AI change management can build on much of what your organization already knows. A clear purpose, useful employee input, practical learning, accountable sponsors and support after launch still belong in the plan. The question is which assumptions underneath those practices need another look.
For generative AI used to draft or analyze work, pay particular attention to how people judge an output, what happens when the tool or its source material changes, and who owns the work after the launch team leaves. Those questions help you adapt an existing change program without treating every previous lesson as obsolete.
Keep the practices that help people do the work
Start with the parts of a previous program that solved a recognizable problem. A sponsor who settled conflicting priorities, a colleague who explained an unfamiliar task, or a feedback channel that actually changed the rollout can all be useful again.
There is evidence for continuity, with limits. A 2023 OECD working paper found that training and worker consultation were associated with more positive reported AI outcomes. Its surveys covered finance and manufacturing in seven countries, with fieldwork in early 2022. These were associations, not proof that the practices caused the outcomes, and the study predates widespread workplace use of today's generative assistants.
Use the following questions to review your existing plan. They are practical prompts for a change lead, rather than a validated adoption model.
| Practice to retain | What it should make possible | What to revisit for generative AI |
|---|---|---|
A clear purpose | People know which work should improve. | Define an acceptable finished result, including checking and correction. |
An accountable sponsor | Someone can settle priorities and unresolved decisions. | Identify who can change scope, pause a use or fund extra support. |
Employee participation | People doing the work can influence its design. | Ask them to bring difficult inputs and failures, as well as successful examples. |
Role-specific practice | People rehearse work they will actually repeat. | Include plausible errors and situations where the tool should not be used. |
Support after launch | Help remains available when ordinary work gets difficult. | Assign maintenance of examples, checks and escalation routes. |
A September 2026 account by change lead Francesco Corda describes a European fintech program combining local AI champions with a small central capability. He says the sponsor needed to provide a mandate, timely decisions and help across organizational boundaries. That is a concrete use for familiar sponsorship work, although one practitioner's account cannot establish a generally effective model.
The discussion also exposes unfinished work. In replies to a challenge about organizing around AI, Corda distinguishes the program's delivery arrangements from the wider organization and acknowledges that transferring the capability into everyday operations needs intentional attention. An effective launch structure does not settle that handoff for you.
Adapt practice to include an unreliable answer
Suppose an apparel wholesaler is introducing AI-assisted product descriptions. A catalogue editor starts with the supplier's approved fabric, sizing and care information, then prepares copy for a buyer to review before publication.
The team can reuse its product-information training, editorial standards and review meeting. The practice exercise needs an additional test: a polished draft that claims a sweater is machine washable when the supplier's instructions say hand wash.
Ask the editor to show where each care claim came from, correct the unsupported statement and explain what to do if the supplier's information is missing. The exercise reveals more than whether the person can generate a paragraph. It also gives the team a useful example to keep beside its normal writing guidance.
NIST's Generative AI Profile describes confidently false content as a risk and recommends reviewing output sources and citations. Applying that guidance here means checking care claims against the approved supplier information. A fabric swatch or a convincing sentence cannot establish washing instructions.
In a public discussion about AI adoption, a Reddit user who described themselves as a UK business change consultant reported seeing expectations of near-instant work alongside people skipping output review. Their account is not a measure of how common this is. It does illustrate why a training plan should include the review work that a successful demonstration can hide.
A useful practice session includes an acceptable result, a mistake worth catching and a way to resolve uncertainty. For this catalogue team, a draft with an unverified care claim stays out of publication until the information is resolved.
Keep monitoring and update what you look for
Earlier technology programs also needed maintenance, feedback and monitoring. NIST makes that continuity explicit in its 2026 discussion of monitoring deployed AI, while identifying continuing challenges around AI behavior and human interaction. It does not offer a universal review interval.
For the catalogue team, revisit the practice example when the approved tool changes, the supplier-information format changes, or reviewers find a new kind of unsupported claim. A named editor can maintain the examples, while the tool owner investigates behavior that ordinary editing cannot fix.
Choose checks that reflect the work's consequences and volume. For instance, the buyer could record which factual claims required correction during the existing approval step. If correction effort rises, the team has a reason to investigate the inputs, instructions or tool before expanding use. A login count would not reveal the same problem.
Keep the review proportionate. A low-volume drafting aid and a system that publishes descriptions automatically need different controls. Changing the level of automation is itself a reason to revisit responsibilities and approval boundaries.
Investigate friction before prescribing more training
Reusing a change practice does not mean reusing the same explanation for every difficulty. If the catalogue editor keeps returning to the old process, ask to see a recent attempt.
- Missing supplier information needs an information owner to resolve it.
- An unclear checking method may need a demonstration and another practice attempt.
- Extra correction work calls for comparing the complete task with the existing process.
- No time to practise requires a workload decision from the manager.
These conditions can occur together. Use the guide to distinguishing inability, reluctance and missing opportunity when the explanation is unclear. Treat employee feedback as information about the design of the work before deciding that encouragement is the missing ingredient.
Review one existing plan before creating another
Take a rollout plan your organization already uses and work through one real task with its owner. Keep the review small enough to change a decision.
- Name the useful practice. For example, retain the catalogue team's buyer approval meeting because it already resolves product-copy questions.
- Identify the assumption to retest. Check whether reviewers can trace AI-generated care claims to supplier information within the time available.
- Agree the adjustment. Add a source check and a route for missing information, then try them on an approved sample of descriptions.
- Assign the continuing owner. Decide who maintains the guidance and who can pause or change the AI use if the check is insufficient.
Record what you will observe at the next review, including correction work and unresolved questions. If the old practice already handles the need well, retain it. If the proposed AI use makes the complete task worse, changing or stopping that use is a legitimate result.
The people and change guide helps turn what you learn into appropriate support. For work across several teams, define the adoption manager's responsibilities alongside the decisions that remain with operational and technical owners.
Questions about adapting technology change programs
Does AI adoption require a completely new change framework?
A generative AI rollout does not automatically require replacing your existing change framework. Start by checking whether it gives you clear purpose, employee input, practice, decision owners and continuing support. Then examine assumptions about output quality and changes after launch. The comparison of practices to retain offers questions for that review, without claiming one framework works everywhere.
Can we reuse training from an earlier technology rollout?
You can reuse relevant task knowledge, learning arrangements and support channels, but update the exercises for the approved AI use. For a catalogue editor, that means practising how to find and correct an unsupported product claim, not merely learning where to type a prompt. Check the current inputs, output standards and escalation route before reusing an old exercise. The product-description example shows what that adjustment can look like.
When can the AI change program finish?
A temporary AI change program can end when its agreed scope is complete and ongoing work has a supported owner. Confirm who maintains guidance, reviews problems, approves changes and helps new users, with enough time and authority to do so. Closing the launch project should leave those responsibilities visible in normal operations. Record unresolved issues and handoff conditions before closing the program.
