Why AI training without workflow redesign often fails

The familiar pattern

The pattern is familiar. An organization buys AI licences for its team. Someone runs a workshop. For a week or two, people experiment: summarizing a report, drafting an email, asking a few questions. Then, slowly, usage drops. A month later, a handful of enthusiasts still use the tools and everyone else has gone back to the old way of working.

The usual conclusion is that people need more training. Often, they don't. They need a different way of working.

What the evidence says

In its 2025 global survey on AI, McKinsey tested 25 factors to see which ones were linked to organizations seeing a real financial return from generative AI. The factor with the biggest effect wasn't the choice of model, the size of the budget or the amount of training. It was whether the organization had redesigned its workflows. Yet only around one in five organizations using generative AI said they had fundamentally redesigned even some of their workflows.

McKinsey's later 2025 survey found the same pattern. The small group of organizations getting significant value from AI were nearly three times as likely as others to have fundamentally redesigned how individual pieces of work get done.

Boston Consulting Group reached a similar conclusion in its 2025 study of more than 1,250 companies. Only about 5% were seeing real value from AI at scale, while around 60% reported little or none. The leaders expected most of their value to come from reshaping their processes, not from adding tools to the old ones.

Meanwhile, demand for training keeps growing. In a 2026 survey of small and mid-sized businesses, 70% said they needed more training to use AI effectively. Training matters. But training on its own is rarely what's missing.

Why training alone doesn't stick

There are four structural reasons standalone tool training evaporates within weeks:

• 1. The training is about the tool, not the job: Most AI training shows what a tool can do: summarize this, draft that, ask it anything. But staff don't wake up wanting to use a tool. They wake up needing to finish a report, answer a client or close the month's accounts. If the training doesn't connect to those specific tasks, it's forgotten by Friday.
• 2. The old process still expects the old output: Imagine someone learns to draft a monthly report in half the time with AI. But the report still has to go through the same five approval steps, in the same format, on the same template. The time saved disappears into the rest of the process, so the new skill feels pointless.
• 3. Nobody owns the new way of working: When AI use is optional and unassigned, it stays optional. If no one is responsible for deciding how a task should now be done, everyone defaults to how it was done before.
• 4. People don't know what's allowed: Without clear guidelines, staff fall into two camps. Some avoid AI entirely for fear of doing something wrong. Others paste sensitive information into public tools without a second thought. Neither is what leadership wanted.

What works instead: train on the work, not the tool

The organizations that make AI stick tend to follow a simple sequence:

• Step 1. Pick two or three real tasks per role: Not "use AI more", but specific, recurring work: the weekly programme update, the first reply to a customer complaint, the monthly supplier reconciliation.
• Step 2. Redesign each task with the people who do it: Map how the task is done today, step by step, including the unwritten steps. Then decide together which steps AI should handle, which stay with a person, and where a human must check the output before it moves on.
• Step 3. Write simple usage guidelines: One or two pages, in plain language: which tools are approved, what information must never go into them, when output must be checked, and who to ask when unsure.
• Step 4. Train on the redesigned task: Now training has a purpose. People practise on their own real work, using the new steps, with the guidelines beside them.
• Step 5. Measure after 30 days: Is the task faster? Is the quality the same or better? Are people still using the new approach? Adjust what isn't working.

An illustration (hypothetical)

Consider an NGO programme team that spends three days each month pulling together a donor report from field updates, spreadsheets and emails.

Training alone: Staff learn to ask an AI tool to "summarize these updates". Some try it, the output doesn't match the donor's template, and they go back to doing it by hand.

Redesign first: The team agrees a standard format for field updates, so the information arrives consistently. An AI step drafts each report section from those updates into the donor's template. A programme officer checks figures and claims against the source data, and the manager signs off. Training then covers exactly this process.

The second approach takes more thought up front. It's also the one still in use six months later.

A note for NGOs and development organizations

For organizations handling beneficiary, health or donor data, workflow redesign isn't only about efficiency. It's where data protection and accountability get built in: deciding what may and may not go into an AI tool, where a human must verify figures, and how outputs are recorded. Guidelines written as part of the redesign are far more likely to be followed than a policy circulated separately.

The takeaway

AI training fails when it teaches people a tool and leaves their work unchanged. It succeeds when it follows a decision about how the work itself should now be done.

Redesign the task. Write the guidelines. Then train people on the work, not the tool.

Planning AI training for your team? Our Team AI Workshop is built around your staff's real roles and tasks. Your team leaves with redesigned workflows, a practical playbook and clear usage guidelines. Write to us at ada@blueberryia.com.