When AI Works, but Adoption Doesn’t

What a real-world scenario revealed about autonomy, user experience and change facilitation
 

We recently completed a solution defence for a potential new client in the airline industry. The opportunity centred on Level 1 (L1) operations, but the client’s definition of L1 went well beyond the traditional IT service desk. It included infrastructure functions across security, network, cloud and endpoints, as well as monitoring for more than 200 applications. Security monitoring was largely defined as ‘detect and notify’, with incidents passed on to the client-managed Security Operations Centre.

Read that scope again and the evolution of the Level 1 service layer becomes clear. Yet the opportunity itself is not the focus of this piece. I want to look at one layer below, at what the infrastructure services world typically calls Level 0 (L0): the largely invisible layer where platforms operate and AI components sit.

At L0, the requirement was clear. We had to show not only where AI would be used, but how it could enable an autonomous operating model. That was not unfamiliar territory for us. Since the rise of generative AI, and more recently agentic systems, helping clients move towards autonomous service delivery has become a consistent north star.

Our initial response explained how we would apply AI and build towards that model. A few weeks and several stages into the opportunity cycle, however, explanation was no longer enough. It was time to provide evidence.

A Familiar Moment at the Check-in Desk

For the demonstration, I chose a scenario that would feel immediately relevant to an airline. Imagine a typical day at the check-in counter. An agent verifies passports and visas, prints a boarding pass and directs the traveller towards the gate. It is airline-specific, but familiar to most of us as consumers.

The check-in process depends on several technologies working together: the computer terminal, baggage conveyor, scanners, or handheld RFID devices in some locations, and the printer.

What happens when the boarding-pass printer fails?

It is often the final step before the traveller leaves the counter. In many cases, including myself, the traveller has already begun to turn towards the exit. Failure at that moment creates an awkward disruption. The traveller becomes restless, while the check-in agent needs to understand whether the issue affects one desk or several. If it is isolated, she must seek help while keeping the customer waiting.

That was the moment in which we chose to employ AI. Because demonstration time is always limited, we focused on perhaps the hardest point: the instant the issue occurs, when the check-in agent is most exposed.

The AI First Responder

In our scenario, the agent calls for help and an AI voice assistant, Aisha, responds first. Aisha identifies the printer problem, proposes possible resolutions and checks what happens after each step. Sometimes a script is run; at other times, the agent takes physical action, such as checking the power cables. The important point is that the conversation happens in real time, with the AI responding seamlessly until the printer is working again.

The demonstration was successful. We showed how AI and autonomous systems could form the bedrock of service delivery before human intervention became necessary.

I was ecstatic.

I have been in demonstrations that did not perform as well as I wanted, and a few that simply failed. This one worked. But my enthusiasm did not last long. I saw the client’s reaction and saw uncertainty on their faces.

When a Successful Demo Raises the Harder Question

I asked whether the demonstration had been clear. The client said it was and even described the scenario as highly relatable. That reassured me, briefly. Then came the red line: they understood how we intended to use AI and autonomous resolution, but they were not convinced it would work on the ground.

Their comment was direct: ‘Amitash, the check-in agents won’t be half as patient as you are.’

At first, I was not sure why. I had deliberately introduced some of the messiness of a real interaction. I initially gave the voice assistant incorrect information, saying that several printers had failed, and then corrected myself to confirm that only one printer was affected. At another point, I was intentionally rude. We wanted to show that the assistant could handle false positives and imperfect human behaviour rather than follow an ideal script.

The client understood and appreciated that realism. They could also see wider possibilities for the voice assistant, particularly for knowledge workers. Their concern was more fundamental: how would this experience work across different user personas?

It was a fair question. I said we would return with further evidence, although I already had a sense of what sat behind their concern.

The Adoption Gap We Do Not Talk About Enough

After the presentation, we regrouped for debrief. Most of us were pleased with how it had gone, but I remained stuck on the client’s remark. I could not fully disagree with it. That, to me, is where extracting the best value from AI becomes difficult.

Since AI entered the enterprise mainstream, much of the conversation has focused on what these systems can do and how their capabilities can be developed to create meaningful value. It is largely a top-down view. From the ground up, however, the question of how employees and end-users will experience and adopt these systems has often been left to the users themselves.

I see a parallel with the modern railway system. Railways were initially designed around freight movement, with the passenger experience considered later. AI is (and has to be) different because, in all its forms, it still requires humans in the loop. Yet the way people will integrate with it, adapt to it and experience it remains insufficiently addressed.

This leads to a simple but important question: how can enterprises realise value from AI if end-users cannot adopt it? Innovators and early adopters will experiment willingly. Reaching the early majority and the larger workforce, however, requires more than a polished launch. It requires meaningful change facilitation.

Making Change Facilitation Part of the AI Experience

Enterprises need to treat adoption as part of the solution rather than an activity that follows deployment. Change facilitation can take several forms: focused user campaigns when the system is introduced, sustained nudges from support teams when users reach out for help, and even gamification where digital dexterity is lower.

At Microland, we have also learnt the importance of genuine management adoption and buy-in. This is not unique to AI; it applies to any major enterprise initiative. But with AI, visible leadership participation matters because technology changes not only processes, but also the way people interact with support, knowledge and decision-making systems.

A few days after the demonstration, we submitted an addendum outlining a change-facilitation approach and a brief adoption plan for different personas. I am happy to say that we are now working with the client to fine-tune the approach for the deployment.

Beyond the Demo

The experience left me with a useful reminder. A successful AI demonstration can prove that the technology works. It cannot, by itself, prove that people will use it.

If autonomous operations are to deliver their promised value, human experience cannot be treated as an afterthought. It must be designed with the same care as the technology itself.