AI Strategy · October 2, 2026 · 4 min de lecture
Your Team Is Already Delegating Decisions to AI. Nobody Decided Which Ones.
AI no longer just drafts. It ranks, filters and recommends, which are decisions in disguise, and few leaders chose which ones to delegate. Here are three questions to settle it.
Somewhere in your company today, someone will ask an AI which supplier to shortlist, which candidate deserves a second interview, or which customer complaint to answer first. They will not call it a decision. They will call it a quick check. By Friday, the AI has shaped a dozen choices, and nobody in leadership agreed to that.
What is actually happening
AI use almost always starts with drafting: an email to rephrase, a meeting to summarise. Then it slides, quietly, into something else: ranking leads, filtering CVs, prioritising tickets, recommending a price. These are no longer tasks. They are decisions dressed up as assistance.
Internal policies, where they exist, cover approved tools and sensitive data. Almost none cover decision rights. You find out the day a choice goes wrong and nobody can say who really made it.
The strategic reframe
Most leaders ask: "Can we trust the AI?" That is a quality question, and it leads straight into endless debates about models.
The question that matters is an accountability question. A highly reliable model can still be the wrong thing to delegate to if a mistake is expensive and invisible. A mediocre tool can be perfectly fine where mistakes are cheap and quickly spotted. The unit of analysis is not the tool, it is the decision.
Three questions are enough to classify a decision:
- Reversibility. If it is wrong, can we fix it this week at low cost?
- Visibility. Would someone see the error before it reaches a customer, an employee or the books?
- Ownership. Is there a named person who would answer for the outcome?
From there, three zones:
Four implications for an SME or leadership team
Name the owner before the tool. For every use that touches a decision, a name, not a department. Without an owner, zone 2 quietly becomes zone 1.
Make the review visible. Require one written line: why I accept, or why I depart from, the recommendation. Without it, human validation turns into a rubber stamp.
Watch the silent delegations. Rank, filter and sort are the risky verbs. They sound like preparation and work like a decision.
Revisit the zones every quarter. Trust is earned from observed errors, not from a vendor demo. A decision can move up to zone 1 when you have evidence, and back down when the context changes.
In the MAKIA framework, this is where two dimensions meet: Actors (who carries the decision) and Alignment (does the decision reflect what the company stands for).
Worth reflecting on
If a customer asked you tomorrow who decided to answer them that way, offer that price, or put them at the bottom of the list, could you answer without hesitating?
Try this week
In 20 minutes, list the last ten decisions where your team used AI (ask in a meeting or scan chat histories). For each one, answer the three questions and place it in a zone. Spot the ones treated as zone 1 that really belong in zone 2 or 3, and name an owner for each.