Harvard Business Review ran a simple test on ordinary business data and found almost none of it holds up. Most organizations are building their AI strategy on top of that failure — because no one asked the data strategy question first.
The exercise behind that number is almost embarrassingly simple. Pull 100 data records your department created or touched. Have someone unfamiliar with the data check each record against a handful of basic criteria — is it correct, complete, and current? Most managers assume they'll fail a few. The HBR researchers found that across the organizations they studied, only 3% of data sets cleared even that low bar.
That statistic is nearly a decade old now. It hasn't gotten less relevant — it's gotten more dangerous. In 2017, bad data mostly meant bad reports and slow decisions. Today, that same bad data is what every AI initiative in the building is being trained on, fed into, and asked to reason over. An AI model doesn't know your customer address field has been wrong since 2019. It just learns from it, scales it, and repeats it with total confidence.
Here's the pattern I see most often when an organization brings in outside help to plan an AI rollout: they hire for AI strategy. Someone evaluates tools, drafts a roadmap, maybe runs a pilot. Data quality gets a bullet point in the risk section of the deck. It rarely gets its own workstream, its own budget line, or its own owner.
That's backwards. AI models cannot perform better than the data underneath them. An AI strategy that isn't co-designed with a data strategy from day one isn't really a strategy — it's a bet that the data will turn out to be good enough. Given the 3% figure, that's a bad bet.
The first thing I look for in a readiness assessment is whether an organization's AI strategy and data strategy were built together. In most, they weren't — and that gap is exactly where the project stalls six months later.
This isn't about having a data warehouse or a BI dashboard. Quality data — the kind an AI system can actually be trusted to learn from — has to be several things at once:
Most organizations pass one or two of these. Very few pass all five — which is exactly why the HBR number sits at 3% and not 30%.
Assumption 1: "We have enough data, so we're ready." Volume is not quality. An organization can have a decade of customer records and still fail every one of the five criteria above. More rows do not fix bad labels, stale entries, or embedded bias — they just mean the AI model has more bad examples to learn from.
Assumption 2: "We'll clean the data during implementation." Data cleanup treated as a mid-project task, rather than a pre-project gate, is how six-month AI rollouts become fourteen-month ones. The MIT xPRO AI Strategy and Leadership research is direct on this point: the organizations that succeed identify the data gap during the readiness assessment — before a vendor contract is signed, not after the first model output looks obviously wrong.
The MIT framework I use with clients treats AI strategy and data strategy as a single deliverable, not two separate ones. That means the same readiness assessment that evaluates which AI use cases are worth pursuing also audits whether the underlying data can actually support them — and if not, what has to happen first.
In practice, that looks like:
None of this is exotic. It's the same discipline behind a good ADKAR rollout: assess readiness honestly before you build momentum you can't sustain. The organizations quietly succeeding with AI right now aren't the ones with the newest models. They're the ones who did the unglamorous data work first — and treated it as part of the AI strategy, not a prerequisite chore handled by someone else.
The AI Readiness Assessment audits your AI strategy and your data strategy together — so you find the gap in week one, not month six.
Start with an assessment →Sources: Nagle, T., Redman, T., & Sammon, D. (2017, September 11). Only 3% of companies' data meets basic quality standards. Harvard Business Review. hbr.org | MIT xPRO. (2024). AI Strategy and Leadership Program, Modules 2, 4 & 9. Massachusetts Institute of Technology.