The 3-Question AI-Ready Data Test

Most AI readiness assessments are vendor sales tools. Here is the test we run on every engagement. You can run it yourself in an afternoon.

AI project failure rate vs. traditional IT

• Traditional IT projects. 40%. fail to deliver expected value
• AI projects. 80%. fail to deliver expected value - twice the rate

AI projects fail at twice the rate of traditional IT because they depend on data quality in ways that traditional IT does not. The three questions in this test identify the most common failure points.

Source: RAND Corporation, 2024

What are the 3 questions in the AI readiness test?

Can you produce one source of truth for any customer in under 30 seconds? Do your top 12 KPIs have one definition the whole company uses? Is every dashboard's source traceable from output to source table? If any answer is no, AI in production will amplify the underlying problem.

I’ve taken a bunch of those AI readiness assessments. Most of the time, they’re just long, tedious surveys that give you some bland score. Then, they push you to buy their product. Truth is, they don’t really show what’s stopping you from moving forward.

Here’s a quick test we like to run at the start of every project. It’s just three simple questions, and you can get through them in an afternoon. The answers reveal way more about your AI readiness than any vendor’s checklist ever could.

• 70% AI projects that fail due to data quality. 70% of AI project failures are attributable to data quality issues - not model quality, not infrastructure, not talent. If you cannot answer the three questions in this test, you are in the 70%.

Source: Traxtech, 2025

Question 1: Can You Agree on Your Key Metric in One Meeting?

This isn’t just about revenue. Pick any key metric you care about. The real question: can everyone in the room agree on what it means, how it’s calculated, and where it stands, right then and there? No emails later, no guessing games.

If you answered yes, your metric governance is working well. If not, your data’s a mess, and your AI will learn from inconsistent data and produce unreliable results.

To be direct, AI models only deliver when the data they get actually makes sense. If your “revenue” number means three different things depending on where you check, your model’s output is going to be garbage. It’s not the AI messing up; it’s that the metric itself is inconsistent. Establish clear, consistent definitions, that’s where the magic starts.

Question 2: Can You Trace a Customer Record Back to Its Source?

Grab a customer record from your CRM or data warehouse. Now, ask yourself these:

• Who actually built this system?
• When’s the last time we even touched it?
• What changes have we made along the way?
• And how did we deal with duplicates?

If you can run through all that in under 15 minutes, your data lineage is solid. If you find yourself hesitating or guessing, it’s a red flag.

AI decisions only work when the customer data behind them is solid and traceable. If you can’t track where the data came from, good luck trying to audit those decisions or keep up with regulations. Compliance is not the goal here, trust is. Without that, the whole system falls apart.

Question 3: What Happens When a Source System Changes?

Imagine you’ve got a source system feeding data into your analytics, ERP, or CRM. Then, boom, it changes its schema. Maybe it adds new columns, renames fields, or swaps data types. How fast do your reports break? And more importantly, how quickly can you fix them and get everything back on track?

If you’ve got automated alerts firing and your change management updates downstream systems within a day, you’re doing pipeline governance right. You’re spotting problems quickly and keeping everything aligned. Simple.

If you find out a report is broken only when someone complains, you’re already behind.

AI models are way more fragile when stuff changes. If a report breaks, you see it immediately. But if a key feature in a model goes off track, it can quietly churn out bad results for weeks before anyone notices.

The 3-question AI readiness test

If you cannot answer all three confidently, your data is not AI-ready.

• Can you answer 'What is our revenue?' in one meeting?. If finance, sales, and the data warehouse give three different numbers, your metric definitions are not AI-ready.
• Can you tell me where a specific customer record came from?. If you cannot trace a record to its source system, version, and transformation history, your lineage is not AI-ready.
• What happens to your reports when the source system changes?. If reports break silently or produce wrong numbers after a schema change, your data contracts are not AI-ready.

Organizations that answer all three confidently ship AI into production at 6× the rate of those that cannot.

Source: Thinklytics Data Foundation Practice, 2026

What Your Answers Mean

Alright, here’s the deal:

• If you said yes to all three, congrats. Your data foundation is solid enough to start building AI. You’re not done yet, but you’ve got a good base to build on.
• If you said no to one, don’t stress. You’ve found a gap that’s totally fixable. Expect to spend about 8 to 16 weeks of focused work closing it.
• But if you’re saying no to two or three, your data layer isn’t ready for AI yet. This one’s bigger. Plan on 12 to 18 months of work, and you’ll need buy-in from the whole org, not just the data team.

It’s really about understanding where you are so you can move forward with a clear direction.

This isn’t a deep dive into AI readiness. Think of it more like a quick check-up. It helps you figure out if your biggest headache is your foundation, governance, or data pipelines, and points you to what to tackle first. Just start there.

Frequently asked questions

What are the 3 questions in the AI readiness test?

Can you produce one source of truth for any customer in under 30 seconds? Do your top 12 KPIs have one definition the whole company uses? Is every dashboard's source traceable from output to source table? If any answer is no, AI in production will amplify the underlying problem.

Why only 3 questions?

Most readiness assessments produce a 40-page checklist that's true but unactionable. These three are the prerequisites that 95 percent of stalled AI projects fail on. Pass these three and the rest can be fixed iteratively while AI ships.

What is a passing answer to each question?

One source of truth: entities resolved across all customer-touching systems. KPI definition: documented owner, definition, refresh cadence, and one tool everyone reads from. Lineage: every dashboard cell traces to a source table in two clicks. If you can demonstrate each, you pass.

How does this test compare to a full 30-day AI readiness assessment?

The test is a self-administered first cut to decide whether to engage. The 30-day assessment goes deep on the gaps the test surfaces, scopes the remediation, and gives a go/no-go on each AI use case you're considering. Most companies that fail the test should run the 30-day.

What is the most common failing answer?

Question two. Companies have hundreds of numbers and a handful of metrics. The handful are the only ones executives look at, but nobody has formally certified them. Definitions drift across teams and AI agents pick the wrong one.

How fast can a company go from failing to passing the test?

12 to 24 weeks of focused remediation. The first 6 to 8 weeks close question two (metric certification). The next 8 to 12 close question one (entity resolution). Question three follows automatically once one and two are settled.

What if we fail one of the three?

That's normal. Most companies fail one or two on the first test. The point is to know which one and start there. Question two (KPI definitions) is the most common single failure and the fastest to fix.

Should we run this test before talking to AI vendors?

Yes. Vendor sales conversations assume your data is ready; the conversation gets cleaner once you know which assumptions are true and which aren't. Reading our AI workflow automation vendor evaluation 14 questions after passing the test sharpens the vendor selection.