A customer asks for a refund. Your AI agent retrieves the order, checks the delivery record and finds the returns policy. So far, everything works.
Then it encounters an exception. The customer received the product late, used it once and now wants their money back. Two internal documents give different instructions. Customer service usually asks someone in sales, who sometimes checks with the managing director. Nobody has formally agreed who can authorise the refund.
The technology has raised a question the business has never properly answered.
This hypothetical example illustrates a challenge leaders should consider before expanding their use of AI: how much of the organisation’s everyday performance depends on people quietly filling the gaps?
The work your processes do not show
An experienced employee knows which document is out of date, which colleague can resolve an unusual request and when a procedure needs interpreting. That knowledge is valuable. It also means that a process can appear more reliable than it really is.
The weakness becomes visible when that employee is absent, the business grows, or an AI system is expected to take on more of the work.
AI agents—systems that can use information and tools to carry out a sequence of tasks—need access to the right information and clear limits on what they may do. When instructions conflict or authority is unclear, they may stop, escalate unnecessarily, or take inappropriate action.
Of course, failures can also come from the technology itself. A model may misinterpret a request, an integration may fail, or essential information may be missing. Leaders need to investigate those causes. But they should also ask whether the deployment has uncovered a problem employees have managed for years.
The cost is already in your business.
Return to the refund request. While the customer waits, several people exchange messages, revisit the policy, and search for someone willing to decide.
The cost extends beyond the time spent handling one case. The customer may need to contact the company again. Employees lose time for other work. Similar requests may receive different answers.
Adding AI could speed up the initial steps while leaving the delay untouched. More requests might reach the same unresolved decision point.
Before approving another automation project, identify where time is actually being lost. Is the work itself difficult, or does it spend most of its life waiting for information, permission or a handover?
That distinction determines what needs to change.
Make responsibility explicit
In the refund example, a workable response might include one authoritative policy, a named owner for exceptions and an agreed financial limit within which routine refunds can proceed. Cases outside those conditions would go to a designated person.
The appropriate limits will depend on the business and the consequences of an error. Some decisions deserve careful review. Others have accumulated approval steps whose purpose nobody can explain.
Clarity means knowing which is which.
It also means deciding who remains accountable when an AI system takes action. Giving a system permission to issue a refund does not remove the need for a person responsible for the policy, its application, and handling mistakes.
This is a Business Design question: how should information, authority and work fit together to deliver a reliable customer outcome?
Reconsider what managers spend their time doing.
AI also raises a practical management question.
If managers need less time to prepare updates, compile reports, or pass information between departments, how should they use that time?
Managers still have to develop people, resolve competing priorities, exercise judgement and take responsibility for difficult decisions. Those contributions become especially important when routine work is increasingly automated.
The useful starting point is to examine activities within a role. Which help people make better decisions? Which support customers or colleagues? Which exist because systems do not connect or responsibilities remain unclear?
That assessment provides a better basis for changing roles than assuming an entire management layer has become unnecessary.
Start with one workflow.
A business doesn't need to redesign its entire organisation to learn something useful.
Choose one recurring workflow that matters to customers and regularly causes delay. A refund, quotation, onboarding process or service request can provide a manageable starting point.
Follow a real case from beginning to end. Include the messages, informal consultations and repeated checks that the process diagram leaves out. Talk to the people doing the work: they are likely to know where it gets stuck and why.
Then address the most consequential obstacle. That might mean clarifying one decision, removing a duplicate check or making a reliable source of information available.
Establish a baseline and test the change. Depending on the workflow, useful measures could include resolution time, the number of handovers, repeated customer contacts, or rework.
This gives leaders evidence for the next investment—and a chance to deliver a useful improvement without committing immediately to a large transformation programme.
Build the ability to respond.
As AI systems take on longer sequences of work, more organisational assumptions will need to become explicit. Businesses that begin examining those assumptions now can learn where automation helps, where human judgement remains essential and which working arrangements need attention.
The broader benefit is a better-equipped organisation that can respond when conditions change. Clear responsibilities and reliable information help people act, whether the trigger is new technology, a customer problem or a shift in the market.
A useful question for your next leadership meeting is:
Which part of our business works because the process is sound—and which part works because someone keeps rescuing it?
Find a practical starting point.
If an AI initiative is revealing delays, unclear responsibilities or repeated handovers, bring one priority workflow to a conversation with MaxMORIX
EXPERTS. We can discuss whether a focused Strategy Diagnostic would help identify the main obstacles and prioritise practical improvements.
