Method
Measure. Model. Install. Verify.
The fourth step is the one nobody else does, and it is the reason the first three can be trusted.
Measure
What the process costs today, before anything is bought. Hours, frequency, people, error rate, rework, and how much of it happens outside nine to five.
Measured rather than estimated. If the baseline is a guess, everything downstream is theatre — and self-reported hours are the largest single source of fake ROI in this industry.
Model
What the automation costs to own and what it returns, over three years, with a payback period and a sensitivity table. Both a conservative lens and an optimistic one, clearly labelled, so you choose which to believe.
The model is built from your numbers, not industry benchmarks. You keep it either way.
Install
One process, built and tested against your last twenty real examples, documented, and handed over with the procedure written down.
Where your existing tools can do the job, we use them. Most businesses already own AI features they have never opened.
Verify
Ninety days later: re-measure, compare actual against modelled, and publish the variance — including when it is worse than promised.
A model that is never checked is a sales document. This step is what makes the first three worth anything.
The three lenses
Every number gets looked at three ways.
Business cost
All the hours at what the work would cost to have somebody else do it. Conservative, defensible, and the lens an accountant will accept.
Opportunity
All the hours at what your hour earns when it is spent on billable or selling work. Bigger, and an assumption rather than a measurement — so it is labelled as one.
Personal
The share of that work done outside nine to five, and what your own time is worth to you. Always shown. Never added to the other two.
The rule that keeps it honest.
Hours saved are not money.
They become money only if the hours are redeployed to revenue work, a hire is avoided at a growth threshold, or the capacity is actually filled with demand. Otherwise they are capacity — useful, but not cash.
Most AI business cases silently assume the first of those and never say so. That one sleight of hand is why so many of these projects look good on paper and disappointing in the bank account.
In the model, when a lever depends on one of those assumptions, you have to pick it explicitly. The model will not let you have it both ways.
An owner spending twenty hours a week on admin is not just losing money. They are losing evenings with their family, and that has a value whether or not anyone puts a number on it.
See it applied to your business.
The audit runs the first two steps on your numbers, and hands you the model.