How do you size and price an engagement when the scope is still uncertain, and what do you do when you are told to cut the number?

Estimating and pricing are two different activities

An estimate is a claim about how much work something is. A price is a commercial decision about what to charge, informed by the estimate, the competition, the strategic value of the account and the margin the business needs. Conflating them damages both halves: estimates get bent to reach a price, so the delivery team inherits a number nobody believes, and prices get set as cost-plus, so you leave money on deals where you were the only credible bidder.

Keep them ordered. Establish what the work is, estimate it honestly with its uncertainty attached, and only then decide what to charge, who signs off the margin, and what shape the contract takes. If someone wants the number lower, that conversation happens on the pricing side or the scope side, never inside the estimate.

Decompose until the pieces resemble work you have delivered

Estimating an unfamiliar whole is guessing; estimating parts you have done before is calibration. Break the engagement into workstreams small enough that each either resembles something your organisation has delivered — with actuals you can look up — or is explicitly flagged as novel. The flag matters as much as the number: a workstream nobody has done is not more expensive on average, it is wider in range, and treating those two things the same is how estimates fail.

For each workstream write the basis of estimate in one line. Not the number, the reason for it: "twelve days per source integration, based on three delivered for the same platform, two of which used the same file format". A number with a stated basis can be challenged productively by a delivery lead. A number without one can only be argued about.

Three-point ranges, with the range kept visible

Give each workstream an optimistic, likely and pessimistic figure. The purpose is not statistical rigour; it is that the spread shows where the risk lives, and that is the input to every decision that follows.

Workstream	Basis of estimate	Best	Likely	Worst
Discovery and design baseline	2 architects, 3 weeks, as on the last two of these	25	30	40
Ingest integrations, 3 source systems	12 days each; 2 formats delivered before, 1 unknown	30	45	75
Data model and migration	4.2m rows; one prior migration of comparable size	35	50	90
Identity and access, existing Entra groups	Done twice on this pattern, low variance	8	10	15
Reporting rebuild, 14 reports	1.5 days per report; 3 have no current specification	18	25	40
Performance and failover testing	Never tested at this transaction volume	15	25	50
Cutover and 4 weeks hypercare	1 engineer plus partial second, from the last cutover	20	25	35
Total person-days		151	210	345

Do not quote 345, because it assumes every risk fires simultaneously and nothing that large is competitive. Do not quote 151 either, because the optimistic case is a fantasy that requires nothing at all to go wrong across seven workstreams. Quote the likely figure with contingency as its own line: at 15 per cent, 210 becomes 32 days of contingency and a total of 242 person-days, and that percentage is defensible precisely because you can point at the two rows driving it.

Keeping contingency visible rather than padding each task by a fifth matters because padded tasks always get spent, nobody knowing which part of a number was buffer, whereas a visible line can be governed and released. It also survives challenge: you can defend a named risk allowance, and you cannot defend inflation you have denied is there.

The two rows with the widest spread — migration and performance testing — are also the honest argument for a paid discovery phase. "Two of these seven workstreams account for most of the range, and a three-week paid design phase would narrow them before either of us commits to a fixed price" is a better commercial position than pricing your ignorance and hoping.

Let the contract shape absorb the uncertainty

The commercial model is a risk-allocation decision, so choose it against how much you genuinely know. Time and materials puts the risk on the customer and is honest when scope is open, though many buyers cannot sign it. Fixed price transfers risk to you and is only sane where scope is closed and change control is real; priced from the likely figure with no contingency it is a loss you agreed to in advance. Capped time and materials splits the difference and is frequently right. And phasing — a fixed price on a narrow phase one, later phases estimated but not committed — converts an uncertainty problem into a sequencing problem, which is far easier to sell than a range. State the assumptions the price depends on in the same breath as the price, because a number quoted without its conditions is remembered without them too.

When you are told to take out fifteen per cent

This request is normal and is not a challenge to your integrity. Somebody has heard a budget figure or a competitor's number, and the professional response is to make the trade explicit rather than to argue or to comply silently. There are four honest levers and one dishonest one.

You can reduce scope, which is the first thing to reach for and the only lever that changes the work rather than the wrapper. You can change the shape — a smaller committed phase one, the rest priced later — which reduces what the customer signs without reducing the eventual total. You can change the team mix, using a different blend of seniority or delivery location, which is legitimate as long as you are honest that supervision overhead and elapsed time both move. And the business can accept lower margin as a deliberate decision, which is a real option but not yours to make quietly: whoever owns the margin approves it in writing.

The dishonest lever is the same scope in fewer days. It reads as cooperation, wins the deal, and hands your delivery organisation an engagement that is under water in month four — at which point the change requests start and the relationship you were protecting is worse than if you had lost.

Make the cut arithmetically visible so it is a choice they are making with you:

Original:  242 person-days (210 likely + 32 contingency)
Request:   "get it to around 205"

Reporting rebuild: 6 priority reports instead of 14        -14 days
Hypercare: 2 weeks instead of 4, then standard support     -12 days
Migration: 24 months of history, not full 9 years          -10 days
                                                          --------
Revised:   206 person-days.  Contingency unchanged at 32.
Deferred to a phase 2, estimated 40-55 days, not committed now.

Three things about that block. The contingency did not move, because the risk did not change, and cutting the risk allowance to hit a number is the same error as cutting the estimate. Everything removed is written down as a later phase, so it is deferred rather than forgotten, which protects you when it is asked for in month five and is also how the deferred scope eventually gets sold. And the customer chose which capability to lose, so that month-five conversation starts from their decision rather than your shortfall.

The sentence that does this work sounds roughly like: "I can get to 206 days. What comes out is ten of the fourteen reports and half the hypercare — I would rather show you exactly what you are giving up than take days off the same work and have us both discover the problem in March." Said early, that is respected. Said after you have already agreed a lower number, it is a renegotiation.

The estimate that loses money is rarely the one that was too low

Post-mortems on unprofitable engagements usually find the estimate was defensible and the failure was elsewhere: an assumption never written down, a dependency with no named owner, no change-control mechanism, or acceptance criteria vague enough that "finished" became the customer's opinion. An estimate 20 per cent light on a contract with working change control recovers. A perfectly accurate estimate on a contract where anything unmentioned is assumed included does not.

Which is why an interviewer probing estimation is probing commercial literacy rather than arithmetic. Anyone can produce a plausible day count. What is being scored is whether you know which numbers are guesses, whether you priced the risk where it lives, and whether you can be pushed on the total without quietly moving the work.

Estimate the work, then price the deal, and keep the two conversations apart — when the number has to come down, something the customer can see must leave the scope, because the alternative is a commitment your delivery team discovers is impossible after the margin has already been booked.