The decision room: what managers get wrong about AI
Bogdan Hruban · Wizeri
We have spent time this year in rooms with owners, founders and management teams, running workshops on AI before anything gets built. The technical questions are rarely the hard part. The hard part is that the people deciding do not share the same picture of what AI can do, what it costs and where the risk sits. This note is about the mistakes that follow from that, and what we have seen work instead.
It is a decision problem, not a technology problem
Most companies we meet are not short of ideas. They are short of a decision. Somebody wants an assistant, somebody else wants the reports automated, the finance lead is worried about data, the founder has read something alarming. The meeting ends with "let's explore it" and nothing happens for another quarter.
Every month of that costs more than it looks, not because the technology is moving fast, but because the manual work keeps being done by hand while the discussion continues. The fix is not more information. It is a room where the people who make the call agree on one shared picture and pick one thing.
Mistake one: starting with the tool
"We need a chatbot" is a tool. "Orders arrive by email and someone types them into the ERP" is a process. Companies that start with the tool end up with a demo that impresses the board and changes nothing on Monday. Companies that start with the process end up with less manual work.
In the workshops we map the manual work inside the company first, then rank it by impact and effort. The tool question answers itself once you know which process you are fixing. Often the answer is a plain integration with no AI in it, and that is a good outcome, not a disappointment.
Mistake two: the all-or-nothing bet
The second pattern is the big programme: a transformation, a platform, a multi-year budget, a vendor who will "do AI" for the whole company. Owners get pushed into these because they sound decisive. They are the opposite. A bet you cannot unwind is a bet you cannot learn from.
What works is a pilot: one process, a clear scope and a way to measure it, chosen by impact, feasibility and payback. If it works, you know what the next one looks like. If it does not, you have lost weeks and learned something true about your own data. We would rather prove value on one process first, keep the risk small, and let you decide what happens next.
Mistake three: treating fear as an obstacle instead of information
Every room has the same fears. Will this expose our data or our clients? Will the team resist it? Will we become dependent on a vendor or a model we do not control? Will it actually pay back? The mistake is to treat these as objections to be overcome. They are the agenda.
Security first means deciding, before anything is built, where the data may go, what stays on your infrastructure and who can see what. Dependence is answered by building on tools and access you already control, with documentation your team can operate. Payback is answered by measuring before and after. Resistance is answered by involving the people who do the work in choosing the process, because the person who types the orders in knows exactly which step is worth removing.
Once the fears have concrete answers, the decision gets easy. Until then, no amount of enthusiasm from the front of the room helps.
Mistake four: deciding without the people who own the process
AI gets discussed at the top and implemented at the bottom, and the layer in between, the department heads who own operations, finance and sales, hears about it last. They are the people who know where the manual work is. A decision made without them is made on a picture of the company that is a few years out of date.
The rooms that work have the owner, the management team and the department heads together, looking at the same process map. Not a course where someone explains AI. A working day where each part produces something: a ranked list of opportunities, a security checklist, a plan for a first pilot. People leave with a decision, not with ideas to explore.
What a good room produces
Three things, in our experience. A map of the processes worth automating, ranked, so the argument about where to start is over. A checklist of the questions to ask before any AI touches your data, so the risk conversation has a shape. And a plan for one pilot: the process, the scope, and how it will be measured.
Whether that pilot involves us is a separate question, and the workshop is deliberately not a sales pitch for it. The point is that the next decision your team makes about AI is an informed one.
If you are the person who has to decide
Pick one process that is visibly slow. Put the people who own it in the room with the people who sign. Agree where the data may and may not go before you look at any tool. Choose the smallest version that would prove something. Then build that, measure it, and decide again.
That is the whole method. It fits on a page, and the companies that follow it stop having AI meetings and start having less manual work. If your team would rather do that first part together, with us in the room, that is what our workshops are for.
Want to talk about automation for your business?
Schedule a free call or start with a quick scan.