People think AI is about technology
We've found it's more of a change management problem.
Yes, you might find it hard to pick the right AI model, the right AI vendor, and build the roadmap. But the value isn't in the model.
The value is getting your people, your organization, to actually use AI.
I spent half a year working as a data entry clerk pushing orders into an ERP while the customer was on the phone or sending me an email.
I've spent time with finance teams who take PDFs and go through a bunch of screens to check a PO and its receipt, then rekey the same data into another system.
The folks that are doing the manual process today are the frontline that will be using AI, training it to do their workflows, and making it improve.
And if they don't buy into the solution, your AI transformation will go nowhere.
The adoption layer
There are three layers to a good AI program.
Everyone obsesses over the Model Layer. Wrong layer.
The model matters. Of course it matters. But the model is becoming more available, more commoditized, more baked into everything. Microsoft will have one. Google will have one. Salesforce will have one. Your ERP vendor will eventually buy one, duct-tape it to a 25-year-old workflow, and call it innovation.
The most important part of AI is the Adoption Layer.
If people use AI, AI learns. If AI learns, it gets better. If it gets better, the business case compounds. If people do not use it, you bought shelfware with tokens. That's it.
- The Model layer: the LLM, Agent, or whatever buzzy term you want to use
- The Data layer: the ERP, CRM, emails, PDFs, and data lake
- The Adoption layer: whether the people doing the work actually use the thing on Tuesday morning.
The first step is resolving fear
The first step is not OpenAI.
It's solving for the rumor mill in the break room.
Every employee asks the same question: is this going to help my job or hurt my job?
Do not dodge it. Executives dodge it because they think the answer is delicate. Bad idea. Silence becomes rumor. Rumor becomes resistance. Resistance eats the rollout alive.
Say the thing directly.
AI should take the worst parts of the job: forms, documents, data entry, status checks.
Nobody dreams of keying order lines into an ERP.
The best inside staff usually know the business cold. They know which customer is a pain but worth saving. They know which product substitutes when inventory is tight. They know which supplier always misses the delivery window.
Then the company makes them stare into a tiny screen and type.
AI should move those people up the value chain. The machine should handle the grunt work. The human should do the work that requires judgment.
The spoon problem
There is an old story about when Milton Friedman went to China. He sees workers digging with shovels and asks the Chinese representative why they weren't using machines. The Chinese rep says that machines would eliminate jobs. Friedman replies, "Then give them spoons and you'll have even more jobs."
The goal of a business is not to preserve spoon work. The goal is to make, move, and sell things better.
Traditional businesses do not have a giant surplus of good people waiting to be eliminated. Manufacturers, distributors, regional banks, utilities, logistics firms, and healthcare operators are already short on talent. They have retiring operators, broken systems, customer pressure, margin pressure, and too much manual work around the house.
AI should not be used to scare good employees.
It should give them better tools.
Like the vacuum and the broom. You still need to clean the house. You just have a better way to do it.
Start with pain, not the board presentation
Most executives want to start with the big use case.
Demand forecasting. Inventory planning. Autonomous pricing.
Makes sense. These are juicy. They show up well in board decks. They sound strategic. They also take longer, cost more, require cleaner data, and demand more organizational trust than the company usually has on day one.
Bad first move.
Crawl: kill stupid work
Start where the pain is obvious.
Order entry. Invoice entry. AP matching. AR collections.
Why?
Because nobody needs a consultant to explain why the current process sucks. The baseline is visible. The ROI is measurable. The user already hates the work. The manager already knows the bottleneck. The CFO already sees the cost.
That is where you get the first win.
And the first win matters more than the first architecture diagram.
A distributor does not need an AI center of excellence before it fixes order entry. A manufacturer does not need a 9-month AI strategy before inside sales can quote faster. A regional bank does not need a 60-slide maturity model before it automates document intake.
Kill stupid work first.
Morale goes up. Cycle time goes down. The next rollout gets easier.
Walk: make AI a co-worker
Employees do not want transformation. They want help. That distinction matters.
Do not tell an AP clerk that AI is here to "reimagine finance operations." She does not care. Tell her AI will read the invoice, match the PO, find the receipt, and flag the exception so she can stop doing manual cleanup.
That lands.
Do not tell an inside sales rep that AI will "unlock revenue intelligence." He does not care. Tell him AI will enter the order, suggest the cross-sell, check the price, and remind him which customer always buys filters with the pump.
That lands.
The AI co-worker needs to live inside the work. Not next to it. Not in a chatbot tab. Not in some sandbox nobody opens after kickoff.
Inside the work. ERP. CRM. Email. Documents.
This is why generic AI disappoints traditional businesses. It can summarize a meeting. Great. It cannot fix the order if the customer email has the PO, the price is in the ERP, the exception is in someone's inbox, and the customer's real preference lives in a rep's head.
Business AI has to absorb the messy reality, not pretend it away.
Run: understand the business
Once AI lives inside the work, the C-suite gets the real prize.
Understanding.
Not dashboards. Dashboards tell you what happened after the fact.
Understanding tells you where the business is breaking while it is breaking.
A CEO can see which customers are valuable but underserved. A CFO can see where cash gets trapped. A COO can see which branch routes around process. A CRO can see which reps discount because they lack confidence, not because the customer demanded it.
I have sat with teams where the "standard process" was not standard at all. The best employee had been manually fixing the same broken workflow for years. Everyone knew it. Nobody had instrumented it. AI makes that visible. That is where the value is.
But the executive use case depends on the frontline adoption. The boardroom answer depends on the order desk. The strategy depends on the clerk.
The executive has to push
A lot of traditional companies have a pull model for IT.
Someone asks for software. IT evaluates it. A tool appears. Adoption is optional. The branch manager ignores it. The regional VP shrugs. Everyone keeps working the old way. That model fails with AI.
AI is not optional homework. If the CEO treats AI like an IT project, the company will treat it like an IT project. And IT projects are where urgency goes to die.
Executives need to push. Push means picking the workflow, naming the owner, measuring the outcome, and telling managers adoption is not optional where the process changes.
The roadmap is not in the board deck. The roadmap is where employees alt-tab, copy-paste, rekey, check, chase, wait, and swear under their breath. That is the field evidence.
The communication doctrine
Executives need to say three things: AI removes bad work, the company will train people, and adoption is mandatory where the workflow changes. Then they need to prove it.
Do not announce AI and vanish. Do not let a vendor run one Zoom training and call it change management. Do not let middle management stall the rollout because the old process feels safer.
Fear fills empty space. Leadership has to occupy it first.
The vendor test
Every AI vendor claims they have the model. Who cares? Ask a better question: who gets our users to change their behavior?
Can they sit with the AP team and understand why the three-way match fails? Can they watch inside sales and see where orders get stuck? Can they integrate into the ERP without turning the project into a funeral march? Can they tie the work to invoices processed, orders entered, cash collected, and quotes created?
If the answer sounds like a webinar, run. If the answer sounds like a work plan, keep talking.
The best AI vendors will look less like software vendors and more like operators with software. They will know the workflow, the systems, and the adoption problem.
Camp A and Camp B
There are two types of C-level executives right now. Camp A wants to get the board off their back. Camp B wants to transform the business.
Camp A should buy Microsoft Copilot, tell the board they have an AI strategy, and wait five to ten years for the ERP vendor to acquire an AI company and wedge it into the legacy stack.
Camp B should move now: specific workflows, forced adoption, trained users, measured outcomes.
Camp A buys permission. Camp B builds advantage.
The scorecard
Do not measure AI by seats. Seats are how software vendors get paid. Outcomes are how businesses justify spend.
A year after kickoff, no board cares how many prompts employees wrote if cash is still trapped in receivables and orders are still sitting in email.
Measure the work: orders entered, invoices processed, cash collected, margin recovered.
That is the scorecard. Not "AI readiness." Not whatever phrase the committee landed on after three rounds of comments.
The work.
The closing reality
Factories did not get the value of electricity by swapping motors into old layouts and calling it a day. They got the value when they reorganized the factory around the new power source.
Same thing here.
AI is the new power source. But most companies will bolt it onto old workflows, call it transformation, and wonder why nothing changed.
The winners will reorganize the work.
They will move employees out of sludge and into judgment. They will make managers enforce adoption. They will make vendors own outcomes. They will use early wins as proof for the next workflow.
AI will not reward the company with the cleanest strategy deck. It will reward the company whose people use it when the phones are ringing, the truck is late, the invoice is wrong, and the customer still expects an answer.