The adoption layer
People think AI is about technology. In practice, it is usually a change management problem.
The value is not just picking the right model, vendor, or roadmap. The value is getting people inside the organization to actually use AI in the work.
There are three layers to a good AI program.
- The model layer: the LLM, agent, or AI capability.
- The data layer: the ERP, CRM, emails, PDFs, documents, and data lake.
- The adoption layer: whether the people doing the work actually use the tool on Tuesday morning.
The first step is resolving fear
The first step is not OpenAI. It is 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. Silence becomes rumor, rumor becomes resistance, and resistance eats the rollout alive.
Say the thing directly. AI should take the worst parts of the job: forms, documents, data entry, status checks, reconciliation, chasing, and copying between systems.
The best frontline staff know the business cold. AI should move those people up the value chain so the machine handles the grunt work and the human handles judgment.
The spoon problem
The goal of a business is not to preserve spoon work. The goal is to make, move, sell, support, and reconcile things better.
Traditional businesses do not have a giant surplus of good people waiting to be eliminated. 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.
Start with pain, not the board presentation
Most executives want to start with the big use case: demand forecasting, inventory planning, autonomous pricing, predictive churn, or strategic workforce planning.
Those use cases can be valuable, but they usually take longer, cost more, require cleaner data, and demand more organizational trust than the company has on day one.
Start where the pain is obvious and measurable. The first win matters more than the first architecture diagram.
Kill stupid work
Begin with workflows nobody needs a consultant to explain: order entry, invoice entry, AP matching, AR collections, claims intake, document review, ticket triage, quote creation, and status checks.
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.
Kill stupid work first. Morale goes up, cycle time goes down, and the next rollout gets easier.
Make AI a co-worker
Employees do not want transformation. They want help. That distinction matters.
Do not tell an AP clerk that AI will reimagine finance operations. Tell her AI will read the invoice, match the PO, find the receipt, and flag the exception so she can stop doing manual cleanup.
The AI co-worker needs to live inside the work: ERP, CRM, email, documents, ticketing, warehouse systems, and the messy handoffs where people already operate.
Understand the business
Once AI lives inside the work, the C-suite gets the real prize: understanding.
Dashboards tell you what happened after the fact. Understanding tells you where the business is breaking while it is breaking.
The executive use case depends on frontline adoption. The boardroom answer depends on the order desk, the branch, the warehouse, the claims team, the banker, the nurse, or the dispatcher.
The executive has to push
A pull model fails with AI. 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 pick the workflow, name the owner, measure the outcome, and tell 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.
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.
Who gets users to change behavior?
Every AI vendor claims they have the model. 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?
The best AI vendors look less like software vendors and more like operators with software. They 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 can buy Microsoft Copilot, tell the board they have an AI strategy, and wait for legacy software vendors to wedge AI into old workflows.
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.
Measure the work: orders entered, invoices processed, cash collected, margin recovered, claims closed, tickets resolved, quotes created, calls contained, and cycle time removed.
The work is the scorecard.
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.
AI is the new power source. 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, make vendors own outcomes, and use early wins as proof for the next workflow.