Most conversations about AI start from the wrong end. They start with the technology and work towards a use for it, which is how businesses end up with something clever that nobody can defend at the next budget review.
This starts from the other end. Where is money already leaking out of the business, how much of it, and would catching some of it pay for the thing that catches it. Answer that and what to build becomes obvious. Skip it and no amount of technical quality will save the project.
Four questions, in this order. The order matters, because each one is only worth asking if the previous one gave you a number you liked.
What is one sale worth, over its life?
Not the invoice. The whole relationship: repeat purchases, the servicing, the contract term, the years they stay. A place on a programme is a year of fees, not a first session. A machine sale carries a decade of parts behind it. A client on a monthly arrangement is the monthly figure multiplied by how long the average one actually stays, which is usually longer than people guess and worth checking rather than assuming.
One number, and it governs everything after it. If one sale is worth a few hundred, then catching a handful more enquiries a month is a rounding error and no system is going to change that arithmetic. If one sale is worth tens of thousands, one extra a quarter reshapes the year, and suddenly a lot of things are worth doing that were not worth doing before.
Push past the first answer you get. The transaction price is the number people quote because it is the one on the invoice. The lifetime figure is the one that tells you what a lost enquiry actually cost.
How do chances at a sale arrive?
A chance is one countable opportunity to make a sale: an enquiry that comes in, a quote that goes out, a booking in the diary, a referral passed on, a past customer who gets back in touch. Not a lead in the abstract sense. A specific, dated, countable thing that either happened or did not.
Map every route they arrive by. The phone, the form on the site, the inbox, referrals from people in your industry, word of mouth, walk-ins, existing customers coming back, outbound if you do any.
Then map the real ones rather than the intended ones. The routes on a website and the routes people actually use are frequently different, and the gap is where the surprises live. The reliable way to find out is to take the last twenty customers and establish, one by one, how each of them first made contact. It takes an hour and it regularly overturns what everyone in the business believed about their own funnel.
Where do those chances die?
This is the diagnosis, and in practice it is nearly always one of five places. Take each one, measure it, and write the number down. Measured, not estimated: the entire value of this exercise is that the numbers come from your records rather than from anyone's impression.
The dormant list is the one worth doing first, and it is worth understanding why. A new relationship warms over weeks: outbound to strangers takes time to produce anything, and time is exactly what a first project does not have. A dormant list produces inside days, from people who already know your name and once wanted what you sell. It is the closest thing to free revenue that exists in a business, it is already paid for, and in most places nobody has touched it because it is nobody's job.
What is one more sale worth against the cost?
Take the lifetime value from the first question. Compare it to what a system would cost you for a year. If one extra sale a quarter does not pay for the whole thing several times over, stop here.
That is a genuine stopping point, not a rhetorical one. It applies to us as much as to any other vendor, and it is the question we ask ourselves before quoting anyone: if the arithmetic is close, the project is not worth doing, because a system that has to be argued for every month will be cancelled in month four whether or not it is working.
The value conversation belongs in a room, out loud, after the numbers are in, and it is built from their figures rather than ours. It does not belong on a slide, and a number computed live on screen from someone's off-the-cuff estimate is worth nothing to either side.
Count the chances, not the closes
One distinction is worth being strict about, because it decides what anyone can fairly be held to.
A system can be held to chances: enquiries answered, quotes issued, bookings made, dormant customers brought back. Those are produced by the system, they are timestamped, and there is nothing to argue about at the end of the month.
It cannot be held to closes. Whether those chances turn into sales depends on your pricing, your reputation, your location, who answers the phone and how the quote lands. That is your side of the line and it should stay there. Anyone offering to guarantee the closing side is either not thinking clearly about attribution or is planning to argue with you about it later.
Worth saying plainly, since it cuts against the sales instinct: this makes the promise smaller and it makes it real. A smaller promise you can prove beats a large one that becomes a dispute.
When to walk away
Four situations where the honest answer is that there is nothing here, and it is better to establish that in an afternoon than in month three.
- The arithmetic does not clear. One more sale a quarter has to pay for the system several times over. If it does not, no build fixes that.
- Your customers are allocated, not won. Where a third party decides who you get, there is no enquiry to catch and no follow-up to make. There is no lever, and this is worth checking early because it is invisible from the outside.
- Nothing can be measured. If the system cannot own the phone number, the enquiry inbox and the web form, then what it produced is a claim rather than a record, and you will be relying on someone's word every month.
- The only value is saved hours. If there is no revenue side at all, the case usually will not carry the cost. That reasoning is set out in where AI actually pays.
Five numbers, this week
If you do nothing else with this, pull these from your own records. None of them need a tool and all of them are in systems you already have.
- Enquiries received in the last ninety days, and how many got a reply within an hour.
- How many arrived outside working hours, and what happened to those.
- Average days from a quote being requested to it going out.
- The count of records nobody has contacted in twelve months.
- What you could deliver next month that is not currently sold.
Those five numbers are the entire diagnosis. Most of the time one of them is visibly worse than the others, and that is where the money is. It is also, without exception, the thing we would point a system at first.