I’ve written before about change management being the unsexy side of digital transformation. But AI implementations have specific characteristics that require adapting traditional change management approaches.

What makes AI different? Several things.

The Uncertainty Problem

Traditional change management assumes you know what you’re implementing. A new ERP system has defined functionality. Training can cover specific features. Success criteria can be set in advance.

AI doesn’t work that way. Outputs are probabilistic, not deterministic. The same input can produce different outputs. Capability boundaries are fuzzy.

This creates challenges:

You can’t document every scenario. Traditional training covers “when X happens, do Y.” AI training must cover “when X happens, the AI might suggest A, B, or C, and here’s how to evaluate those suggestions.”

Success looks different. “The system works correctly” isn’t the right success criterion. “Users can effectively work with AI suggestions” is closer but harder to measure.

Edge cases are unavoidable. AI systems encounter situations they weren’t trained for. Users need judgment about when to trust AI and when to override it.

The Trust Calibration Problem

People either trust AI too much or too little. Both are problems.

Over-trust: Users accept AI outputs without critical evaluation. This leads to errors when AI is wrong, which it regularly is.

Under-trust: Users reject AI assistance entirely. They do everything manually despite having AI tools available. The investment doesn’t deliver value.

Calibrating trust appropriately – knowing when AI is likely reliable and when it’s not – is a skill that takes time to develop. Traditional change management doesn’t address this.

The Continuous Change Problem

Most technology implementations have a go-live date after which the system is stable. Users learn it once and apply that knowledge going forward.

AI systems change continuously:

• Models are updated with improved capabilities
• Outputs shift as underlying models change
• New features and interfaces are added regularly
• Performance characteristics evolve over time

Users must adapt continuously, not just once. This requires different support structures than traditional implementations.

Adapted Change Management Approaches

Given these differences, here’s how to adapt:

Reframe Training as Skill Development

Don’t train on AI features. Train on skills:

Critical evaluation. How to assess AI outputs for accuracy, completeness, and appropriateness. When to accept, modify, or reject suggestions.

Effective prompting. How to communicate with AI systems to get useful outputs. This is a skill that improves with practice.

Error recognition. Understanding common AI failure modes and how to spot them. Knowing when AI is likely wrong.

Appropriate application. Judgment about which tasks benefit from AI and which don’t. Not everything should involve AI.

These skills transfer across AI tools and survive model updates. Feature-specific training becomes obsolete quickly.

Build Trust Gradually

Start with low-stakes applications where users can verify AI outputs:

• Draft generation that humans review before sending
• Analysis suggestions that humans validate
• Automation of tasks where errors are easily caught

As users develop trust calibration, expand to higher-stakes applications. Don’t push AI into critical decisions before users have developed judgment.

Track trust calibration explicitly. If users accept AI outputs without checking, they’re over-trusting. If they never use AI features, they’re under-trusting. Both require intervention.

Design for Continuous Adaptation

Instead of change management projects, build change management capabilities:

Regular AI updates communication. When models change, communicate what’s different. Not technical details – practical implications for users.

Ongoing learning resources. Short, regular updates rather than comprehensive training. Microlearning that fits into work rather than separate sessions.

Community of practice. Users learning from each other about effective AI use. Internal sharing of tips, prompts, and approaches.

Feedback mechanisms. Ways for users to report AI issues and suggest improvements. Close the loop between user experience and system development.

Address Fear Directly

AI provokes anxiety that other technology doesn’t:

• “Will this replace my job?”
• “Will I be blamed for AI errors?”
• “Am I falling behind if I don’t use this?”

These fears are often unstated but affect adoption. Address them directly:

Job impact transparency. Be honest about how AI affects roles. If some roles will change or reduce, say so. If the intent is augmentation not replacement, make that clear with evidence.

Error responsibility clarity. Establish how errors involving AI will be handled. Users need to know they won’t be blamed for reasonable AI failures.

Psychological safety for learning. Create space for people to experiment with AI without fear of looking incompetent. Early adopters shouldn’t be penalised for visible mistakes.

The Support Structure

AI implementations need ongoing support that traditional implementations often don’t:

AI champions per team. Local experts who help colleagues with AI application. Not IT support – business users who understand the work.

Rapid response to issues. When AI produces problematic outputs, quick investigation and communication. Unaddressed issues erode trust.

Regular effectiveness reviews. Periodic assessment of whether AI is delivering value for different roles and tasks. Willingness to remove AI from situations where it’s not helping.

Evolution involvement. Users should have voice in how AI tools develop. Their experience should influence configuration and capability requests.

The Investment

This adapted approach costs more than traditional change management. Budget for:

• Skills-based training development
• Extended support period (ongoing, not time-limited)
• Community building and facilitation
• Regular communication production
• Champion network development and support

A reasonable budget is 25-30% of the AI implementation cost, compared to 15-20% for traditional technology change management.

The investment is justified because AI adoption failure is expensive. Tools that aren’t used don’t deliver value. Tools that are used poorly create problems. Getting adoption right is essential for AI ROI.

Final Thought

AI isn’t just another technology implementation. It requires users to develop new skills, calibrate trust appropriately, and adapt continuously. Traditional change management frameworks don’t address these requirements adequately.

The organisations succeeding with AI adoption are those that recognise these differences and adapt their change management accordingly. Those applying standard approaches are struggling with under-adoption and misuse.

Change management for AI is harder. It’s also essential. Plan accordingly.