Hunting for Hidden Tech Debt

Every established company runs on old shortcuts and aging systems: a CRM several versions behind, a spreadsheet that moves numbers from the accounting system into it, or a connection between two systems that only one person understands. These are forms of tech debt, the cost of choices that were easier at the time and harder to live with later, along with its close cousin, the process nobody wrote down. Both can stay hidden for years, until an AI project brings them into the spotlight.

This differs from the problem I described in Bad Data Kills Good AI. That article was about the records. This one is about the systems themselves and the plumbing between them.

In most companies I talk with, much of how work gets done lives in people’s heads. Say an AI tool prepares quotes from the prices in your CRM, but the real prices are in a spreadsheet someone updates by hand. The quotes will sound convincing and be wrong. A second risk is harder to spot: automating a step can remove a workaround that was doing a job, such as someone catching errors before invoices went out. A third is more basic: an outdated system may not connect to modern AI tools at all.

In one large program described by Boston Consulting Group and CAST, AI-generated changes to software nobody had mapped were about 30 percent accurate over the first year, and often created more work than they saved. Once the team mapped how each application actually worked, accuracy rose to 85 percent.¹ That was a large company’s software, but the lesson holds for a mid-sized business: AI is only as reliable as its picture of how you really operate.

You do not need to fix everything first. Map only the systems your first AI project will touch, and fix what creates immediate risk. Paying down the rest will be a longer-term process.

Before your next AI project, go hunting. List the systems it will touch, and for each one ask who knows how it connects to the rest and whether that is written down. If the answer depends on one person, you have found one component of your tech debt. Finding it now gives the project a far better chance of working.

Sources

¹ Julien Marx, Vincent Delaroche, Michael Fraser, Jean-François Bobier, Fabrice Bardon and Charles Grenet, “Six Myths CIOs Must Avoid in AI-Based IT Modernization,” Boston Consulting Group and CAST, August 2026.