AI Is Coming to Your Project — And Your Team Isn't Ready

Shane Tyrrell | Head of Advisory and Capability Enablement, Prosci ANZ

A finance team has just had an AI tool deployed into their month-end close process. The tool reconciles transactions in minutes instead of days. The business case showed a 70% reduction in manual reconciliation time. Leadership approved the rollout in a single meeting — a no-brainer.

Three months in, half the team is still running the manual process alongside the AI output, checking the "machine's" work line by line because nobody trusts it yet. The other half stopped using it altogether after the first error went unexplained. The tool works. The team doesn't believe it does.

This is not a technology failure. The model performed exactly as designed. What failed was everything around it:

No one explained why the change was happening. No one addressed what it meant for the people whose judgment was being partially replaced by a machine. No one built the trust required for the team to actually let go of the old ways of working.

If you're running a project right now that touches AI in any way, this scenario should make you uncomfortable — because the data says it's closer to the norm than the exception.

The Problem Was Never the Model

Prosci surveyed thousands of professionals on what actually derails AI implementations. User proficiency — the human side of learning the tool, building trust in it, and adapting old habits and ways of working — accounts for roughly 38% of reported difficulties. Purely technical issues account for about 16%.

With user proficiency issues With purely technical issues 38% of reported AI implementation difficulties 16% of reported AI implementation difficulties

Read that gap again. The thing most likely to break your AI project is more than twice as likely to be a people problem than a technology problem. Yet most AI project plans still allocate the majority of their budget, time and attention to the technology.

Harvard Business Review reached the same conclusion in its November 2025 analysis of organisational barriers to AI adoption. Treat AI adoption as a change problem first and a technology problem second. Most organisations are still doing the opposite.

The Trust Gap Nobody Is Measuring

Prosci asked organisations to rate leadership support for AI initiatives on a scale from -2 to +2. Organisations with smooth AI rollouts scored their support at +1.65. Organisations that were struggling scored it at -1.50. That is a spread of more than 3 points on a four point scale. This is not a small gap. It is the difference between adoption and abandonment.

There is a second, quieter gap underneath that one. On the same scale, frontline workers' trust of AI was +0.33. Executives' trust was +1.09. The people closest to the daily risk of a bad AI decision are the most sceptical. The people approving the budget are the most confident. That mismatch is exactly where projects go wrong, because the rollout plan gets built by the people with the least exposure to what could go wrong.

Go back to the finance team. Half of them kept running the manual process in parallel. That is not stubbornness. That is a rational response from people who were never given a reason to trust a system they did not help design and were not trained to verify.

Training Is Not the Same as Readiness

Most organisations think they have solved this by running an AI training session before go-live. The data says that training is not closing the gap. Docebo's 2026 enterprise learning research found that 85% of employees say the AI training they received does not help them apply it to their actual job. One in five received no AI training at all.

A training session is a Knowledge activity in ADKAR terms. It tells people how to use the tool. It does nothing for Awareness of why the change is happening, Desire to actually use it over the old way or Ability to apply it under real conditions with real consequences. Skip these steps and the training session becomes a tick box exercise on a project plan and not a behaviour change on the frontline.

This is precisely what happened with the finance team. They sat through a 90 minute walkthrough of the new tool. Nobody addressed what it meant for their role, their judgment or their job security. The training covered the buttons. It never touched the fear.

Three Things to Do Before Your Next AI Rollout

First, find out what the team actually fears before you write the training plan. Run structured conversations, not a survey nobody reads and ask people directly what they think this AI tool means for their role. You will get answers like job security, loss of expertise or being blamed for a mistake the machine made. Address those answers by name in your communications. If you skip this step you are training people on software while leaving the real objection completely unmanaged.

Second, put your sponsor in front of the team before go-live and not just in the announcement email. The leadership support gap between smooth and struggling AI rollouts is not closed by a memo from someone the team has never spoken to. It is closed by a sponsor who shows up, explains the decision in person and stays visible through those bumpy first weeks. If your sponsor's involvement ends at the approval signature expect the frontline trust score to sit near zero.

Third, design for the parallel run period instead of pretending it will not happen. People will check the machine's work until they trust it. That is not a failure state, it is a predictable phase of adoption. Build a defined verification window into your project plan, give it an end date and use that period to show people exactly where the tool got it right. Trust gets built through evidence, not instruction.

What This Means for You as a PM

AI projects are showing up on more of your project plans this year than last year and this trend is not slowing down. Most of you are being asked to deliver these projects on the same governance frameworks you used for ERP rollouts and system migrations five years ago.

Those frameworks were not built for a change that touches trust, role identity and judgment the way AI does. A new finance system asks people to click different buttons. An AI tool asks people to hand over part of their professional judgement to something they did not choose and may not understand. That is a fundamentally different category of change and it requires a fundamentally different level of people investment.

You do not need to become a change practitioner overnight. You do however need to stop treating the people side of an AI rollout as a comms task that happens after the technical plan is locked. Build it in from week one, resource it properly and measure it with the same discipline you apply to your technical milestones.

Your next project plan has a line item for an AI tool. Does it have one for the people who have to trust it?