Posted by |
Getting your Trinity Audio player ready...
|
One can barely have a conversation without the mention of AI. I am hearing it from clients, from educators, from friends, from kids. FOMO is permeating the system and thriving on curiosity. But it is also overwhelming people as they are trying to figure out what it means to them. Some are experimenting at work without clear guidelines or impact considerations. Some are burying their head in the sand. How will AI adoption expand—and what does a structured AI adoption framework actually look like in practice?
Where Are We Now?
In the recent State of AI in Enterprise Report by Deloitte for 2026, some numbers are difficult to get my head around:
- 25% of organisations have moved 40% or more of their AI experiments into production
- 34% are pursuing deep transformation, with the remaining focused on surface-level efficiency
- 66% are achieving efficiency and productivity gains, but only 20% have increased revenues through AI
- 84% have not redesigned jobs for AI, and only 27% of firms have mature governance
I don’t know about you, but for me, these numbers point to a much lower adoption than the one we seem to hear about whenever we open news, business publications or LinkedIn commentary. And whilst I do not think this is because of hype, I do find more and more questioning when discussing with C-level executives. They are not questioning the existence, the transformative power or the need anymore. They are questioning how to start and how to impact the business.
How do I Start?
One of the questions I get more often is
“There is just too much going on that I don’t know how to start!”
By start, it does not mean companies are at zero (though the data suggests a good part of them are). Often, it means the step after getting basic licenses, putting employees on mandatory training, and having some basic content created in ChatGPT. These baby steps get done out of experimentation and curiosity.
The real question for business leaders is how to take it from experimentation to a formal pilot and scaling. There is a rush to provide licenses, to tell everyone to get training and start with the program. But the question managers are asking is then – what program?
When you look at the stats and see that only 60% of workers with access to AI are using it in their daily workflows, you can’t help but wonder. Why are people who are given sports cars choosing to walk to work? It shows that giving workers an AI license is not a strategy per se. Fluency and job redesign are not keeping up with the software provisioning.
Once the fun dating stage is done, how do people move into a serious relationship?
From Experimentation to Scale: An AI Operating System
What I see right now is not a lack of effort or even a lack of willingness. It is a lack of structure. And without structure, all this effort gets stuck in experimentation loops, duplicated efforts and unclear impact.
Instead of asking “how do I start?”, it is perhaps more helpful to ask “how do I make this a reality in my business?”
Over the past 2 years, I’ve been evolving a simple framework to think about this. Not as a rigid model, but as a way to make sense of what I see working, and not working, across organisations.
I think of it as stages, with many of them running in parallel and others missing altogether:
AI Operating System – an AI Adoption Framework
Source: Inside the Business Mind
On paper, the first stages could really allude to an IT roadmap. Roll out tools. Train people. Manage access. But that assumption is exactly where many organisations go wrong.
Enable and Educate: This is Not an IT Problem
A key challenge is that AI is not a new technology that the IT department needs to implement. It is a whole new way of working that everyone needs to understand so they can see how processes, roles and companies will be transformed. At last week’s CEO Forum on Driving AI Transformation, promoted by the D3 Institute, a partnership between Harvard Business School, Nova SBE and Nova Medical School, this was one of the key messages:
AI will not be done for you, you have to do it.
Karim Lakhani, HBS
Unlike many other IT transformations, this time, it is not about rolling out a tested technology. It is not about IT selecting a software and getting a video training on how it works and the features it has. The speed of evolution and the standardised personalisation provided by AI means that each job, function, role, and enterprise can be looking at it through different lenses and extracting different sources of value.
In traditionally structured enterprises, one would even question whether the conversation is being held at the board level or staying at the operational levels of the organisation. AI is not an operational decision to start with. It is a matter of strategy, of business model, of survival. The devil will be in the details of operationalising it, but the strategic drive needs to ensure it permeates the company as a new way of thinking, not as a new desktop item.
And, for the record, many of these companies don’t even have this IT department. Their size or existing operations do not justify having this function in-house.
The Importance of Exploration… when I still need to do my work
So if IT is not doing it for you, who is? The reality is that with AI, you have to experiment in the workplace while still having to deliver on yesterday’s company goals. Management teams need to push for adoption and testing while still having strict budget goals for their teams. It is easy to think of implementation in companies with large consultant resources and blessed with technical talent that can help bridge the conversation. But in most companies, each person will have to figure this out while also doing their day job,
In overstretched teams, this is a significant challenge. Some try to do patches and get a GenAI subscription that can respond to the questions coming “from above”. They replace Google search with an AI-related model, asking questions on the go, uploading information without giving it much thought and getting mixed results. They remain unconvinced, and often the subscriptions linger without the usage really paying off.
Because there is not enough time to invest in investigating deeply, many stay at the surface. As expected, some of the results are well, superficial!
Furthermore, with all the AI perils that are widely discussed and quickly evolving, and with less than a quarter of the companies having proper governance in place, many sceptics try it only to get their “fears” confirmed, querying the AI in ways that speak to their expectation that it does not work and it can never replace a human.
So they go back to work. That is what they are paid for.
FOMO is Driving Experimentation, But is it Enough?
So how do you start? This is where Exploration can take over, driven less by strategy and more by curiosity, pressure, and often, FOMO.
As with any large technological revolution, you need to identify the game-changers. Those who are keen to experiment, fearless about testing and focused on what the impact is. Those can be empowered and act as champions across teams and departments. Champions help get more people on board.
Think of them as the annoying person in the room who has done all their homework. They call you up to their desk to see their new agent. They respond to all their emails in a structured, clean manner. They are producing more reports than ever before. They have debugged the entire backlog of errors. And they are talking about it. With it, they create FOMO.
Another speaker at the conference mentioned how people who were using AI were more ready for promotions, and therefore, people were seeing it as a career accelerator. So what does it mean for those not using it? In an ideal scenario, it means they, too, get inspiration to experiment, to change their own processes, to make their own agents.
So is this all you want? Wide adoption per anyone’s wishes and desires?
Taking it up a notch: disseminate from individual to business
Here is the thing. Left to its own devices, FOMO-driven exploration creates noise as much as it creates value. The champion calling you to their desk is brilliant, but if their breakthrough never leaves their desk, it was an individual win, not a business one.
This is where Disseminate becomes critical. It is the deliberate act of taking what works and making it travel: cross-divisional forums where people share what actually moved the needle, peer-to-peer learning that goes beyond forwarding a LinkedIn post, and internal showcasing that turns individual champions into company-wide proof points.
The difference between a company stuck in experimentation and one that is genuinely scaling is not in the tools, not in the training, but in whether someone is intentionally assessing and spreading what works.
I feel like a lot of companies have not been able to enter this stage. They have bought a few licenses, provided a few trainings and asked people to experiment. A few early adopters are talking about it and making noise. Management gets to feel good about what they are doing. But is that it? Or are we just creating more inefficiencies in the process of becoming more efficient?
Experiments Without Results: What Will You Measure?
Exploration alone does not always yield measurable results. And as a side effect, you risk getting duplication and inefficiencies. If everyone is given free rein, you end up with multiple agents for the same role, similar agents across departments and other agents that are very role-specific but have very limited business impact. Who is tracking?
As companies get stuck in this Enable-Educate-Explore cycle, how do they then move to preparing for scale? How do they meaningfully select the pilots with a real business impact? Importantly, how do you make that choice and assess the business impact? Frequency of the task, time or cost reduction potential, and impact on the client should be the most important drivers. And after this, the ability to scale to other teams.
In practice, individual experimentation can generate significant individual benefits, but it is only economical if it can generate more value than what it is consuming, not only for the user, but for the corporation in general.
So rather than just saying “AI is changing everything”, perhaps we need to start by saying “AI has changed X for the business”. That should focus the conversation on real business impact.
- 2x increase in speed of response to client complaints
- 3x increase in the number of tenders for large clients
- 4x reduction of time spent on manual processes for management reports
- 5x increase in content production for digital channels
- 1.5x higher conversion from increased consistency of communication, all else equal
All these are made up, but all these are also under-estimated on what AI can really do in a business. We just need to start measuring it and focusing on it. Metrics are not meant to be an afterthought. They need to be built into the process from day one — before the pilot, not after.
No guidelines, no framework, no guardrails. Get Governance Going.
I don’t want to be the party-pooper, but it is easy to see how this can go wrong. Companies rushed to allow experimentation and see “how it goes” and they are faced with data spread all over different models and little guardrails around it.
We went for a policy of apologising later, but it is hard to continue in that mode anymore. 3 years into the widespread release of ChatGPT and with innovation duplication every 6 months, companies need to establish a few clear rules for employees to use these tools without jeopardising the infrastructure, the service delivery and, more importantly, the client relationship.
Data governance, software usage and risk management need to be part of everyone’s lingo, not just the general counsel’s or risk officers.
The risk increases even more in the presence of agents that are autonomous to do pieces of work inside the user’s computer, having access to more documents and holding the ability to edit information directly at the source. Not wanting to be like the EU regulating before we even know what we are regulating for, we are now at a stage where we can, at a minimum, mitigate what can go wrong.
Re-imagine. Yes, go back to the drawing board.
Underneath the adoption of AI, there is a deeper issue. We are trying to scale AI on top of processes that were never designed for it. This is the step most organisations skip: re-imagining the process itself. I know, boring. Who wants to talk about workflows and process design?
When I first worked on a framework for AI 2 years ago, my instinct would make me move straight from experimentation into dissemination, measurement and governance.
However, that is like applying an old software on an old mainframe system. It is not compatible.
Processes have historically been designed for human interactions and limitations. How can we scale AI on top of processes that were never designed for it? When we have agents handling processes, we need to reimagine what these can look like. I am not assuming a removal of humans in the loop; I am assuming an automation of replicable decisions that will be under human oversight. Though we kind of know that many processes will indeed become autonomous. We are just not entirely there yet. Because we still think of some of these processes as inherently human.
Re-imagining these processes will also mean we re-imagine the jobs of the future. There is talk of job losses or hiring freezes across industries. How do we keep employees motivated to do these key business changes if they don’t see a role for themselves in the future?
In the panel about the future of the workforce, there was more emphasis on reskilling than replacement. While I think part was driven out of a conservative and politically correct perspective, there is an element of moving up the value chain that is expected to happen. But that will require re-imagining functions, roles, and departments altogether. The honest answer is that Re-Imagine requires leaders to have the conversation about roles before people hear it in the press, not after.
For businesses, it will mean going back to the drawing board. If you were to set-up the business today with the existing tools, how would you do it? Because if you don’t ask that question, someone else will.
On Try AI for Growth, I explore how you can start exploring and seeing impact today. Listen in any of your favourite podcast apps or read on the blog:
Photo by Markus Winkler on Pexels.com