We are doing it all wrong… again.
I'm getting serious déjà vu at the moment as I watch organizations attempt to rollout out AI, and to be honest, it is frustrating the shit out of me.
Back in 2014, I wrote a book called Digital Adaptation where I talked about the shortcomings in how organizations were approaching digital transformation. This was the time when organizations were finally waking up to the fact that their website wasn't just another brochure and that digital had become essential to all elements of their operation.
Digital transformation was never really a technology problem. The technology was the easy part, and most of it was already sitting there waiting to be used. The hard part was that users had changed how they behaved, and organizations were being dragged along behind them, trying to keep up with expectations set by Amazon and Google rather than by anyone in the boardroom. Catching up meant changing culture, processes, and the way people worked, which turns out to be a lot harder than buying a new content management system.
AI adoption is the same problem all over again. Yes, there is a shiny new technology involved. But, the thing that will decide whether it works in your organization is people. Not just your users, but your staff too.
We appointed a chief electricity officer once, too
In the book I compared digital teams to the "chief electricity officer" that some companies appointed when electricity first arrived. It made sense for about five minutes, until everyone realized electricity was something every part of the business used and putting one person in charge of it was a bit daft.
Today organizations are busy appointing Chief AI Officers and setting up central AI committees. I understand the impulse. It feels responsible, and it gives the board someone to point at when things go wrong. But, AI is going to end up woven into almost every job in the building, and no central team understands the daily work of finance, customer service, and the warehouse well enough to decide how AI should fit into each of them.
IT built it, so staff ignored it
We saw what happens when that distance exists with digital. IT designed systems for people they had never met, and when those systems did not work, staff routed around them using their own off-the-books workarounds. The same thing is happening with AI, except the workaround is now a personal ChatGPT account on a phone under the desk.
The CEO has had a revelation. Everyone else hasn't.
However, not everything is the same. When I wrote the book, it was really aimed at helping digital teams convince management that change was required. Now the direction has flipped. Often it is the CEO who has become the convert, and the rest of the organization is lagging behind.
I spoke to the CEO of a cargo airline recently who had binned a reporting project that had been "in production" forever without ever actually being finished. That was because, within days of trying AI, the CEO had working dashboards built with Claude. That is a nice illustration of another point from the book, by the way. Big projects that try to specify everything up front rarely finish, and building, testing, and iterating beats them almost every time.
Understandably, that CEO is now evangelical. The problem is that a dashboard for the CEO means nothing to the person processing shipping paperwork. Everyone needs their own revelation about their own role, and you cannot hand somebody a revelation in an all-staff email.
Why your staff are right to be nervous
This is where AI differs from digital, and where I see most rollouts fall apart, because your staff are nervous about AI, and understandably so.
Is this about replacing me?
AI makes every individual employee wonder whether the CEO's enthusiasm is about replacing them. When leadership announces an AI initiative, a lot of staff hear "we are working out how many of you we need." Now, you can reassure them as much as you want, but they have read the same headlines you have.
What's in it for me?
Then there is the question of what is in it for them. If staff are still going to be sitting at their desks for eight hours but are now expected to produce more, the honest answer is nothing. They take on the learning curve and the risk, and the organization keeps the benefit. I would be reluctant too.
Decades of habit don't vanish overnight
We also underestimate what we are asking when we tell people to give up ways of working they have built over decades. That procedural knowledge is a big part of how people see themselves as good at their jobs, so being told a tool can now do it is hardly a neutral message.
They are already using it anyway
And on top of all that, with some people, adoption is already happening without management permission. Plenty of staff are using AI right now to make their lives easier. With these people, the job is less about getting approval and more about catching up with what they already do and making it safe.
So what do you actually do about it?Start with the people who already care
First, start with the enthusiasts rather than the resisters. Every organization has a handful of people who are already experimenting. Bring them together, give them space to share what is working, and let their combined enthusiasm do the heavy lifting.
In the book I described an internal conference where staff showed colleagues what they had built, and it did more for adoption than any memo from the top. That works just as well for AI.
Make it safe to mess up
Second, make it safe to experiment. That was a big theme in the book, and it matters even more now. Lots of people are afraid to try something different in case they get it wrong, and if your organization has a culture of fear, no amount of memos or shiny new tools will persuade them to play. Playing with new tools is how we learn, so people need to know that a failed experiment will not count against them.
Give people time to play
Feeling safe is only half of it, though. People also need the time and space to experiment. Most of the staff I meet are so stretched that trying something new feels like a luxury they cannot afford. So, if you want people to explore AI, you have to give them the hours to do it rather than expecting it to happen somewhere between back-to-back meetings.
Aim it at the boring bits
Third, frame AI around the parts of the job people hate. Nobody is emotionally attached to reformatting reports or chasing timesheets. At that same airline, a technical writer was convinced AI would replace her, and she became one of its biggest advocates the moment it made the tedious parts of her work faster.
Skip the one-size-fits-all training
Fourth, be wary of generic training. I have seen organizations run one-size-fits-all AI sessions aimed at their most resistant staff, and the result was exactly what you would expect. People typed in vague requests, got vague answers, and left convinced AI was rubbish. Garbage in, garbage out.
Interfaces matter more than you think
Fifth, pay attention to the user interface. My wife ignored AI entirely until she used Grokbot, which presented it as a team of helpers. Same technology, completely different reaction, which is a reminder that adoption is a design problem as much as anything.
Sort out governance (without a committee)
Sixth, sort out governance rather than forming another committee. Lock down only the data that genuinely needs protecting, with permissions set at the data layer, and then trust your staff to do the right thing with everything above that. Most people want to do a good job and have no interest in leaking customer data. What they need is a lightweight governance framework that sets out a few clear principles and who is responsible for what, and the responsibility matrix from the book applies almost unchanged here.
Don't lock it down until it's useless
Finally, do not lock things down so tightly that AI becomes useless. Yes, sensitive information needs protecting. But, I keep seeing organizations where AI is so restricted that nobody can do anything of real value with it. I am working with a UX team at the moment who want to use AI to build a user research repository, and their Copilot license is so restrictive it will not let them. They are not allowed to bring in any other AI tools without going through a complicated committee process, so the whole idea is stuck, despite management telling them they should embrace AI.
Over-locked AI gives you roughly the same result as no AI, except you are still paying for the licenses.
So, lock down the data that genuinely needs it, then make it easy for people to get on with everything else, because overdo the governance and you kill creativity and adoption along with the risk.
Make sure UX is in the room
If you work in UX, you have an incredibly important role to fulfill here. I am not suggesting you should be the one leading AI adoption. There are HR issues, technology issues, and strategy questions that sit firmly with senior management, and none of those are ours to own. But, UX should very much be in the room and part of the conversation. After all, UX was born out of human-computer interaction, and AI is human-computer interaction in its purest form. We are now having actual conversations with machines, which is about as close to our core discipline as it gets. We also understand behavior, motivation, emotion, and resistance to change better than most people in an organization, and those are exactly the things that will decide whether AI adoption works. In the book I asked web teams to lead digital change. This time I am asking UX people to make sure they have a voice in it, because I am not convinced anyone else will speak up for how staff actually feel about all this.
I am curious whether you are seeing the same patterns where you work. If enough of you are, this might turn into a series, and maybe one day another book, assuming I can stay focused long enough to finish one.
I'm thinking about running a deeper workshop on this in January, for anyone who's been handed the job of getting their organization on board with AI. If that sounds useful, just reply and I'll let you know if it happens.
It agrees with everything you say, it can't read the politics in your organization, and it won't warn you before you make a mess of that stakeholder conversation. I will.
Six coaching packages, newsletter readers only. 4 hours of one-to-one coaching for £660, booked before the end of October. When they're gone, they're gone.
Reply to this email to grab one.
