passive recipients of whatever the tech team decides. The commercial context that makes data useful lives in marketing. The technical infrastructure that makes it accessible lives in IT or data engineering. Neither side can do this alone, and the organisations that figure out how to make those two functions work together are the ones that get real value from AI.

Agile frameworks, when applied well, can help here. Forrester’s work on agile scaling points to the importance of cross-functional ownership as organisations grow. Data strategy for AI is exactly the kind of problem that benefits from that kind of structure: small, accountable teams with clear mandates, iterating against defined commercial outcomes.

If you are working through how data strategy connects to your broader commercial planning, the Go-To-Market and Growth Strategy hub covers the strategic frameworks that sit around these decisions, including how to align marketing investment with business outcomes.

Where to Start

The most useful thing most marketing organisations can do right now is not buy a new AI tool. It is conduct an honest audit of their existing data: what they have, what condition it is in, who owns it, and what commercial questions it could answer if it were clean and connected.

That audit will be uncomfortable. It will reveal gaps, inconsistencies, and ownership disputes that have been papered over for years. It will also reveal the actual starting point, which is the only place a real strategy can begin.

From there, the sequence is straightforward even if the execution is not. Define the commercial decisions you want AI to improve. Identify the data those decisions require. Close the gap between what you have and what you need. Build the minimum infrastructure to connect data to AI application. Measure the commercial output, not just the AI performance. Iterate.

That is not a revolutionary framework. But most of the organisations struggling with AI are not struggling because the framework is wrong. They are struggling because they skipped the first step and went straight to the tool. The data strategy is the work. The AI is what you do when the work is done.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions