Two years into the AI wave, most businesses have one of two things: a chatbot in the corner of the website that answers questions nobody asked, or a pilot that impressed everyone in a demo and never reached a workflow.
Neither creates leverage. Leverage is when a fixed input — a person’s hour, a marketing dollar, an enquiry — produces more output than before. AI does this in specific, unglamorous places. Knowing where is most of the value.
Where the leverage is
Response time. An enquiry answered in two minutes converts at a different rate from one answered the next morning. Not marginally — categorically. An AI that reads the enquiry, asks the two qualifying questions, and books the consultation or routes the lead to the right person is worth more than most websites. It works at 11pm, on Sundays and during the busy season.
Reading and writing at volume. Proposals, quotes, summaries of calls, replies to routine emails, extraction of details from documents. This work consumes skilled people’s hours and is highly repetitive in structure. AI drafts; a person reviews. The hours come back.
Qualification and prioritisation. When forty leads arrive and a team can call ten, which ten? A model that scores intent from what the prospect wrote, where they came from and what they looked at makes that decision consistently, every time, in seconds.
Personalisation at scale. “You last visited in March; here is the service you usually book, with a time that suits your pattern.” Trivial for one customer, impossible by hand for five thousand, routine for a system with the data.
Decision support from your own data. Not generic insight — your pipeline, your margins, your seasonality, summarised for the person who decides. The value is in the connection to the data, not in the model.
Where it isn’t
A chatbot with no access. If the assistant cannot see the calendar, the CRM or the order system, it can only talk. Talking is not leverage.
Automation of a process nobody has defined. AI does not fix an undefined process; it makes it fail faster. Design the workflow first.
Novelty. Anything whose business case is “customers will find it impressive.” They will not, for long.
Replacing judgement that carries risk. AI should draft the proposal, not sign it; recommend the discount, not grant it; flag the anomaly, not reverse the transaction. The human handoffs are part of the architecture, not a compromise.
The three conditions
In our work, AI creates measurable leverage when three conditions hold:
- It is integrated. It reads from and writes to the systems of record — CRM, calendar, documents, orders. An AI system that lives outside the workflow is a toy.
- It is supervised. Actions are logged, reviewable and reversible. Sensitive actions require a person. This is what makes the business comfortable letting it run.
- It is measured. Response time, hours saved, conversion at the stage it touches. If the number does not move, the AI is removed or redesigned — like any other component.
What this means for investment
The businesses that get real returns from AI are rarely the ones with the most ambitious AI strategy. They are the ones with a clear picture of their own workflow, a system of record the AI can work inside, and one or two well-chosen places where speed, volume or consistency turns directly into money.
Start there. The impressive things can come later, funded by the boring ones.
