
Business units follow an AI adoption strategy when each unit gets four things: one workflow it chose itself, an owner inside the unit, a redesigned process and a result it can report as its own. A strategy that arrives only as licences and a mandate leaves the old process in place. The teams closest to the work then keep doing it the old way.
In June 2026, Deloitte's AI Pulse Check surveyed nearly 3,700 professionals. Of those, 48% say their organization introduced AI without redesigning the workflows or roles around it. Only 12% report redesign at scale, with a new operating model behind it. Closing that gap means changing how each unit runs.
A mandate asks a unit to change its tools while its targets, roles and daily process stay the same. The people in the unit carry the risk of the change and get little of the credit, so waiting it out is the sensible move. BCG's AI at Work survey, published in June 2025, asked more than 10,600 workers in 11 countries. Regular AI use among frontline employees had stalled at 51%.
The same survey shows what raises it. Where leadership support was strong, the share of employees who feel positive about generative AI rose from 15% to 55%. Yet only about a quarter of frontline employees said they get that support. Regular use was sharply higher among employees with at least five hours of training and access to in-person training and coaching. BCG's rule of thumb for AI programs puts 70% of the effort on people and processes. The rest is split between technology, data and algorithms.
Where a company sits overall matters too. Our AI adoption framework sets out the five stages and a check for which one you are in. This post covers the part inside each business unit: who owns the workflow, what the unit gets, and how the work changes. The company-wide choices, such as which goals AI serves and in what order, belong in the enterprise AI strategy.
Deloitte found that among organizations making changes, 37% begin by taking full ownership of one workflow, testing it and then scaling up. The unit should choose that workflow from its own week, because a unit defends the problems it feels at the next budget review. Good candidates repeat every week, take hours of skilled people's time and have a result anyone can count. Invoices approved, tickets triaged and reports produced are all good examples.
The AI Use-Case Canvas puts one workflow on a page before anyone builds it. The automation ROI calculator turns hours saved, build cost and running cost into the month the build pays for itself. The unit head can take that date to finance.
The rollout below follows one workflow from the unit's choice to the point where the agent works with less supervision. Click through the steps to see who decides at each one, what the unit gets and what happens if you skip it.
The order matters more than the speed. Redesign the work before you automate it, and record the baseline before go-live. Give the agent more autonomy only after the logs show it handles the work. An AI pilot runs the redesigned workflow on your test (staging) systems for 2 to 6 weeks, against a pass mark set in the first week. It ends in a go/no-go report on accuracy, response time (latency), token cost and total cost of ownership. If the report says go, an AI consulting firm that also builds and ships the software can take the same workflow into production without handing it to another team.
Ownership splits four ways, and the split works only when each role has a named person before the build starts. The workflow owner sits in the business unit, and that is the role most strategies skip.
| Role | Who it is | Owns | Answers for |
|---|---|---|---|
| Sponsor | The executive over the unit | Priority and budget for the workflow | Whether the unit has time to do the redesign |
| Workflow owner | A lead inside the unit | Rules, thresholds and changes to the workflow | Results against the baseline |
| Builders | Engineers, in-house or a partner | The integration, the agent and its monitoring | Uptime, failure alerts and fixes |
| Governance | Risk, security or an AI council | Which actions need approval and who sees what data | The autonomy level of each agent |
The workflow owner needs real authority over the workflow. When we built daily MI alerting for IES Limited, we put the alert bands in a control centre the business sets itself. The people who read the alerts decide what counts as unusual. Deloitte puts the question plainly to leaders: who owns the outcome when an AI-driven action goes wrong? Its advice is to make ownership explicit before autonomy expands further.
Redesign means deciding, step by step, what the agent does, where a person decides, who handles exceptions and which steps disappear. McKinsey's State of AI in 2026 looked at AI high performers, the companies getting the most value from AI. Nearly three-quarters of them have fundamentally redesigned workflows because of their AI use, against one-quarter of other respondents.
Build the redesigned workflow where the unit already works. IES Limited's content pipeline runs inside the Notion workspace its marketing team already used. One trigger starts the research, the drafting and a compliance check. The team reviews each draft after the compliance agent has passed it, in the tool it already worked in, as the IES Limited case shows.
The incentives that work are ones the unit can see in its own numbers, and you have to plan them before the build:
Autonomy is how much an agent may do without a person approving it, and it is where adoption and governance meet. Deloitte calls the steps up an autonomy trust ladder, and most organizations sit low on it. Of the respondents, 34% still have people approve every action, and 35% allow only low-risk actions that can be undone.
Each workflow moves up on its own evidence. Before every step up, answer Deloitte's three questions. Which risks are reversible, and which need escalation? Who owns the outcome when an AI action goes wrong? What evidence would justify changing the guardrails? An AI governance framework writes those answers down before the agent goes live, and our agentic AI roadmap shows how autonomy grows across many agents.
Training sticks when it happens on the workflow the unit will run. Our corporate AI training programs build each session around one task from the team's own week, with an engineer beside the team. Each team leaves with a working automation and an owner who keeps it running. Leaders get a separate track on which processes go first, who reviews AI output and in what order teams get the rollout. For the program view across a whole company, see enterprise AI training.
An AI adoption strategy sets which workflows get AI first, who owns each one, how the work changes and how results are measured. A strategy business units follow is built workflow by workflow, with an owner inside each unit.
Give them a workflow from their own week, train them on the live system and make the result theirs to report. BCG found employees were far more positive about generative AI where leadership support was strong. Regular use was also sharply higher among those with at least five hours of training.
A lead inside the unit should own each workflow: its rules, thresholds and results. An executive sponsor owns priority and budget, the builders own the integration and monitoring, and governance owns which actions need approval.
Most add AI to the old process instead of redesigning it. Deloitte found 48% of respondents introduced AI without redesigning the workflows or roles around it. Only 12% report redesign at scale.
Both, split by layer. Shared pieces such as data access, logging and governance rules work best when run centrally. Each unit picks and owns its own workflows. Our five-stage adoption framework covers when to build the shared platform.
If you want to start with one workflow in one unit, book a discovery call. A senior partner will ask which workflow the unit would pick. They will then tell you whether to begin with the AI Audit, which ranks use cases across units in 4 to 8 weeks, or go straight to a pilot.