Change Management For AI
Traditional Workday change management assumes a process change. AI introduces trust, role, and judgement changes. Old playbooks under-serve this.
Welcome
You already know how to run a Workday change programme. You know the comms cadence, the champion network, the training waterfall, the hypercare plan. None of that goes away — but if you lift it straight onto an AI rollout, it will miss the things that actually make or break adoption.
This module is about the delta. What's different when the change isn't a new screen or a new business process, but a new relationship between a person and a system that makes suggestions they can accept, edit, or override? That's a harder change, and it rewards a different approach.
Learning objectives
Why traditional change management under-serves AI
A standard Workday change plan answers: what's changing, who's affected, how do we train them, how do we support them through go-live? That framing assumes the change is a process change — the task stays, but the clicks, fields or approvers shift.
An AI rollout isn't that. The task might not change at all. What changes is the relationship between the person doing the task and the system. The agent suggests something. The person has to decide whether to accept it, edit it, or ignore it. That decision sits on top of every task, every day, forever.
If your change plan treats the agent as a new button, you'll under-train judgement, under-communicate uncertainty, and over-promise accuracy. People will either blindly accept the agent's output and inherit its mistakes, or blankly ignore it and you'll never realise the value.
Traditional CM optimises for "can they do the new process correctly?". AI CM optimises for "can they work with the agent well — knowing when to trust it and when not to?". Those are different competencies and they need different enablement.
The three change vectors
Every AI rollout asks three questions of every affected user. If your change plan doesn't answer them explicitly, the user will answer them on their own — usually badly.
Trust
Will it be right? What happens when it's wrong? How do I tell the difference?
Role
What does my job become if the agent does part of it? Am I still needed? Am I still valued?
Judgement
When do I accept the suggestion, when do I edit it, and when do I override it entirely?
Trust is earned through transparency — show the data the agent used, show the confidence, show the audit trail. Role is navigated through honesty — say clearly what the agent does, what it doesn't, and what the human is still accountable for. Judgement is built through practice — give users low-stakes reps before the system is load-bearing.
A good AI change plan works all three in parallel. A weak plan works only trust, assumes role sorts itself out, and leaves judgement to chance.
Stakeholder map
Every stakeholder group asks a different question. Same rollout, six conversations. Get the messaging wrong and each group blocks a different bit of the programme.
End users
Question: Is this here to help me or replace me? Message: Here's the bit of your day this removes, here's what you still own, here's how it makes you better at the part that matters.
Managers
Question: How do I coach my team on something I don't fully understand myself? Message: We'll enable you first. You'll know the agent's strengths and limits before your team does.
Compliance
Question: Can we evidence the control environment? Message: The agent inherits tenant controls, every action is audited, and the human approval step stays where it already is.
Data privacy
Question: Where does personal data go and on what legal basis? Message: Data stays inside Workday's trust boundary, DPIA updated, no new processors introduced.
Risk
Question: What's the failure mode and the blast radius? Message: We've defined what the agent can and can't do, where humans sign off, and how we roll back if needed.
Exec sponsor
Question: Is this going to deliver value, and can I defend the investment? Message: Measurable outcomes tied to pilot cohorts, adoption metrics beyond activation, and a credible path to scale.
The Kainos AI Navigator frames these conversations around Use, Adapt, or Build. Each stakeholder group needs to understand which posture you're taking and why — because their concerns change shape depending on the answer.
Communication approach
AI rollouts fail quietly. People don't revolt — they just don't use it, or they use it badly. Communication is the single biggest lever you have, and the one most commonly under-played.
Cadence
More frequent than a traditional rollout, for longer. Go-live isn't the end of the comms plan — it's the middle. Expect to keep communicating for two to three release cycles after the agent is live, as behaviour settles and edge cases surface.
Channels
Match the channel to the stakeholder. End users want short, in-product and Teams-native nudges. Managers want a written playbook they can reference in a one-to-one. Compliance and risk want documents they can archive. The exec sponsor wants a dashboard and a story.
Tone
Honest about uncertainty, specific about boundaries. Don't say "the agent is always right" — it isn't, and saying so destroys trust the first time it's wrong. Don't say "the agent might be wrong sometimes" either — that's too vague to act on. Say "the agent is strong at X, weaker at Y, and here's how you'll know the difference".
Confident, direct, human. This is change about people, not systems. If your comms read like a release note, rewrite them.
Resistance patterns & responses
Three patterns show up in almost every AI rollout. Name them early, address them directly, and they stop being resistance — they become conversations you can have.
Pattern 1 — Fear of replacement
What it sounds like: "If this agent drafts the contract, why do you need me?" or, quieter, people just stop using the tool because using it feels like voting against themselves.
Why it happens: The narrative around AI in the market is often replacement-first. Unless you counter it explicitly, users assume the worst.
Response: Be specific about what the agent does and doesn't do. Be specific about what the person is still accountable for — and make it the harder, more valuable bit. Don't hand-wave. People see through "AI will free you up for higher-value work" unless you say what the higher-value work actually is.
Pattern 2 — Over-reliance
What it sounds like: Silence. People accept every agent suggestion without review because it's easier, and because the agent is right enough of the time that they stop checking.
Why it happens: Humans automate cognitive effort wherever they can. If the agent is 90% right, the 10% will be missed unless the process forces a check.
Response: Build the check into the workflow, not the training. Show the confidence or the source data inline. Sample and review agent outputs as a standing operational practice, not a one-off audit. Reward the edit, not just the acceptance.
Pattern 3 — Scepticism from past AI failures
What it sounds like: "We tried an AI thing two years ago. It didn't work. Why will this?"
Why it happens: Most organisations have an AI scar. Often it was a bolt-on pilot that never reached production, or a vendor demo that didn't survive real data.
Response: Don't dismiss the history — inherit it. Ask what went wrong last time, and show specifically how this rollout is different. Usually the difference is that the agent runs inside Workday, inside the tenant, on data users already trust — not beside it. That's a credible answer.
Pilot to scale as a change strategy
Pilots are usually framed as a technical de-risking exercise — does the agent work, does the integration hold, is the accuracy acceptable. That's necessary but not sufficient. The bigger value of a pilot is as a change-management rehearsal.
What you're actually piloting
- The messaging. Does the story land with end users and managers, or do you need to rewrite it before you go wider?
- The training. Are people learning to work with the agent, or just learning where the button is?
- The support model. When the agent is wrong, where does the user go? Does that path work?
- The measurement. Are you picking up adoption, trust, and judgement signals — or just activation?
Scaling without losing the change work
The common failure is to nail the pilot and then scale by switching on the agent for everyone else. The pilot cohort had intense change support; cohort two gets a training email. Adoption collapses and the programme gets blamed on the technology.
Scale the change envelope, not just the toggle. Every new cohort gets the same vector-by-vector treatment — trust, role, judgement — even if the agent itself has been in production for six months. Use the Kainos AI Navigator's Use, Adapt, or Build lens to keep each cohort's change plan proportionate to what's actually changing for them.
What success looks like
The cleanest signal that AI change management has worked isn't usage numbers. It's a sentence. Ask a random user:
"When do you trust the agent, and when don't you?"
If they can answer that in their own words — specifically, with examples — the change has landed. They have a working model of the agent's strengths and limits, they know where their judgement still matters, and they're using it accordingly.
If they can't — if the answer is "I always use it" or "I never use it" or "I don't really know" — something in the change plan has missed. Either trust wasn't built deliberately, role wasn't addressed, or judgement wasn't practised. Go back and find which.
You haven't finished the change until users can articulate, unprompted, when to trust the agent and when not to. Everything else is a leading indicator of that outcome.
Check your understanding
Three questions. Each explains why every answer is right or wrong — the reasoning matters more than the score.
Next steps
- Pair this with Module 10 — Adoption, not activation: the measurement side of the same problem. What to track, and why activation is the weakest signal.
- Operational playbook — Module 8: the adoption playbook translates the principles here into specific cadences, artefacts, and checkpoints for a live engagement.
- Customer-facing positioning — Module 4: how to frame the Kainos AI Navigator and the Use, Adapt, or Build posture so that stakeholder conversations start in the right place.