Which deals to refuse, and the arithmetic behind it

Two quotes are on the desk in the same week. One is for 300,000, one is for 70,000. The instinct is that the first one is the break, and it is wrong often enough to justify an hour of arithmetic. The reason is not that large customers are bad. It is that a large deal and a small deal can consume the same indivisible thing: the month of a person who cannot be in two places.

The unit that does not divide

A person-month is a physical quantity, not a budget line. If a delivery needs the same two people for the same two months, the 300,000 deal and four 70,000 deals cost the same input. What differs is what comes back, how fast, and how much arrives at all.

Most small practices price the work and forget to price the calendar. They compare 300,000 with 70,000, conclude the first is four times better, and only discover in month two that the second option was four jobs, four references and four chances to reuse what they built.

The arithmetic, with the assumptions showing

Take a two-person shop. A 300,000 engagement that drags both people for two months consumes four person-months. Assume 40 per cent up front — 120,000 — and 180,000 on completion. One practitioner reports that in his own sample, the probability of never collecting the final payment is above 60 per cent. At 40 per cent collection, the expected value is 120,000 plus 72,000, or 192,000.

The same four person-months, sold as four 70,000 jobs of one month each, stage-paid and collected, returns 280,000 — and leaves four verifiable references and whatever component got reused each time. The gap is 88,000 before procurement overhead, the rework a large scope reliably generates, and the fact that the large deal's requirements usually grow during delivery because nobody with authority froze them at the start.

Only two numbers come from a source: the one-to-two-month drag, and the collection observation from one person's sample. The rest is a model. The model is there to show the shape, not to predict your month.

The honest counter-argument: if the 300,000 deal has a named business owner, a fixed scope, half up front and an acceptor who signs at each stage, the arithmetic can reverse. The band is not the rule. The three variables are who owns it, what is frozen, and when money moves.

Four bands

Band	How it looks	What it costs	What to do
Under 50,000	easy to sign, fast to close	rarely covers the selling and communication time; fine as evidence, poor as a business	take it only if it produces a reusable asset or a reference you cannot get otherwise
50,000 to 200,000	unglamorous	one to two person-months, stage-payable, final payment small enough to survive	default yes; this is the band a small practice can actually staff
200,000 to 300,000	feels like a step up	often crosses into multi-department approval or public procurement; the cycle doubles and you carry the float	split it into two separately acceptable stages, or decline
Above 300,000	the one you want to tell people about	the same person-month is fully occupied, the final payment is uncertain, and the opportunity cost is highest	no prepayment and no named acceptor means no

Six deals to refuse

Each has a tell that shows in the first meeting, before any technical work.

1. Nobody owns the problem. Only IT is in the room, and no one from the business can describe what changes on Monday. Without an owner, acceptance has no author.
2. You may look but not touch. No real data, no account, no sample that reflects production. A demo built on invented data proves nothing about the system that will run.
3. The first meeting is the whole company. Every department, every use case, one budget. That is not a scope; it is an invitation to spend a quarter discovering one.
4. One person for pre-sales, build and support, with no defined boundary on support hours, no backup, and no answer for who covers a holiday. If scope and cover stay unclear, the role is a trap with a salary.
5. The demand locust. Three traits, and two are enough: work that sits outside the customer's stated strategy, a customer who prices you per head like cheap outsourcing, and internal politics heavy enough that the project's survival depends on one faction winning.
6. A pilot with no graduation standard. If nobody can say what result ends the pilot and what happens the day after, the pilot is a permanent audition that you fund.

Only one of the six is about money. The rest are about whether the engagement can end: an owner who can accept, data that can be checked, a scope that can close, a person who can say yes.

Refusal is a capacity decision

Refusing is not a moral posture. It is the only mechanism a small practice has for keeping its calendar for work that compounds. Every marginal deal accepted is a month that cannot be sold twice, and the cost never appears in the spreadsheet, because it arrives as revenue.

The practical form of a refusal matters as much as the decision. "No" closes a door; "not now, and here is what would change it" names the missing condition — an owner, a frozen scope, a paid validation round — and tests whether the customer can produce it. Most cannot, and the ones who can were never the bad deal. The seven-figure quote that died the same day is what happens when a team spends its energy on coverage instead of on the conditions of payment.

The stakeholder map is how you find out in week one whether an owner exists, and scoping and pricing is how the conditions get written down before work starts.

The test

Before signing, compute one number: expected collection divided by person-months. Expected collection is the upfront amount plus the balance multiplied by the probability you have actually experienced collecting it — not the probability you would like. If you cannot state that probability, you have not done enough deals to price this one; assume the worse one.

Then compare it with the best deal you declined this quarter. If the new number is lower, the deal is a loss dressed as growth, and the calendar already knows.

Refusing a deal is the other half of quoting for one. Three pricing models covers what to do with the deals you keep; this page covers the ones where no model rescues the arithmetic.