Why you don’t need the data you think

“You need massive, clean and structured data for AI to be useful.” It’s the most common argument marketing leaders hear — in boardrooms, at conferences, on every “where do we start with AI” discussion. The argument is two years out of date.

This issue is about what actually gates AI adoption in marketing. The data warehouse is not always the answer. The operating model is, and most teams haven’t started rebuilding it yet. Let’s look at what changed, and what didn’t.

Pre-2023, “AI in marketing” mostly meant predictive models — lead scoring, churn prediction, attribution. These required clean, dense, first-party data. Most teams didn’t have it, so waiting made sense back then. Generative AI moved the requirement.

Three things you can already do

• Generative Engine Optimization. Search is shifting from clicks to answers. Your job is no longer just ranking on Google — it’s getting cited by ChatGPT, Gemini, and AI Overviews. The mechanics are different: clear authorship, structured content, expert sources. The bar is editorial discipline, not data scale.
• Content personalisation, localisation, and adaptation. AI tools let teams deliver content by segment, campaign, and language at a fraction of the cost. The cost drops when the brand style guides and decision rights are sorted upstream.
• Synthetic market research. LLMs generate synthetic personas and run synthetic interviews against them. The output doesn’t replace primary research; it compresses early-stage discovery from weeks to days. Three days to test five messages, six personas, two value propositions — then you validate the strongest signals with real humans.

None of these three require a perfect data warehouse. All three require something else.

The 70% nobody budgets for

According to BCG’s Leader’s Guide to Transforming with AI, success breaks down on a 10-20-70 rule: 10% algorithms, 20% technology and data, 70% people and processes. The technology and data is the small part.

The 70% is where it actually breaks. Who’s accountable for the output. Where the human review sits. Which workflows are getting redesigned, not just augmented. How teams build the habit of shipping AI workflows, not piloting tools annually.

What to do this month

• Week 1 — Prioritise by value, not by hype. Look at the three use cases above. Don’t pick the trendiest; pick the one with the biggest gap between your current performance and what AI could deliver. Synthetic research if your insights cycle is too slow. Localisation if content cost is the blocker in new markets. GEO if your organic visibility is declining.
• Week 2 — Define ownership before you onboard. Who owns the output, who reviews, and what happens when the model gets it wrong. This is the work that gets skipped but decides whether the pilot scales or gets shelved in six months.
• Week 3 — Run one workflow end to end. The full workflow, including the human review, the brand check, the publish decision. Time it, measure it, and identify the value delivered.
• Week 4 — Scale the value, not the headcount. Once you’ve proven value on one workflow, the question shifts. Not “which tool next” but “which workflow next, and which team learns it.”

You don’t need bigger data to start. You need a sharper operating model and a team that ships. The teams that win the next two years will be the ones with the clearest operating models and the fastest learning loops.

← All insights