There's a machinist I think about sometimes. He ran a five-axis CNC mill at a plant I visited during a consulting engagement. The engineers who designed the machine probably understood its specifications better than anyone. But this guy — 22 years on the floor — knew something they didn't. He could hear when a tool was about to fail. He knew which programs needed a warm-up cycle that wasn't in any manual. He'd adjusted feeds and speeds over thousands of hours until the machine performed better than its published tolerances.
You don't walk up to that person and tell them they're doing it wrong. You watch. You ask questions. You try to understand what they know that nobody wrote down.
I think AI should work the same way.
The shame cycle
Enterprise technology has a messaging problem. For 30 years, the pitch has been some version of: your tools are bad, your data is messy, your processes are broken — and we're here to fix you.
"Get off Excel." "If it's not in the ERP, it doesn't count." "Your data needs to be clean before AI can touch it."
Meanwhile, the people doing the actual work have been quietly building the tools they need. According to Gartner, 41% of employees acquired or created technology outside of IT's visibility in 2022 — and that number is heading to 75% by 2027. Shadow IT accounts for 30-40% of enterprise IT spending in large organizations.
That's not a rounding error. That's a third of the budget going to tools people chose for themselves. It's not rebellion. It's not laziness. It's the organization solving problems faster than its official systems can.
Excel as signal
Here's a number that should give every enterprise software company pause: approximately 500 million people pay to use Excel. Bloomberg called it an "AI-resistant, multitrillion-dollar empire." Packy McCormick wrote that "Excel never dies." Even Sam Altman's AI hasn't put a dent in it.
Why?
Not because people are stuck in their ways. Not because they don't know better. Because Excel doesn't judge. It doesn't force a process. It doesn't require a six-month implementation, a certified admin, or a data migration project. You open it, and it does what you need. It might be the most empathetic tool ever built.
When a supply chain planner builds a demand model in Excel instead of using the ERP's planning module, they're not cutting corners. They're telling you that the ERP doesn't fit how they actually think about the problem. The workaround IS the process — and it works.
That's a signal worth paying attention to.
The numbers on "rip and replace"
The enterprise software industry's preferred strategy — rip out the old thing, install the new thing, mandate adoption — has a brutal track record:
- 70% of large-scale transformations fail (McKinsey, 2022)
- 55-75% of ERP implementations fail, with manufacturing hit hardest (Godlan/KPC, 2025)
- Only 16.2% of software projects are delivered on-time and on-budget (Standish Group)
- 67% of enterprise software features go unused (Gartner, 2024)
- 42% of companies abandoned most AI initiatives in 2025, up from 17% the year prior (S&P Global)
The pattern repeats with every wave of technology: impose a system from the top down, demand people change how they work, watch adoption stall, blame the users.
What if the users aren't the problem?
A different posture
There's a concept in change management research that's so obvious it shouldn't need to be stated: projects with excellent change management are up to seven times more likely to succeed (Prosci). The way you introduce something matters more than what you introduce.
We think about it as three stages — not as a framework we put on a slide, but as a posture:
Start by connecting. Plug into the workflow that already exists. The Excel model, the emailed report, the homegrown tracker. Don't ask anyone to stop what they're doing. Start where the work actually happens.
Then extend. Add what the current tool can't do on its own. Traceability. Collaboration. AI that reflects the user's expertise rather than replacing it. The existing workflow gets better without anyone learning a new system.
Then earn. Over time, the user stops needing the workaround — not because you killed it, but because you understood them well enough that they didn't need to create the next spreadsheet. They just asked the system to handle it. This stage is earned, never demanded.
What tacit knowledge has to do with it
There's a concept called Polanyi's Paradox, named after the philosopher Michael Polanyi: "We can know more than we can tell." The machinist who hears a failing tool, the supply chain planner who feels a demand signal before it shows up in the data, the quality engineer who senses a bad part before any measurement confirms it — they all carry knowledge that can't be written in a manual or extracted into a database.
This is the most valuable knowledge in any organization. And it's exactly the knowledge that "rip and replace" destroys, because when you throw out someone's workflow, you throw out the tacit understanding embedded in it.
A 2023 NBER study by Brynjolfsson, Li, and Raymond found that when AI was trained on top-performing customer service agents' interactions — capturing their tacit expertise — it boosted novice productivity by 34%. Workers with two months of experience performed as well as those with six months. The AI didn't replace the experts. It mirrored their judgment and made it available to everyone.
That's the model we believe in. Not general intelligence that tries to be good at everything. Reflective intelligence that mirrors how your best people actually think — their seasonality intuitions, their supplier instincts, their process knowledge — and makes it available to the team, the next shift, the next hire.
The reframe
There's a phrase in tech that has always been a warning: "you are the product." It means the platform harvests your data and sells it. You're the supply, not the customer.
What if it could mean something else? What if "the product is you" meant that you externalize yourself into a system that reflects you — your expertise, your judgment, your way of seeing the problem — and for that reason alone, it's valuable? Because you are, and always have been, the most valuable thing in the building. Your systems just never appreciated that.
The next time someone tells you to get off Excel, or that your data needs to be clean before AI can help, or that your workarounds need to be eliminated — consider the possibility that your workarounds are the smartest thing in the building. They're telling you exactly where the real work happens and what your "proper" tools got wrong.
The question isn't how to replace them. It's how to listen to them.