No one's got it worked out
When it comes to AI adoption in enterprise, what you see in the media or at events is usually the glossy stuff or just plain scaremongering.
When it comes to AI adoption in enterprise, what you see in the media or at events is usually the glossy stuff or just plain scaremongering.
My conversations behind closed doors are materially different. And the people I’m talking to are in the thick of it, doing the real work.
These are customers, partners, and friends. Mostly in large, complex organisations, including government and enterprises.
What I’m hearing is that no one’s got it completely worked out. And what’s interesting is that most people feel like they’re behind, but usually they’re really not.
I’ve noticed a few patterns that keep coming up.
1. The pressure is personal
Yes, there’s pressure from the top. Boards and CEOs want to know what the organisation is doing about AI. And when they feel like it’s moving too slowly or it’s not enough they’re forcing their agenda.
Often IT departments are feeling the pinch because they’re already under-resourced and have a full work program and now they have to shoehorn something in that is not well understood - especially the cybersecurity and data risks without proper controls in place.
But a lot of the pressure people are under, they’re putting on themselves. Self-inflicted.
I’m talking to a lot of senior people who kept AI at arm’s length for a while. “I’ll get to it.” They had real work to do, existing fires to put out, and major changes to implement. Now it’s everywhere, and there’s this feeling that it’s passing them by and they have to get on top of it. Fast.
And the people who’ve been handed AI as a job are squeezed. They feel like they don’t have authority or the resources they need, and they’re being dragged from pillar to post with unrealistic expectations put on them.
What that produces is a lot of reacting.
And that’s usually where “just roll it out” comes from. Someone prescribes a solution instead of working on an outcome.
They just want the dashboard to go green.
2. No one owns it
I’ve walked the halls of dozens of organisations recently and asked people point blank, “Who owns AI here?” The answers I heard were usually IT, or a working group that meets when it can. Sometimes it’s an executive or the CEO, but they’re not actually involved other than being a sponsor.
So there’s no one to make or challenge decisions. The main issue with this is that there’s no one to say no.
The licences (Copilot, ChatGPT, Claude usually) paint the picture. A department has some budget, so it buys a few. Then another one does. Multiply that across every business unit and you’ve got hundreds of people on it, and no one owns the total.
And once a team has paid for its licences or switched on an AI feature somewhere, good luck taking it back. So the role of IT becomes enablement, not policing.
Data’s the same. The business needs to own its data, be responsible for it and make decisions on it, rather than handing it to IT as an IT problem.
When it all defaults to IT, you’ve got a technology function defending a business decision it was never given the authority to make.
3. Time saved isn’t value captured
I was in a meeting with an executive team and their private equity overlords recently and AI came up.
This was not a technical group and the meeting was about value creation. The private equity guy said something so simple but so important:
“Whatever you do with AI, it has to tie back to some sort of commercial outcome. It’s not about people finishing work half an hour earlier if there’s no real difference to the business.”
Ask staff what they’re doing with the time AI saves them and you get honest answers. In some staff survey results I saw recently, people were using it for important personal things, like picking the kids up from school or fitting in a catch-up.
That is actually valuable and if there’s strategy behind it, it can be tied to a commercial outcome. Just think about what that could mean for attrition and retention metrics.
But that sort of value is rarely captured because it wasn’t considered in the first place.
Generally speaking, if you go wide and shallow with AI use cases, everyone’s just doing things like meeting summarisation or drafting email and that’s about it. There’s no obvious, measurable value.
Going deep is better, but it’s not the whole answer either. A very deep and narrow use case is good because it builds a muscle. But if you stop there, it atrophies over time. Point solution, point solution, point solution.
It’s a bit of a trap, too. You can do more, so you do more, but it doesn’t mean you should do more.
For a lot of organisations, the play isn’t efficiency. That can be a race to the bottom. It’s revenue growth, new revenue streams, new value creation for their customers.
So what are the good ones doing?
If you look back over those three patterns, they’re really the same problem showing up in different places. A decision that no one has made.
No one decided what AI was actually going to be used for, so whatever is convenient (Copilot licences or something a vendor offers) gets rolled out to make the dashboard go green.
No one decided how calls about AI get made, so AI tools sprawl haphazardly across the business.
And no one decided where the saved time was meant to go, so it ends up wherever people choose to put it. And that could be really valuable, but it’s not deliberate and no one is measuring it.
The organisations getting somewhere have made those calls, and made them early.
Someone in the business owns it, and their first job is helping set out what the organisation is actually trying to achieve with AI and what matters most.
The outcomes they’re after, the risks they’ll accept, and where the time AI frees up is meant to go. It doesn’t need to be complicated, it just needs to exist.
Once it does, decisions have something to stand on. Whether it’s a new tool, another round of licences or a pilot that isn’t going anywhere, the answer comes from what the organisation has agreed matters, rather than whoever’s loudest in the room.
Sometimes that answer is yes, sometimes it’s no, and either way people can see why. It speeds things up too, because decisions stop sitting in limbo waiting for a working group that meets when it can.
Then they pick something they’re already good at and go deep on it before moving to the next thing. A little bit over a long period of time, and that’s where the compounding comes from.
None of it needs a big program or a 40-page strategy. It mostly needs someone senior to sit down and make a few calls they’ve been putting off.
If you want somewhere to start, have a first pass at writing down some basic criteria to assess AI requests. What’s it for? What data does it touch? Could it affect someone’s job or the outcome they get? Share it with your leadership group. What you want is a reaction, which drives thinking, which drives discussion, and now you’re on your way…
Want a head start on those three questions?
They come from the screening template in my AI Governance Toolkit, one of eight templates I use with my own customers to get AI governance up and running. They’re in Word and Excel, and none is longer than five pages. The others cover setting up a working group, finding out what AI is already in use, what happens when it goes wrong, and a policy people will actually remember.
It also includes a Kickstart session: an hour with me where I’ll explain the templates and we’ll work out exactly where you should start. $695 + GST.