Skip to content
Back to Insights
An elegant business owner reviewing one clear decision card as disconnected message and document fragments quietly organise in the background, business partner cobalt gold and warm neutral editorial style.

AI Should Remove Friction, Not Leadership

Business Partner · 14 October 2026 · 6 min read

Two very different promises are often mixed together in conversations about AI and business decisions. One says that, given enough information, a system can make the decisions an owner used to make. The other says that AI can remove some of the effort involved in reaching those decisions, while leaving judgement and accountability with the owner.

The second promise is narrower, but more useful and easier to evaluate honestly. It recognises that an owner does not necessarily want to delegate an important decision; they want to stop spending the morning hunting for the information that decision requires. The friction before a decision is often a better target than the decision itself.

The cost of assembling the picture

Consider a customer who has become unusually quiet. To work out whether it matters, the owner may need to search messages, review an earlier agreement and remember what happened during the last conversation. Only then can they weigh a call against the other matters competing for attention.

The useful judgement is whether to intervene, what to say and how the relationship might be affected. Much of the preceding work is retrieval, comparison and reconstruction. It is necessary work, but it is not the part where the owner's experience creates the most value.

The same pattern appears in supplier commitments, overdue invoices and unresolved quotes. A relevant fact might be in a spreadsheet while the qualification sits in email or in somebody's memory. Rebuilding context again and again is costly precisely because it looks like ordinary administration.

Assistance should explain what it noticed

Software can help where the underlying information is available and the task is properly bounded. It may identify an overdue invoice from a supported report, connect it with a recorded owner response or prepare a draft follow-up for review. The quality of that assistance depends on whether the evidence is current and the conclusion is appropriately qualified.

A useful result says what it observed and why it brought the matter forward. The owner should be able to inspect the document, date or recorded conversation supporting the recommendation. When information is incomplete, the system should ask for clarification instead of inventing a confident explanation.

This is not a demand for lengthy technical reasoning on every screen. A short business explanation and an accessible evidence trail are usually more useful than a flood of internal detail. The owner needs enough context to challenge the recommendation, not another report to decipher.

A recommendation is not a decision

A prepared message can save time, but it should not be mistaken for a message sent. A suggestion to pay a supplier is not permission to move money, and a proposed customer discount is not a revised commercial agreement. Those distinctions are fundamental to maintaining trust.

There are actions an owner may authorise within defined limits, but that authority must be explicit, scoped and revocable. An AI tool should not infer broad permission from a request for advice, nor treat silence as confirmation where meaningful approval is required. The consequence of an action matters as much as the convenience of preparing it.

This also protects the owner from false reassurance. If a message was opened in an external app, that does not prove it was sent or delivered. If an employee reports that an action happened, the system should record the source of that information without silently promoting it to independently verified fact.

Helping an owner act is different from claiming the action already happened. Good assistance preserves that distinction in its language and its records.

Leadership depends on context that may be missing

No system can assume it knows everything the owner knows. A longstanding customer may be under unusual pressure, a supplier may have agreed to an exception, or an apparently attractive order may create delivery problems the records do not reveal. The right choice can turn on history, judgement and commitments that have never been digitised.

A strong tool therefore makes it easy to correct what it believes. It should treat an owner's clarification as relevant evidence, while remaining clear about whether the resulting state is reported, inferred or verified. That is more useful than forcing the owner to accept or reject a recommendation without being able to explain the difference.

Sometimes the sensible recommendation is to do nothing yet. A routine delay may not deserve a call, while an unexplained change in account details may require an immediate hold. The ability to defer, investigate or stop an action is part of competent management, not a failure to automate.

Leadership includes knowing when not to act.

More time for the decisions that matter

The commercial value of reducing friction is easy to miss because it often appears as time not spent. The owner does not have to reconstruct a quote history, rediscover a payment promise or search through unrelated messages before considering a next move. Each avoided search leaves more attention for customers, people, quality and growth.

That benefit should still be measured honestly. If evidence is missing or an integration is unavailable, the work has not been removed; it may simply require a different route. If the AI suggestion needs significant correction, that correction counts as work too. Useful systems should become clearer about these limits as they are used.

The longer-term opportunity is an operating relationship in which the business becomes easier to understand and important matters are less likely to disappear. That does not require the owner to hand over the choices that define the business. It requires support that notices what it can, remembers what is established and prepares the next responsible step.

The owner remains the person who decides what the business stands for and what risks it is willing to take. AI should clear the path to that decision, not claim the right to make it.

================ END WEBSITE BODY: DO NOT PASTE BELOW INTO WEBSITE BODY ================

================ LINKEDIN ARTICLE VERSION: COPY SEPARATELY ================

AI Should Remove Friction, Not Leadership

There are two different promises hidden in many conversations about AI.

One is that AI will make the owner's decisions. The other is that AI will remove much of the searching, checking and preparation that happens before a decision.

Those are not the same promise.

Consider a customer who has become unusually quiet. Before deciding whether to call, the owner may need to search messages, check a quote and recall the previous conversation.

The judgment is whether a call makes sense. The friction is reconstructing the context.

AI can be useful when it helps with that friction, provided it can show the evidence and the limits of what it knows.

A trustworthy assistant should be able to explain:

  • What it observed.
  • Why the matter deserves attention.
  • What information is missing.
  • What the owner can do next.

And it must preserve a vital distinction: preparing an action is not completing it.

Opening a drafted message is not proof that it was sent. Suggesting a payment is not permission to make it.

The owner also knows things that may not be in any connected system. A relationship, an exception or an unrecorded promise can change the right answer.

Leadership includes the right to correct, defer or reject a recommendation.

The goal is not to hand more decisions to technology. It is to spend less energy getting to the decisions that belong with the owner.

See your own Morning Brief

Send the aged debtors report your accounting system already produces and see who to chase first.

Try it with your report

Keep reading