Everyone has an AI strategy. Who is redesigning the work?
Everyone has an AI strategy. Who is redesigning the work?
Which piece of work is done differently because of AI?
Which step in that work no longer exists?
These questions are harder than asking how many employees use an AI tool or how many use cases have been collected. Tool usage is visible. A list of pilots is easy to present. Workflow change requires leaders to make choices about roles, risk, quality and the use of time saved.
In our AI in the Room 2025 survey, 83 percent of respondents said leaders talk about AI at town halls. Fewer than one in five said leaders use AI tools daily in their own work. Only 16 percent of firms described AI as deeply integrated, while 42 percent said they were still dabbling.
The figures came from more than 100 professionals across technology, finance, manufacturing, retail, consulting and the public sector. They point to an alignment question. What leaders think is happening, what they say about AI and what teams do with it can be three different things.
When a use case leaves the workflow unchanged
Take a familiar task such as preparing a monthly business review.
An AI assistant can summarise reports, compare performance with the previous month and draft commentary. That may save an analyst several hours. If the review process remains unchanged, the analyst still produces the old pack, leaders still read it late, and the meeting still spends its time explaining numbers.
The tool has improved one activity. The surrounding work remains the same.
A redesigned review might give leaders access to standard analysis before the meeting, ask the analyst to investigate anomalies, move routine commentary out of the agenda and reserve meeting time for decisions. This changes preparation, roles, information flow and the purpose of the meeting.
The distinction matters because productivity gains often disappear into existing work. People fill the saved time. Leaders see faster drafting and wait for financial value that was never designed into the process.
McKinsey’s 2025 global survey on AI found that workflow redesign had the strongest association with reported EBIT impact among the organisational attributes it tested. Yet only 21 percent of respondents whose organisations used generative AI said that at least some workflows had been fundamentally redesigned.
Buying the tool may be the simplest decision in the programme.
Where AI strategies get stuck
Use cases are collected before work is understood
Employees are invited to suggest AI ideas. The list grows quickly: proposals, customer support, recruitment, forecasting, coding, learning and knowledge management.
Some ideas solve a real problem. Some automate an activity that should be removed. Some shift work to another person. Some save minutes in a process delayed by days of approval.
Before approving a use case, map the work around it. What triggers the process? Who uses the output? Where does it wait? Which errors matter? Which exception consumes most of the effort?
AI applied to a poor workflow can make poor work arrive sooner.
Training is separated from responsibility
A general workshop can help people understand the tools. The account manager still needs to know whether client data may be used, whether an AI-generated proposal needs review, and whether the time saved should be spent on account research.
Training becomes useful when it is tied to a role and a sanctioned workflow. The employee should know what they may do, what they must check, where human judgement remains essential and whom to ask when the situation falls outside the rule.
Policy is written for every possible risk
In the absence of a clear operating choice, policies try to cover the whole company. They become long, cautious and difficult to apply. Employees either avoid useful tools or use unsanctioned ones with more convenient rules.
Start with specific workflows and their actual risks. A public marketing draft, an internal performance assessment and a clinical recommendation should not share the same approval path.
Productivity is declared without deciding where the time goes
Suppose a team saves 20 percent of the time spent on a recurring task. What happens to that capacity?
Will the team handle more volume, improve quality, reduce turnaround time, spend more time with customers, or reduce cost? Each answer requires a different redesign and a different measure.
“Productivity” is incomplete until leaders make that allocation choice.
Ownership sits with the team that bought the tool
Technology can own platforms, integration and technical controls. It cannot decide how a sales manager should review an AI-generated account plan or how a recruiter should weigh an AI-assisted screening result.
The business owner of the workflow has to own the outcome. Risk, technology, HR and legal contribute boundaries and expertise. The person accountable for the work decides how those inputs fit together.
Six questions before an AI pilot
Choose one workflow and answer these questions in a room with the people who perform, manage and receive the work.
1. What outcome are we improving?
Be precise. Faster proposal turnaround, fewer missed contract risks, better forecast accuracy and higher first-contact resolution are outcomes. “Use AI in sales” is an instruction to search for an outcome.
2. Which part of the workflow causes the current constraint?
AI may help with information retrieval while the process waits for approval. It may generate an answer while the employee lacks authority to act on it. Find the real constraint before selecting the tool.
3. Which activities will change or disappear?
Name the current steps. Decide which will be automated, assisted, combined, moved or removed. If every old step survives, the company may be adding AI work to existing work.
4. Where is human judgement required?
Identify decisions involving material risk, ethics, sensitive data, ambiguous context or a promise to a customer or employee. Define who checks the output and what evidence they need.
Human review should have a purpose. Asking a person to approve every output without time, criteria or authority creates a ceremonial control.
5. What will we do with the capacity created?
Make the allocation explicit. If saved time is intended for customer conversations, adjust goals and calendars. If it should increase volume, change capacity assumptions. If it should improve quality, define the quality measure.
6. Who owns adoption and value?
Name one business owner. Give that person authority over the workflow, access to the relevant measures and a way to resolve cross-functional issues. Tool deployment, employee usage and business value are three separate measures.
Start with one piece of real work
A practical AI redesign can be done in five stages.
Observe the current workflow, including waiting, rework and exceptions.
Establish a baseline for time, quality, volume and risk.
Redesign roles and steps before configuring the tool.
Pilot with a small group using real work and clear guardrails.
Review outcomes, employee behaviour and unintended effects before expanding.
The pilot should produce a decision. Scale it, change it or stop it. An endless pilot protects everybody from having to make that decision.
Leaders have work to redesign too
Employees notice how leaders use AI. They also notice whether leaders keep asking for the same reports, approvals and presentations after announcing an AI transformation.
Pick one leadership workflow: the monthly review, investment proposal, customer escalation, hiring review or strategy update. Apply the six questions. Let the organisation see what changed, what was learned and which safeguards were necessary.
This does more for credibility than another speech about an AI-first culture.
At the next review, count your pilots if you need to. Then ask which workflow has fewer steps, which role has changed, which decision is better and where the saved capacity went.
If the room can answer those questions, the AI strategy has entered the work.
The Execution Readiness Assessment can help leadership teams examine the gap between AI ambition, communication and everyday use.