Putting the Cart Before the Horse

Business leaders are hearing a lot about AI—usually through polished decks, confident roadmaps, and big promises. But before you scale anything, there’s one question that matters most:

Is your data foundation strong enough for AI to be trusted in the real world?

AI isn’t a strategy. It’s a capability. The strategic decision is simpler—and harder—than most roadmaps admit:

Will this reliably improve a decision you care about (pricing, risk, service, fraud, forecasting) at scale?

What this means in the boardroom

If you can’t explain—in plain English—where the data comes from, who owns it, and why it’s fit for purpose, that’s not a small gap. It’s a reason to pause.

Before you turn a pilot into a programme, pressure-test the initiative with four questions:

• What decision will change?

• What would “better” look like—and how will we measure it?

• What data will the model use (and what data is it missing)?

• What could go wrong—and how would we know early?

Those questions are the difference between a promising demo and a durable business capability.

The parable, reimagined

Picture a merchant investing in a beautiful new cart—polished wood, perfectly balanced wheels. He hitches it up, ready to go… and nothing moves.

Not because the cart is flawed.

Because there’s no horse.

That’s many AI programmes in a nutshell. Organisations fund platforms, hire specialists, and announce transformation—then discover the model can’t be trained or trusted because the data is missing, messy, biased, or inaccessible, or all of the above.

The technology is rarely the limiting factor. The data are.

Why the shiny cart is tempting

AI is easy to sell because it’s visible. You can watch a chatbot answer questions, see a slick demo, and put a confident roadmap on a slide.

The work that makes AI succeed—data quality, integration, governance, access controls, lineage—rarely looks exciting. It’s slower, more technical, and often less visible.

But it’s also the work that determines whether AI creates real value… or just headlines.

Four ways data quietly sinks AI projects

1) Completeness: gaps in coverage

An AI model can only learn from what it can "see". If your data misses key channels, time periods, customer groups, or operating contexts, the model fills the gaps with guesswork.

The result is an AI that sounds confident about a world it only partially understands.

2) Trustworthiness: bias and provenance

Untrustworthy data is like a contaminated well: it may look fine at the surface, but the problem shows up in the output.

If historical data reflects bias, inconsistent processes, or weak provenance, the model will reproduce those issues at scale.

AI doesn’t fix bad data—it amplifies it.

3) Accuracy: errors and drift

Small data errors rarely look dramatic in a spreadsheet. In a model, they can become expensive.

A typo, a stale record, or a drifting sensor can steadily erode performance. And accuracy isn’t static: markets shift, customers behave differently, and models degrade unless they’re monitored and refreshed.

4) Availability: access, timeliness, and integration

Even good data fails if it can’t be accessed in time, in the right format, with the right permissions.

If relevant data is trapped in silos, locked behind approvals, or delayed by legacy systems, the AI programme stalls before it even gets started.

What “putting the horse first” looks like

The organisations that succeed with AI are rarely the ones with the flashiest models.

They’re the ones who were disciplined about their data foundations before the excitement began. They ask:

• What decision are we improving?

• Do we trust this data enough to act on it at scale?

• Who owns it, governs it, and maintains it?

• How will we keep it up to date as the business changes?

These aren’t glamorous questions. But they separate real transformation from expensive experimentation.

A practical checklist

Before you build, make sure you can answer:

• What decision will this solution change?

• How will success be measured?

• Which data sources will it use?

• Is the data complete, accurate, representative, and current?

• Are there bias, privacy, security, or compliance risks?

• Who owns the data—and who signs off on go-live?

• How will performance be monitored after launch?

Closing thought

This isn’t really a story about horses and carts. It’s about sequence and discipline.

The most powerful AI is useless if the underlying data can’t be trusted or accessed. The organisations getting value from AI aren’t necessarily moving fastest—they’re investing early in data foundations, agreeing ownership and controls, and only then scaling what works.

If you want AI that delivers, start with the question:

Can we trust what our data is telling us?