The one-lever rule: every AI idea must move one of three numbers
More money in, fewer hours spent, or better quality out. If an AI project can't name which one it moves, from what number to what number, and who watches it, it isn't a project yet. It's a demo.
Most AI ideas I hear from business owners sound like this: "We should use AI for customer service." Or, "Can AI help with marketing?"
They aren't bad ideas. They just aren't decisions yet, because nobody can say what would count as success.
Three levers, borrowed and sharpened
Corey Ganim, who teaches AI services to consultants and operators, filters AI ideas through three levers:
- Effectiveness: make more money.
- Efficiency: save time or labour.
- Quality: improve the product, service or customer experience.
It's a useful filter. My twist is to make it harder to fake. An AI idea only passes when you can fill in one sentence:
This will move [lever] from [today's number] to [target number], and [person] will check it every [week or month].
If you can't fill in the numbers, you don't understand the problem yet. If you can't name the person, nobody will notice when it stops working.
What each lever looks like in my own work
Money: what a search campaign actually cost
When I ran Google Ads for my own product, the lever was cost per sign-up. The first campaign landed at SGD 19.28 per free-report sign-up. Once I could see that number per theme, I kept both of the marketing angles I'd been debating, because their costs were almost identical: SGD 18.15 and SGD 18.39. The number made the decision. How I got those numbers.
Time: from a day of replies to a review queue
Replying thoughtfully on X was worth doing but ate a full day for 50 to 100 replies. The lever was hours. My agent team now drafts and checks the replies, and my job is reviewing and approving. It has published more than 15,000 that way, and the daily review now takes about 20 minutes. The case study.
Quality: a pass mark before the customer sees it
For a chatbot I built, the lever was accuracy. Every answer was graded out of 100 against the source documents, and only those scoring 90 or above were sent. Everything else went back for revision or was handed to a person. How that works.
Why one lever, not three
It's tempting to promise all three at once: "It'll save time, bring in more leads and improve service!" In my experience, projects that promise everything get measured on nothing.
Pick the one lever that matters most for this project. The others may improve too, and that's a bonus. But the primary lever decides whether the project continues after the first month.
How to use it with a vendor, or with me
When anyone pitches you an AI tool, including me, ask them to fill in the sentence:
- Which lever?
- From what number, to what number?
- How will we measure it, and who checks?
A good answer sounds like: "Quote turnaround from two days to two hours, measured from the enquiry timestamp, checked weekly by your operations lead." A weak answer sounds like "increased productivity" or "AI-powered insights".
This is also how I structure an AI workflow audit: every opportunity gets a primary lever, an estimate of hours reclaimed or money gained, and an owner. Anything without a number goes on the "later" list.
A worked example (hypothetical)
Imagine a renovation firm that gets 60 enquiries a month through WhatsApp and its website. The owner quotes each one personally, and replies take about two days.
- Lever: money, because slow replies lose jobs to faster competitors.
- Today: a 2-day average reply, with about 1 in 5 enquiries becoming a site visit.
- Target: a same-day draft quote for the owner to approve, and 1 in 4 enquiries becoming a site visit.
- Owner: the office manager checks weekly.
The owner can now judge any tool, including an enquiry-to-quote system, against a clear before and after. And after a month, everyone knows whether it worked.