In 2016, we nearly bought a company that would have cost us far more than capital.

It came through a well-regarded intermediary we had closed several good acquisitions with. Let's call the company "CoreVision." CoreVision served a highly specialized vertical - government records management. The economics were textbook VMS: 94% gross margins, recurring revenue exceeding 85%, decades-long customer tenure. At first glance, it was exactly the kind of business we were built to buy. An AI deal-screening tool, optimising for pattern matches with past acquisitions, would almost certainly have ranked it highly.

We met the founders. They were sharp. One was an ex-civil servant with deep domain knowledge, the other a technical lead who’d built much of the product in the early 2000s. They’d bootstrapped, never taken outside money, and claimed to have grown EBITDA at double digits for the last seven years. They were looking to retire, and the price, while rich, was not outlandish.

So far, it looked like a fit.

But something didn’t sit right.

The first clue came during a site visit. Our team was talking with select senior staff. The staff were courteous but cautious. More than one person mentioned being unsure what their long-term role would be, even though the company supposedly hadn’t communicated any changes yet. When one of our team asked how product decisions were made, a senior engineer laughed. “Depends who’s yelling louder that week.”

It wasn’t data - but it was a signal. A small alarm bell ringing somewhere in the gut.

Turning Gut Into Data

We don’t make decisions on gut alone. But we pay attention to it. Because good gut feelings are just fast, unconscious pattern recognition - reflexes trained over time. When something feels wrong, our job is to dig until we find out whether the feeling is justified. Gut is a hypothesis. We went looking for the data. An AI model, presented with the same financial profile, would not have known to look.

Gut is a hypothesis. We went looking for the data.

We pulled their ticketing system data. The customer support backlog was massive, trending upward for six consecutive quarters. Their customer churn, which they hadn’t mentioned, was quietly increasing - just a few percentage points each year, but the pattern was consistent. The NPS scores were sporadic and selectively reported.

But the real red flag came when we conducted blind customer interviews.

One customer, a mid-size municipality, told us: “We’ve been using CoreVision for years, but we’re planning to migrate. They’re just not responsive anymore. We can’t wait three weeks for a ticket to get looked at.” Another said, “Every time they release a new feature, it breaks something else. We’ve stopped upgrading.”

CoreVision was not dying. But it was deteriorating in slow motion.

And deterioration is rarely visible in financials - until it's too late.

Still, the financials showed strength. We stress-tested the forecasts. We asked ourselves: could we fix this post-acquisition? Possibly. But that wasn’t the right question.

The right question was: should we need to fix it at all?

We make our returns by buying good businesses - not fixer-uppers. CoreVision might have once been a gem. But it had become a liability hiding inside a historical P&L. A beautiful carcass with a decaying nervous system. We walked.

A beautiful carcass with a decaying nervous system.

The deal closed three months later - at a higher price - with a private equity firm that specialized in software turnarounds.

Within two years, CoreVision had lost its top five customers. A whistleblower surfaced. A regulatory issue, previously undisclosed, emerged regarding data handling in one of their largest contracts. The PE firm had to write down the asset. And quietly, they shopped it again, this time to us.

We passed. Again.

I’ve thought a lot about that deal. On paper, we would have made it. The financial profile, the size, the industry. But the gut said no. And when we listened to the gut, then searched for supporting data, we found truth that wasn’t in the CIM.

What AI Screening Cannot See

I think about CoreVision when colleagues ask whether AI-powered deal-screening tools might sharpen the acquisition process. The answer is: perhaps, at the surface level. A model trained on the financial signatures of successful VMS acquisitions would, in all likelihood, have scored CoreVision very highly. 94% gross margins, recurring revenue above 85%, decade-long customer tenure: these are precisely the patterns such a system is calibrated to surface. What it cannot detect is a senior engineer laughing when asked how product decisions are made. It cannot weight the significance of staff who are evasive about their own futures. It cannot interpret the silence around an NPS figure that is conspicuously absent from the reporting pack. The due diligence that saved us was human judgment operating on soft signals, signals that a financial model is not designed to process, and that an AI screening tool, calibrated on what prior winners looked like, would almost certainly have filed under "noise."

The lesson is simple. Never ignore the gut - but never follow it blindly. Use it to ask better questions. Demand better data. Your instincts, when trained on enough reps, know more than your spreadsheet will tell you. But verification is the duty that keeps instinct from becoming arrogance.

Had we done that deal, I suspect we’d still be picking up the pieces today.

Instead, we walked - and walked into better deals with fewer scars.