AI won’t fix yesterday’s org chart

AI won't fix outdated decision-making. Paul Donkers on redesigning approvals, human judgment and accountability before you approve your 2027 plan.

A faster report is useful. A decision that no longer needs three meetings may be worth considerably more.

AI is on the agenda of almost every leadership team I work with.

A pilot here. A licence there. A working group exploring the opportunities. People writing faster, summarising more and preparing better slides.

Good. There is real value in that.

But underneath, the same people still approve the same things. The same reports go through the same meetings. The same decisions wait for the same person.

We have improved the tools. We have not necessarily improved the way the business works.

As you prepare your 2027 plan, that distinction matters.

Key takeaways

Adding AI can improve an existing process. It should also prompt you to ask whether that process still needs to exist.
When AI moves from preparing information to taking action, decision rights and accountability need to change with it.
Start with the decisions that matter—not a shopping list of tools.

The approval nobody questions

Consider a routine customer quote.

Someone prepares it. Someone checks the margin. Someone else approves a discount. A manager reviews it before it goes back to the customer.

Now introduce AI.

The quote is prepared in seconds. The margin check is automated. The manager receives a beautifully written summary.

Then it waits in their inbox until Thursday.

The individual tasks are faster. The customer may notice very little.

Perhaps that approval is essential. Perhaps it protects the business from a commitment that is difficult to reverse.

But perhaps it exists because of a mistake made seven years ago, under a manager who has since left.

AI cannot make that distinction for your leadership team. You have to examine why the approval exists and what would justify keeping it.

Before you automate the work, ask which part of it still needs doing.

Success makes this harder

Established companies have advantages. Customers, expertise, relationships, capital. I would not dismiss any of them.

But they also have arrangements that nobody remembers choosing.

Reports that once solved a problem. Approval limits that have not moved with the business. Meetings whose original purpose has disappeared, while the calendar invitation survives.

These arrangements are harder to question when the company is doing well.

“We have always done it this way” sounds weak.

“This is how we built a successful business” sounds much more convincing.

Sometimes it is a good argument. Sometimes it protects a process long after that process has stopped protecting the business.

The challenge is to know the difference.

When the tool starts acting

Much of the first wave of AI helped people prepare work: write, search, summarise, analyse.

AI agents can go further. Within defined permissions, they can carry out several steps, use systems and take action without asking for a new prompt each time.

That changes the discussion.

Drafting a customer response is one thing. Sending it is another.

Recommending a supplier is one thing. Committing expenditure is another.

Identifying a pricing exception is one thing. Deciding to accept it is another.

The boundary matters more than the label on the software.

For each use, someone needs to decide what the system may do, when it must stop and who handles the exception.

And someone must remain answerable for the result.

“The agent did it” is not an explanation a customer, regulator or board should have to accept.

Where does judgment belong?

This is also the question behind VELIQA: where does your personal judgment genuinely add value?

It is a useful starting point for your own working week. AI makes the same question relevant to the design of the wider organisation.

Some work creates considerable value and requires experience, context and difficult trade-offs. Give people more room for that work.

Other decisions are repeatable, with clear rules and outcomes you can check. Those may be candidates for automation.

But apparently routine does not mean harmless. A simple decision made incorrectly thousands of times can become a serious problem.

Look at the consequences, the quality of the information and whether a mistake can be reversed. Decide how you will detect failure before you decide how much freedom to give the system.

Delegate the action where appropriate. Do not leave accountability unassigned.

The concern behind the polite answer

There is a human question underneath all this.

If someone helps you automate a substantial part of their job, what happens to them?

You may be thinking about capacity and competitiveness. They may be thinking about their mortgage.

Do not mistake a polite response for confidence in the change.

Equally, do not promise that every job will remain unchanged if you cannot stand behind that promise.

Explain what you know. Be honest about what you do not know. Involve people in redesigning the work and give them practical support to develop the capabilities you will need.

That does not remove every difficult consequence.

It is a better basis for trust than asking for enthusiasm while avoiding the obvious question.

Before you approve the 2027 plan

Choose ten recurring decisions that materially affect your business. Not ten AI projects. Ten decisions.

For each one, ask:

Who owns it today? Include the people who check, approve and can overrule it. What genuinely needs human judgment? What could a system prepare or execute, under which limits? What would we change around it? An approval, a meeting, a role, a measure—or the decision itself.

Redesign decisions, not just tasks: map ownership, set the boundary for AI and redesign the work. Keep a named person accountable and test the whole outcome.

Then test one redesigned process. Measure the whole outcome: time to decision, quality, exceptions, customer impact and the work created elsewhere.

Not just how quickly the first draft appears.

An organisation chart may eventually need to change. Moving the boxes is not where I would start.

I would start with a question:

If AI gives your people time back, have you decided what you want their judgment for?

Try VELIQA on three activities from your own week. Then bring the same discussion to your leadership team.

Try VELIQA: reflect on three activities from your week

Over to you.