Mistakes and lessons from an AI transformation specialist

By Bjorn Schliebitz, AI Transformation Specialist and Umbrella Club member

I have spent years helping organisations move from AI ambition to AI outcomes.
In that time, I have watched the same mistakes play out again and again, across industries, across company sizes, and across very different starting points. None of these mistakes is new either. Digital transformation had its own version. So did cloud, and mobile before that!

What is different this time is the speed. Boards want to see AI activity now, and that pressure is creating a genuine split. Some organisations are moving too fast, without a strategy behind the technology. Others are stuck in planning mode, too cautious to actually start. Both end up in roughly the same place: a lot of activity, and very little real value.

In this article, I’ll share how to find that happy medium where AI and automation are implemented with purpose and set up for success.

Strategy first, technology second

Automating for the sake of automating is almost guaranteed to fail. I want to be direct about that, because it is the mistake I see most often. The technology is rarely the problem. The absence of a clear reason for using it is.

You want to first look at the strategy of your organisation, and then, through that lens, ask yourself how you could use this technology to create more value for your customers or clients. That question has to come before any decision about tools, vendors, or platforms. Skip it, and you end up automating tasks rather than generating outcomes.

This also shapes which tasks are actually worth automating in the first place. It is the menial, heavily standardised tasks, where you have standard operating procedures well documented and understood, that are the ones worth starting with.

You do not want to automate something in a new business area, or a process where you do not quite know why it is currently ineffective, because you could simply be automating the existing mess and making it a lot larger.

Fear of missing out is driving a lot of what is happening right now

If I had to name the real driver behind most current AI activity, it would not be strategy. I think it is mostly a fear of missing out. Boards are pressuring CEOs to do something with AI, and then CEOs need to be seen doing something with AI, so they end up starting anything.

That gap between activity and substance shows up clearly in the data too. MIT’s 2025 study on enterprise AI adoption found that 95% of generative AI pilots fail to deliver a measurable return, with most organisations stuck evaluating generic tools rather than integrating AI properly into their actual workflows. It backs up exactly what I see in practice: plenty of movement, very little transformation.

That gap between claimed success and actual outcomes runs right through leadership too. CEOs who say they have had a successful AI rollout will often find that employees disagree, because in a lot of cases it is still limited to buying licenses and rolling out Copilot, without preparing staff on what they are meant to do with it, or giving them training on how to correctly use these tools.

The conversation most leaders avoid

Of everything I have seen go wrong, this is the one I think is the most avoidable. The biggest impediment that companies create for themselves is that they do not speak about the impact on job descriptions and job roles.

Are we going to lay people off or not?

That is something a leader needs to tackle up front, even if the honest answer is uncertainty. It is fine to say, “We know we need to do something here, we do not want to let people go if we can avoid it, but we may need to slow down hiring”.

Being honest is something people appreciate, because the alternative is uncertainty, worry, and, eventually, people actively sabotaging AI rollouts.

Why ‘please don’t do that’ doesn’t work as a safety measure

There is a technical mistake I see almost as often, usually from people who understand the strategy well but have not spent enough time on the engineering side. Soft guardrails really are not guardrails. If you politely ask a model in a prompt to never do a particular thing, that is not going to hold, because these are probabilistic systems.

At some point, that system might give someone a gift voucher over a thousand dollars, even though it was only ever supposed to approve amounts up to a hundred.

You have to put in place hard guardrails that check an action before it executes, not soft instructions that hope the system behaves. That distinction matters more than most non-technical leaders realise.

Crawl, walk, then run

The tension between moving too fast and moving too slowly shows up constantly. I see organisations either going too fast or too slow on AI rollouts, and a lot of that comes down to focusing too heavily on internal cost elimination as the entire goal, rather than the starting point.

The fix is not complicated, even if it requires discipline. It is important to learn how to crawl, walk, and then run, rather than committing to goals that are highly aspirational before you have proven any value at all. Begin with something that produces real, measurable value, and build from there.

A quick way to check where your organisation actually sits: