A copilot in Sales. An AI assistant in HR. An agent in Procurement. Every team is faster, and enterprise performance has not moved. That gap has a name: AI change management. Not the training kind. The kind that changes how a process runs across teams. It is unavoidable. It can also be much smaller than most companies think. Change does not disappear: it gets a size you can own.
Every team got its AI. Individual productivity moved. Not enterprise performance.
Teams adopted tool after tool. Today, AI and agents are everywhere. A rep drafts an email in seconds. A manager gets a call summary before the next meeting. An employee finds the travel policy without opening a ticket. Each of these saves minutes for one person.
The results tell another story. In McKinsey’s State of AI survey, published in August 2026, only 37% of respondents attribute any EBIT impact to AI, the same share as a year earlier.
AI for individual employees is easy to roll out and saves each person time. That is productivity. It does not scale revenue or results. The quarter, the close and the onboarding cycle do not depend on how fast one person works. They depend on how work moves from one team to the next. That is performance, and it is where most AI projects stall: no measurable ROI, and pilots that never end. We made the same point for sales teams in AI made every rep faster. Why didn’t the quarter move?. It holds for every team.
Adoption is free, transformation pays: the rule of every new technology
AI is not the first technology to show up everywhere except in the results.
The first central power stations opened in 1881. Yet in 1899, electric motors still drove less than 5% of the machines in American factories, and productivity in manufacturing did not take off until the early 1920s, four decades later. The economist Paul David explained why in 1990. At first, factories added electric motors to their existing shafts and belts and kept the old layout. The gains came when each machine got its own motor. Plants could then be redesigned: single-storey buildings, better materials handling, machines placed and moved where the work needed them. It happened through a long, decentralised learning process, factory after factory, not through one big switch.
Computers followed the same path. In 1987, the economist Robert Solow wrote: "You can see the computer age everywhere but in the productivity statistics." The gains showed up in the 1990s, once companies reorganised how work ran around the new tools.
Erik Brynjolfsson, Daniel Rock and Chad Syverson gave the pattern a name in 2021: the productivity J-curve. A technology that changes everything first asks for investments nobody sees: new processes, new skills, a new organisation. Measured productivity dips, then climbs.
Nearly 9 in 10 companies use AI. 37% see an impact on EBIT. That gap is today’s version of the same rule. The dip is the time it takes to change the process.
Where the change sits: in the coordination between teams
The change that pays does not sit inside one tool or one team. It sits between them.
Take quote to cash. A qualified deal crosses Sales, the deal desk, Legal, Finance and Customer Success before it becomes cash. Each team has its own tool, its own inbox and its own calendar. A discount gets approved in a thread. A signed contract waits two days before Finance sees it. A commitment made to win the deal never reaches Delivery. MGI Research puts the loss at 3 to 5% of revenue at least, mostly between quote and contract, and between contract and cash. Our quote-to-cash guide walks through each of those leaks.
Onboarding tells the same story in HR. A new hire touches HR, IT, Facilities and the manager. Each team does its part well. The link between them is often someone’s memory or someone’s spreadsheet.
AI assistants save time for individual employees and make each step faster. They do not move the work from one team to the next. That is why the first 48 hours after a deal is signed still decide so much of next year’s revenue.
"No change management" is half true
Two stories dominate the market. Both are half true.
The first says AI needs no change management. For using AI, that is right. When AI answers and acts in Slack or Microsoft Teams, nobody learns a new tool. Adoption is close to free. But the process stays the same, so the results stay the same. Picture a rep who drafts a discount request in thirty seconds with an AI assistant. The request then sits four days in Finance’s inbox, waiting for someone to look at it. The AI saved ten minutes. The process lost four days.
The second says AI needs a transformation program. That is right too: something has to change. But a program has a sponsor, a budget, a steering committee, and often nobody who owns the work end to end. It is designed far from the work, and by the time it lands, the process has already changed twice. Picture an eighteen-month roadmap, process-mapping workshops, a target operating model signed off by a steering committee. Meanwhile, the pricing policy changed twice, two approvers changed roles, and the team kept working around the plan.
Each one does something well. AI assistants save time for individual employees and get people comfortable with AI. A program brings alignment and a governance frame. Neither changes the coordination between teams, which is where the results are.
Companies feel it. In the Ofelia Agentic Transformation Summit 2026 barometer, change management ranks among the top three blockers to scaling AI, cited by 48% of respondents.
AI change management, sized: the process owner leads
The right size for the change is one process, led by the person who owns it.
The capability moves to the business. The owner of the process leads the change: RevOps for quote to cash, HR Ops for onboarding, Finance Ops for the month-end close. They know where it breaks. They own the KPI. Until now, they needed a developer or a consultant to change anything.
IT keeps control. IT decides who can build and publish, which data and systems a process can reach, and how every step is traced. IT governs the frame. The business moves inside it.
The process will not stay as it is. With AI and agents inside it, a process can do more. An agent checks a request against the policy, prepares the onboarding plan, flags a missing document. A person validates each result. Bringing in these capabilities becomes an edit to the process, not a new project.
What it looks like in practice. Take onboarding. The HR Ops lead starts from the onboarding checklist the team already uses. She adds one agentic step: an agent prepares the IT access request from the contract, and the IT manager validates it. IT approves the connection to the HR system once. In the first month, she measures what she already tracks: the time it takes a new hire to be productive. When a new site opens, she adapts the process herself, the same week. The expected result: an onboarding cycle 42% shorter.
What IT keeps. Sizing the change to the business owner does not mean taking it away from IT. IT decides once which systems can be connected, who can build, who can publish, and what gets logged. Every agentic result has a person who validates it, and every step leaves a trace. IT stops being the bottleneck for every change and stays the guardian of the frame.
Signs you sized it right. One person can explain the process in two minutes. It already has a KPI someone tracks. It can run in weeks, not quarters. And it can change again next month without a new project.
Why it works. The people who feel the friction are the ones who fix it. The change is small enough to test, and it is measured on the KPIs the process already tracks: request-to-decision time, days from signature to cash, time to productivity. Adoption follows on its own, because the right path is also the fastest one.
“Adoption is not a campaign. When the right way to run a process is also the fastest one, and it sits where people already work, nobody needs to be convinced.”
Loic Barbaux, Product Marketing Manager, Ofelia
How Ofelia Agentic makes change simple to lead and easy to evolve
That is exactly how Ofelia Agentic was designed. Ofelia Agentic is the governed agentic execution layer for enterprise operations, inside Slack and Microsoft Teams. It works with your CRM, your HRIS and any system you use, and leaves every decision to a person. Each of its three experiences takes one part of the change off the table.
First, a return before any change. Ofelia answers and guides. "Can I offer this discount on this deal?" "What does a new hire need on day one?" "Who approves a contract exception?" Ofelia answers from your approved documentation and your connected systems, and guides the person to the next step without leaving Slack or Teams. Nothing in the process has changed yet, and people already get value from day one.
Then, a change nobody has to be convinced to adopt. A process needs more than answers, so Ofelia coordinates it as a deterministic workflow: the same steps every time, each owner receiving theirs with a deadline, and every decision traced. Take a deal at the offer stage, call it Acme. The rep launches the quote-to-cash workflow for Acme in Slack with one message, asking for a discount above their limit and non-standard payment terms. Ofelia checks the request against the policy. What is standard goes through. What is not goes to Finance and the sales director at the same time, each with a deadline. An agentic task checks the customer’s payment history and brings Finance a recommendation. Finance decides. If an approver does not answer, Ofelia reminds them, then escalates. Once both say yes, the quote is updated and the contract goes to Legal, and the deal moves on through quote to cash. Nobody learned a new tool. The rep sent one message, and every approver answered where they already work. The right path was also the fastest one.
Watch one deal orchestrated from lead to renewal, in four minutes.
And the change itself stays small, because Ofelia puts it in the hands of the process owner. Ofelia generates the workflow from what you already have. The owner shares the procedure or describes it in plain language. Ofelia proposes the model: steps, owners, validations. The owner adapts it in the no-code Workflow Designer, tests it and publishes it into Slack or Teams. No developer, no project. When the process needs to change, the owner edits it and publishes again: everyone runs the new version from the next request, and the procedure Ofelia answers from is updated with it. Bringing in a new AI or agentic capability is one more edit, and a person still validates its result. IT keeps control of who can build and publish, of access and connections, and of the full audit trail. That is AI change management at the size of one owner: the person who feels the friction changes the process, the same week, inside a frame IT set once.
Run this way, the time from deal to delivery is halved and billing errors drop by 70%. Under the hood, Ofelia builds on Bonita BPM, which has run enterprise processes for more than 15 years at more than 130 enterprise customers.
Download the Ofelia Agentic one-pager (PDF), or see how operations and sales teams run their processes with Ofelia.
Start with one process, the one where the handoffs hurt most. For most companies, that is revenue. Change does not disappear: it gets a size you can own.
Sources: McKinsey, The State of AI: Global Survey 2026, August 2026 (nearly 9 in 10, 8 in 10, 37%). Paul A. David, “The Dynamo and the Computer”, AEA Papers and Proceedings, May 1990. Robert Solow, “We’d Better Watch Out”, New York Times Book Review, 12 July 1987. Erik Brynjolfsson, Daniel Rock, Chad Syverson, “The Productivity J-Curve”, American Economic Journal: Macroeconomics, 2021. MGI Research, Revenue Leakage, December 2025 (3 to 5%). Ofelia Agentic Transformation Summit 2026 barometer (48%). Expected results: Ofelia Agentic use cases Revenue Lifecycle, Quote-to-Cash and Employee Onboarding.
Schedule a personalized demo and see how Ofelia can transform your operations.
