TL;DR
- Not all resistance to AI is about protecting a job. A large part of it is a considered judgment that the model cannot yet do the work.
- The underwriter is not transcribing data. Every cell they copy by hand is checked against a pattern in memory, and anything that stands out gets scrutinized on the way in.
- Underwriting an asset is still an art, in the strict sense: hard to reproduce without an individual's practice. The moment it can be mass-reproduced it stops being art and becomes production.
- The way in is not to displace the ritual but to clear the ground around it. Sell the power drill, not the humanoid.
Let’s get past the obvious push-backs – both the unspoken and the vocal – to save one’s own job. That is understandable. But there is another. There is a pushback that comes not from fear of relevance but lack of trust in AI’s ability to do the job.
For those with the vested interest – the companies that want to embrace AI and the ones that sell AI solutions – the resistance may seem antidiluvian. We tend to lump all resistance as resistance to change, as though AI is unmistakably perfect.
I will give you examples.
The underwriter and the telehandler
We are studying the equipment leasing market. Someone I know sells AI to lenders. The parallels are just the same. We want to sell the best underwriting software that looks at two dozen variables and predicts asset value or the quality of the leasee.
For someone who has looked at equipment of all shapes and sizes all-day long for 20 years, it is unrealistic that AI can figure the asset’s residual value all by itself. At the simpler end, an underwriter pulls data from auction sheets, economic outlook data from internal economist’s reports, price benchmark sheets that come out annually, and 5 more sources that I cannot remember. What happens when they transpose the data into their underwriting sheet by hand is, every cell’s input is checked against a pattern stored in memory.
“Huh! That’s a 2022 Merlo ROTO 50.26 S Plus Telehandler which means US EPA Label maybe absent”.
There is a mental note, a verification, and a discount math all going on with just the title of the asset they are looking to underwrite. In each subsequent step, as they copy over the data, they are finding something that matches the pattern they know or stands out. Anything that stands out, gets scrutinized, not when they sit to run the model but as they put the data in.
An AI underwriter can harmonize and assemble the data in 2 minutes instead of 45 minutes. It can do a dozen pattern-matches and keep track of all variables that may affect the residual value of that asset.
Why this work is still an art
Humans call something an art when it is hard to reproduce without an individual’s practice and skill. If it can be mass-reproduced it is a production act and not an art. Underwriting an asset today is art. How do I know what politics, tariff, oil prices, and weather’s impact on next season’s crops would do to the price of an asset in the secondary market? I don’t. So I will apply my mastery and artfully arrive at a price.
The place to start is not confrontation
AI can do a better job, for sure. But the place to start isn’t one of confrontation. You cannot ask an underwriter to move aside from their ritual without proving that AI can do better. But to prove that you can do better, you need to be on the driver’s seat, right?The way to bring about change is to not ham fist your way into the heart of the problem, but enable the underwriter to perform the art. Pull data from different sources, fix the label problems, assemble the data into their spreadsheet format, open one new easter egg every few days, and make them realize that AI has made them focus on the art and not hand-copy the notations into a new sheet, the night before the performance. Why the sequencing matters as much as the capability is the subject of Instrumenting Work to Avoid Expensive AI Rollouts, and what the underwriter actually knows that the model does not is the argument in Don't shortsell your know-how to AI vendors.
A bicycle for the performer's mind
For every organization there is a coveted prize; the place where AI will open the floodgates of accuracy and scale of action. The path to that prize is not cleared by a bulldozer driven by the C-suite but a bicycle for the performer’s mind.
Resist selling AI as the humanoid. It's tempting. What will get accepted is a power drill.
Where Moative fits
Moative works this way by design: we start by removing the manual work around the expert rather than aiming at the expert's judgment. That ordering is slower to demo and considerably more likely to survive contact with the people who do the job. In healthcare revenue cycle management the same pattern holds, where the experienced biller is doing pattern recognition on the way in and not at the end. The resistance you hear in the room is only half of it. The other half is the colleague who never argued and quietly built their own tool instead, which is the subject of Your AI readiness blindspot.
Frequently asked questions
Why do employees resist AI tools?
Two different reasons get treated as one. Some resistance is about job security, which is understandable and predictable. A great deal of it is a technical judgment: the person doing the work has seen the model fail at the part that actually matters, and is telling you the tool is not ready.
Is resistance to AI just fear of change?
Often not. An expert who rejects a model is frequently making an accurate assessment of what the model cannot yet do. Treating every objection as change resistance means discarding the most informed quality signal available.
Can AI replace an experienced underwriter?
It can do the assembly far faster: harmonizing sources, matching patterns, tracking variables. What it does not yet do is the judgment about how politics, tariffs, commodity prices and weather will move a secondary-market price, which is the part the underwriter would call the art.
What does the expert actually know that the model does not?
Mostly the checks that happen during the work rather than at the end. Recognizing from an asset title that a compliance label may be missing, and applying a discount for it, happens while the data is being entered. That judgment is not written down anywhere the model could read.
How do you introduce AI without triggering resistance?
Start with the work around the expert, not the expert's decision. Pull the data, fix the labels, assemble the sheet in their format. Earn trust through repeated small wins before asking for any of the judgment, because you cannot prove the model is better without first being allowed to run.
What is the difference between an art and a production act?
An art is hard to reproduce without a specific person's practice and skill. A production act can be mass-reproduced. The distinction matters for AI adoption because it identifies which parts of a job can be automated now and which part is the reason the person is still there.
