Most AI investments don’t fail because the technology doesn’t work. The tools available today are genuinely capable, and the platforms have matured enough that the technical build isn’t where most initiatives fall apart. What falls apart is earlier, in the state of the business before any technology is introduced.

Henry Hays has a framework he applies before any AI initiative gets off the ground: three things have to be in reasonable shape before an investment has a real chance of delivering. Get all three right and almost any well-chosen AI application will produce results. Get one wrong and the best tool in the market won’t save you.

The First Thing: Your Technology Has to Be Ready

Technology readiness doesn’t mean having the most modern systems. It means having systems that are stable, connected, and in a state where adding an AI layer makes sense.

The most common technology readiness problem isn’t outdated software. It’s disconnection. AI gets dramatically more useful when it has access to the information and context that lives in the systems a business already runs. When those systems don’t communicate with each other, when data lives in silos that require manual effort to bridge, the AI ends up operating with incomplete information, and incomplete information produces incomplete results.

Before any AI initiative begins, it’s worth mapping out which systems the AI will need to touch and whether those systems are in a state where that’s practical. Are the integrations required straightforward, or will they take significant work to build? Are the underlying systems stable and well-maintained, or do they have known issues that an AI layer will inherit? Is there anyone internally who can support the technical side of the implementation, or will that capability need to come from outside?

None of these questions require a technical audit to answer. They require honest conversation with the people who manage the systems day to day, and a realistic look at what the implementation will actually involve.

The Second Thing: Your Data Has to Be Ready

Data readiness is where the most AI investments fail, and it’s the area where the most self-deception happens. In most growing businesses, data quality is assumed rather than verified. The CRM has records. The spreadsheets have numbers. The assumption is that everything is probably in good enough shape to be useful.

Sometimes that assumption is right. Often it isn’t. And the way to find out is to look before you build.

Hays is direct about the cost of skipping this step: if everything else is ready and the data isn’t, you are burning money. AI doesn’t generate insight from nothing. It operates on what you give it, and if what you give it is inconsistent, incomplete, or scattered across systems that don’t talk to each other, the outputs will reflect that. No tool compensates for genuinely problematic data.

The practical data readiness questions are: Where does the relevant data actually live? Is it in one place or spread across several? Who owns it and who maintains it? When was it last reviewed for accuracy and completeness? Is it structured in a way that makes it accessible to an AI system, or would significant cleanup and reorganization be required before it could be used?

If the answers reveal problems, that’s useful information. Data cleanup is almost always faster and less expensive to do before an AI implementation than during one, because during the implementation, every data problem becomes a project problem.

The Third Thing: Your People Have to Be Ready

People readiness gets the least attention in most pre-investment conversations, and it matters as much as the other two. Sometimes it matters more.

There are two dimensions to people readiness that both require attention. The first is understanding. Do the people who will be working alongside this AI know what it is, what it’s for, and what it isn’t? Do they understand that the goal is to make their work better, not to eliminate their roles?

Hays makes this point directly from his work in the field: employees need to know that AI is coming to augment what they do, not replace them. Without that clarity, even a well-designed AI implementation runs into resistance at the point where it touches actual people doing actual work. That resistance doesn’t always show up as open opposition. More often it shows up as inconsistent use, workarounds, and a quiet drift back to the old way of doing things.

The second dimension is capability. Using AI well is a skill. It involves knowing how to give it clear instructions, how to evaluate whether the output is actually good, how to push back when it misses, and how to iterate toward a result that’s useful rather than just plausible. Teams that are new to AI need time and guidance to develop that skill. Building that development time into the plan, rather than assuming the team will figure it out on their own, is the difference between an implementation that reaches its potential and one that gets used inconsistently and eventually deprioritized.

Why All Three Have to Be True at Once

The three readiness areas aren’t independent of each other. They compound when they’re all in shape, and they compound in the wrong direction when any one of them is significantly off.

Good technology and good data won’t produce results if the people using the system don’t know how to work with it or are resistant to adopting it. A capable and motivated team won’t get far if the data they’re feeding the AI is a mess. Clean data and a skilled team are limited by technology that isn’t stable enough to support what they’re trying to build.

This is why the assessment has to look at all three before the investment is made. A company that finds one area significantly out of alignment isn’t necessarily in the wrong place to invest in AI. It’s just in the right place to do some foundational work first, which is almost always a better outcome than discovering the problem six months into an implementation when fixing it is significantly more expensive.

What a Realistic Assessment Looks Like

Running through this assessment doesn’t require a lengthy consulting engagement. It requires honest conversation with the people who know the business best, and a willingness to hear answers that might not be what you were hoping for.

Technology: talk to whoever manages your systems day to day and ask them directly whether the systems the AI will need to access are in a state where adding something on top of them makes sense.

Data: pull a sample of the data that will be feeding the AI and look at it. Not a summary, not an aggregate. The actual records. See whether they’re consistent, complete, and structured in a way that an AI system could work with.

People: have a direct conversation with the team members who will be working with the AI about what they know about it, what concerns they have, and what they’d need to feel confident using it. The answers will tell you a lot about what the implementation plan needs to include.

The businesses that move through this assessment honestly and act on what they find are the ones that build AI capabilities that actually hold up. The ones that skip it find out what they missed during the implementation, which is the most expensive time to find out.

Listen to the full podcast episode here.

Related Resources

• The 90% Problem: What the MIT Study on AI Pilot Failure Actually Means for Your Business: Ninety percent of AI pilots fail, and most of those failures trace back to one of these three readiness areas. This post breaks down what the MIT research actually reveals and what to do about it.
• How to Evaluate Any AI Investment Before You Spend a Dollar: Readiness assessment is the first step. This post covers the full pre-investment evaluation framework, including the red flags that signal a company is about to make a bad investment.
• The Three Questions Every Business Owner Should Answer Before Investing in AI: Once readiness is confirmed, the next question is whether a specific use case is worth pursuing. This post walks through the ROI framework for making that call quickly and clearly.
• Who Should Own AI in Your Company When You Don’t Have an IT Department: Someone has to own the assessment process and the implementation that follows. This post covers how to find the right internal owner when you don’t have a dedicated technical team.