Stop buying AI. Start redesigning work.

By Kim Moore, Executive Associate, People Solutions Division at Change Logic

Operating model, capability and process, not technology alone, will decide who turns AI investment into measurable value.

Every executive conversation about AI eventually arrives at the same question: how do we move from pilots to measurable business value? The growing discussion around “AI pilot purgatory” suggests that many organisations are still struggling to make that transition.

The organisations seeing real progress are not simply spending more on AI. They are changing the work around it.

Buying an AI tool is the easy part. The harder question is what changes once it arrives. Which tasks should move to AI? Which decisions still need human judgement? Who is accountable when work is shared between people and technology? Unless those questions are answered, the organisation has bought capability without changing performance.

A pilot tests the technology. Production tests the organisation.

Pilots are useful because they show what the technology can do. They do not, however, show whether the business can run it safely, repeatedly and at scale. That requires integration, data, controls, ownership and adoption, as well as a clear view of the workflow the organisation is trying to improve.

Too many pilots begin with the tool and search for a use case afterwards. A stronger approach begins with the work. Which process is slow, costly or inconsistent? Where are decisions delayed? Which steps require judgement, and which are repeatable enough to automate? Once those questions are answered, the business can decide where AI belongs and what must change around it.

This is also why I am wary of organisations pointing to a long list of pilots as evidence of progress. A pilot only matters if it moves into production, changes how work is done and fits into a broader operating-model strategy. Without that, activity increases, but the value remains fragmented.

The value sits in the redesign

The technology is rarely the part that holds an AI programme back. In my experience, the real problems tend to sit around it. The data is unreliable, the process was never properly designed, nobody is quite sure who owns the decision, and leaders assume adoption will take care of itself.

An AI tool can produce an answer quickly. That does not mean the business is ready to use it. Someone still has to decide whether the answer can be trusted, what action it should trigger and who carries the accountability if it is wrong.

This is why simply giving people access to AI changes very little. If the same approval layers remain in place, the same information is trapped in different systems and employees are left to work out the rules for themselves, the organisation has added another tool without fixing the work.

I tend to look for a much simpler test. Is the process now faster? Are there fewer handovers? Are people making better decisions? Has the time saved been put to better use? If the answer is no, usage may be rising, but the business has not changed.

Agents are changing the shape of work

The next phase of AI raises the stakes. Agents can do more than draft content or answer questions. They can complete multi-step tasks, use business systems and coordinate activity across a workflow. That means organisations must decide where an agent may act, where a person must intervene and how responsibility is maintained when work is partly automated.

These decisions reach directly into organisational design. Jobs are bundles of tasks, and AI will not affect every task in the same way. Some will be automated, some will be accelerated, and others will become more important because they depend on judgement, empathy, creativity or commercial context.

The right response is not to redraw the entire organisation around a prediction of what AI may eventually do. It is to build enough visibility to understand the work now. When roles are mapped into tasks, skills and capabilities, leaders can see where AI can absorb repeatable work, where human capacity can be redeployed and where new responsibilities are emerging.

In my experience, four areas determine whether an AI initiative can move beyond the pilot stage:

1. Operating model. Organisations need to decide who or what should perform each part of the work. That may require changing decision rights, roles and handovers rather than forcing AI into the existing structure.
2. Data. AI cannot work reliably when information is fragmented, outdated or poorly governed. Data quality needs a clear owner, particularly when an AI system is expected to act on that information rather than simply report it.
3. Process redesign. The strongest use cases usually begin with a small number of important workflows. Adding AI to one step of a badly designed process often allows the same problem to happen faster.
4. Change and adoption. People need to understand what is changing in their work, what remains their responsibility and how performance will be measured. That cannot be left to a launch campaign.

Stop managing headcount. Start managing capability.

Many workforce decisions still begin with the question, 'Who do we need to hire?' AI makes a different question more useful: 'What capability does the organisation need, and where can it come from?' The answer may involve an employee, a contractor, a shared service, software, an agent or a combination of them.

This does not reduce people to interchangeable resources. It creates a more precise view of where human contribution matters most. Repeatable steps can move to automation, while people take on work that requires judgement, relationship-building, problem-solving and accountability. The aim is not simply to remove roles. It is to redesign them around the work humans are best placed to do.

A capability view also improves redeployment. When the organisation understands the skills it already has, it can move people towards emerging work before recruiting externally. Workforce planning, learning and reward can then operate from a shared language rather than from job titles that may no longer reflect the work being done.

Adoption is a leadership job

The people side of AI is often assigned to HR or change teams after the technology decision has been made. By then, many of the choices that determine adoption are already fixed. Leaders have selected the tool, designed the process and set the measures without involving the people expected to use it.

Sustainable adoption begins earlier. Employees should be involved in identifying friction, redesigning workflows and testing how responsibility will work in practice. Line leaders must be accountable for whether new ways of working take hold, while executive leaders remain responsible for the strategic choices behind them.

Measurement must also move beyond logins and licences. The organisation should track whether AI is producing a business outcome and what happens to the time or capacity it releases. Unless that capacity is deliberately reinvested, the promised value may never reach the income statement or the customer.

The work starts after the technology arrives

AI does not fix a badly designed process. It does not resolve unclear ownership, clean up poor data or decide where human judgement still matters. Those questions remain, and in most organisations they are harder than choosing the technology.

This is the work I believe deserves far more attention. Which decisions can be automated? Where does a person still need to intervene? Who is accountable when an AI-supported process produces the wrong outcome? Which workflows need to be rebuilt rather than simply sped up?

None of this makes for an impressive demonstration, but it determines whether AI becomes useful or simply adds another layer to an already complicated organisation.

By 2026, buying AI will no longer count as an AI strategy. The harder and more important work is deciding which processes need to change, how roles and accountability must shift, and what the organisation will do with the capacity the technology releases. That is where measurable value will come from.