According to Gartner, a majority of AI projects get abandoned not because the AI doesn't work, but because the data and systems underneath can't support it.

If you run an established business, you've probably felt this already. You've seen the demos. You've read the case studies. Maybe you've even signed up for a few AI tools and started experimenting. The potential is obvious: smarter document processing, automated workflows, dashboards that actually surface useful insights.

But when it comes to plugging AI into the software your teams use every day -- the CRM, the operations platform, the internal tools you've been running for years -- things get complicated fast.

Here's the problem no one talks about enough: AI is only as good as the system it connects to.

The foundation problem

I've talked to enough businesses about their product ideas to see this pattern come up again and again. A company wants to add AI-powered document processing, or an intelligent dashboard, or an automated workflow. The concept is solid. The business case makes sense. Everyone's excited.

Then the technical reality hits.

The internal system was built years ago. There's no easy way to pull data out of it. The CRM and the operations platform don't talk to each other. Customer data lives in different places, in different formats, and nobody has a single source of truth.

No amount of AI on top will fix that. You'll end up with a smart layer on top of a broken foundation. The AI will hallucinate, give inconsistent answers, or simply not have access to the data it needs to be useful.

This isn't a hypothetical. MIT research found that 95% of enterprise AI pilots deliver no measurable business impact. The root cause isn't the AI models. It's that generic tools "don't learn from or adapt to workflows." The organizations where AI actually works are the ones that fixed the plumbing first.

What "getting ready for AI" actually looks like

If you're serious about integrating AI into your operations, there are three things that come up in almost every project.

You need an API layer. If your internal system can't expose data cleanly to other services, you can't connect AI to it. Period. Sometimes that means wrapping an existing system in an API layer without rewriting the core. Sometimes, if the system is old enough, it means rebuilding parts of it so there's something to connect to in the first place. Either way, someone needs to look at what you have and figure out the shortest path to making it accessible.

Think of it this way: AI needs a door into your software. If there's no door, it doesn't matter how smart the AI is. And if the wall is made of concrete, you can't just hang a door on it -- you might need to rebuild that section of the wall first.

Your data needs to be structured. AI agents, LLMs, automation pipelines -- they all rely on being able to read, retrieve, and act on data. If your data is siloed across departments, inconsistent in format, or trapped in systems that don't share it, the AI will reflect that mess right back at you.

This doesn't mean you need a perfect data warehouse before you start. But you do need to know where your data lives, how it flows, and what needs to be cleaned up before an AI tool can work with it reliably.

Someone needs to own the technical integration. This is where most AI projects stall. The AI consultants design the strategy. They map out what's possible, build the models, maybe even run a proof of concept. And then they hand it off and say "now connect this to your system."

Who does that? Who builds the software layer? Who modernizes the existing system so it can talk to the new one?

That gap between "AI strategy" and "working software" is more common than you'd think. And it's the number one reason promising AI projects die on the vine.

What this looks like in practice

Think about a mid-sized company that runs on a mix of tools: a CRM from five years ago, a custom operations platform someone built internally, and a bunch of spreadsheets filling the gaps. They want to add an AI assistant that can answer questions about orders, flag issues, and automate reports.

The AI part is actually the easy part. The hard part is making sure the AI can actually access the right data, in the right format, from all those different systems. That's the work nobody sees in the demos.

The businesses getting real value from AI right now aren't the ones with the fanciest AI tools. They're the ones that did the unglamorous work of getting their systems ready first. The AI is the last layer, not the first.

The question to start with

If you're planning any kind of AI integration into your internal operations, the first question isn't "which AI tool should we use?" It's "is our software ready to connect to it?"

Most businesses that start exploring AI quickly realize the real challenge isn't the AI part. It's getting their existing systems, data, and infrastructure ready for it. That's a software engineering problem, not an AI problem. And there aren't that many teams out there who understand both sides -- the AI layer and the legacy software underneath.

The order matters. Get the foundation right, and the AI part becomes straightforward. Skip it, and you'll end up in the majority of companies that abandon their AI projects before seeing any return.

If any of this sounds familiar, shoot me a message. I happen to know a team that does exactly this kind of work, and I can point you in the right direction. Let's get in touch!