Before You Measure Whether AI Makes Money, You Need to Know Whether It Makes the Right Decision
By Malin Gaertig · September 25, 2026
Before you measure whether AI makes money, you need to know whether it makes the right decision. Bunnings is already measuring Buddy that way — and the definition of a good AI decision becomes something the business needs to own.
- —Before you measure whether AI makes money, you need to know whether it makes the right decision.
- —Bunnings measures Buddy beyond conversion — whether it gives the right advice, works as expected, how customers respond, and whether it produces unsafe behaviour.
- —An AI assistant can increase conversion and still be a bad system if the advice it gives is wrong or unsafe — trust is part of the product.
- —A good AI decision is judged on whether the system understood the need, used the right evidence, recommended an appropriate product, avoided unsafe advice and produced the outcome the business wanted.
- —Because those criteria are business-specific, the definition of a good AI decision is something the business needs to own.
| The obvious metrics | The right decision | |
|---|---|---|
| What is measured | Conversion, average order value and other financial outcomes. | Whether the advice is right, the system works as expected, and customers respond well. |
| What success proves | That the assistant is used and sells. | That the assistant can be trusted — the same system can increase conversion and still be a bad one. |
| The question asked | Did AI make money? | Did AI make the right decision, for the right reason, in a way we are prepared to trust? |
| Who owns the definition | The platform or vendor. | The business — the criteria are specific to it, so it owns the definition of a good AI decision. |
Before you measure whether AI makes money, you need to know whether it makes the right decision.
Bunnings is already treating that as part of the job. At the ARC AI Summit, Vivek Pradhan, GM of Data & AI at Bunnings, talked about how the business measures Buddy, its AI shopping assistant.
What metrics does Bunnings use to measure its AI assistant?
The obvious metrics are there: conversion, average order value and other financial outcomes.
But Bunnings is also watching whether Buddy is giving the right advice, whether the system is working as expected, how customers respond to its answers, and whether it produces potentially unsafe AI behaviour.
Why is trust part of the product for Bunnings?
That last part is particularly important.
For Bunnings, trust is part of the product. An AI assistant can increase conversion and still be a bad system if the advice it gives is wrong or unsafe.
So the question is: 'Did AI make the right decision, for the right reason, in a way we are prepared to trust?'
What does a good AI decision look like?
That requires a different kind of measurement and really goes to the heart of critical evaluation: did the system understand the customer's need, use the right evidence, recommend an appropriate product, avoid unsafe or misleading advice, and produce an outcome the business actually wanted.
Those criteria are specific to the business. Which means the definition of a good AI decision becomes something the business needs to own.
What should brands measure about AI decisions instead?
That's what MARK is all about — it measures whether AI selects your brand and products and what evidence appears to drive that selection.
Bunnings is asking whether its own assistant made the right decision. Brands outside the conversation should be asking the mirrored question: when an AI decides on a customer's behalf, is the evidence available to it good enough to choose you? That is the difference between being found and being chosen, measured as Share of Model.