Every ROI figure on this site comes from the same four-step method. It is the same method we use inside an AI ROI Roadmap Session , so this page is the working part of that engagement, described in public.
We publish it for a simple reason: a number you cannot check is not evidence. If we tell you a forecasting system is worth $213,000 a year and you cannot see how we got there, you are being asked to take it on trust — and most AI business cases are exactly that, a large number with no arithmetic behind it.
The four steps
1. Find the cost driver, not the annoyance. The starting question is never “what is frustrating?” but “where does the money actually go?” Those are different lists. The thing everyone complains about is often cheap; the expensive thing is usually silent — profit per client nobody calculates, capacity nobody measures, customers who leave without ever complaining. We map the process first and let the cost tell us where to look.
2. Take the baseline from your own data. Every estimate starts from a figure your business already has: what you spend on holding stock, what a departing employee costs to replace, how much revenue moves through a channel. Not an industry average, not a benchmark — your number. If the number does not exist yet, that absence is itself a finding, and usually an early one.
3. Apply a conservative improvement rate. This is the only step where judgement enters, so it is the one to scrutinise. The rate comes from what the technique reliably delivers, set at the low end of that range rather than the middle. A forecasting system that could plausibly cut holding costs by 15–25% is modelled at 15%. If the honest range is wide, we say so and model the bottom of it.
4. Subtract what it costs to run. Build cost, licence cost, the time your team spends feeding it. A model that saves $200,000 a year and takes $150,000 a year of someone’s attention to keep alive is not a $200,000 win, and a business case that ignores the second number is not a business case.
What that looks like
Take the sales forecasting example . The retailer spends $250,000 a year on holding costs — that is step 2, their figure. Better forecasting reliably takes 15–25% off that; we model 15% — step 3. So $250,000 × 15% = $37,500. One line, one assumption, arithmetic you can redo in your head.
Four lines like that make a total. If you disagree with a rate, change it and the total moves — which is the point. The tables on those pages separate inputs from outputs precisely so you can argue with the inputs.
What it deliberately leaves out
- Second-order effects. Better retention probably improves morale, which probably improves retention further. Probably is not a number, so it is not in the total.
- Anything we cannot trace to a stated input. If a benefit has no line in the inputs table, it does not appear in the outputs table, however real it is.
- Revenue we would have to assume you win. Cost you already carry is knowable. Sales you might make are a forecast on a forecast.
- The best case. Every rate is the conservative end. A model that only works at its optimistic setting is not a model, it is a hope.
Why the figures on this site are modelled, not client results
The systems described under Solutions are real: we designed and built them. The numbers attached to them are not the clients’ actual figures, because those are confidential and will stay that way.
So we do the next most useful thing — run the same method against a stated, representative business and show every step. You get to see how the estimate is constructed, which is the part that transfers to your situation. A real client’s $3.4M saving tells you about their business. A worked model tells you about the method, and lets you substitute your own inputs.
That is also the honest limit of these pages, and we would rather state it than imply otherwise: they demonstrate a method at a representative scale, not outcomes we are claiming on your behalf.
Doing this on your own business
The method is not secret and you do not need us to apply it. Take one process, find the annual cost, apply a rate you would defend to a sceptical CFO, subtract what it would cost to run, and see whether the remainder justifies the work. Most businesses have at least one process where the answer is obviously yes and nobody has ever done the arithmetic.
If you would rather have it done properly across the whole business — every function scanned, the two or three that actually hold the money worked through in detail, with the assumptions written down so your team can check them — that is what the AI ROI Roadmap Session is.