AI Implementations Live or Die on Change Management

Implementing AI, in many cases, is asking a large group of people to change how they do their job. I’m sure you’ve heard all the hype about how “AI will change everything.” As for the CM part, what I usually see is a couple of emails and very average training. Which people often don’t attend if they can get away with it.

Many AI projects are failing - still 80% as of 2026.

The big irony here is that according to Prosci’s research, an initiative with good change management is ±6X more likely to hit its goals than one without.

1: Awareness can lull you into a false sense of security

Most companies are approaching AI in an earnest fashion and make strong inroads into staff awareness & readiness. They hold AI discussions, purchase compliant AI tool(s) and write an AI policy document alongside doing some vendor-sponsored training. They may even appoint an AI manager and integrate their policies into existing frameworks they use. From an administrative and governance standpoint, this is all very sensible.

It doesn’t get you very far though.

2: Desire is the most important part of any change management, adoption or enablement

Training can build awareness and knowledge, but it doesn't automatically build desire. Sometimes there are hidden de-motivators too. With AI, there’s quieter worry that they’re making themselves replaceable by using it.

One in-house counsel put it to me directly: “If you show something [AI] to the lawyer [that works well], they go ‘holy sh*t I can’t do it better, so … my role is defunct now?”

Implementations that build genuine desire target the parts of the job people already hate doing. Automate the painful, unglamorous task and adoption follows. Automate the thing someone's identity is built around and you'll get quiet resistance.

3: Knowledge training is being delivered, but it's the wrong knowledge

I’ve talked about AI training being lacking before. When I've dug into what's covered, at worst its very general and covers basic compliance and “check your work,” at best there’s a little stuff on what a prompt is, how to verify and common risks. One partner-track associate said to me: lawyers are taught IRAC in law school, structured reasoning, but almost nobody teaches equivalent structured prompting. The knowledge gap isn't "does AI exist?" it's "how do I use it well enough that I could defend the output under questioning?"

4: Ability is another common failure point

This is the gap between "we did the training" and "people actually do it differently."

A newly trained associate at an insurance firm knows the correct prompting method but hasn't yet built the instinct for where it tends to fail, so they accept an AI-drafted clause comparison without running the edge cases a senior might spot. Because the change management for the AI implementation was lacking, this newbie wasn’t differentiated from the others.

5: Reinforcement is almost universally absent

I’ve seen plenty of tech companies reinforcing AI usage by tying them to KPIs and so on, but very few non-tech firms doing this.

In fact one law firm CEO suggested that the bill-by-hours model creates a perverse incentive NOT to be too efficient. Reinforcement isn’t a training problem. It's an incentive design problem, and most governance frameworks don't touch incentives at all.

What does good look like?

• Awareness that goes beyond buying the tool and running a vendor module. Great training with lots of applied examples - both good and bad - goes a long way towards achieving this.
• Building genuine desire through a focus on things that genuinely benefit employees. For example, a law firm that pilots AI completing first-pass document review (a task everyone hates) would be a relief rather than a threat.
• Knowledge that's role specific. If junior staff at a law firm get training on verification habits, partners get guidance on risk exposure and what to sign off on. Its worth noting that this doesn’t rule out generic training altogether - just narrows things a bit.
• Ability built through practice and experience, not left to accumulate by accident. This will differ for every company, but even simple check-ins, assessments and follow-up training help ensure it happens.
• Reinforcement that includes real incentives. Example: time spent checking AI output gets logged as billable or otherwise credited, so the incentive to verify properly doesn't sit at odds with the incentive to bill hours.

—

Monboard helps firms with the messy part of AI rollouts: change management. Drop us a ping if you want to know more.