Ask a consulting firm why an engagement lost money and you will hear about the client: they expanded the ask, they were slow with approvals, their data was a mess. All true, and all beside the point. The engagement was lost earlier — in the week the scope was written, when nobody could say precisely what was being sold.
Underbidding is not a pricing failure. It is a scoping failure that pricing inherits.
Why bespoke work underbids by default
Product companies scope once and sell many times. Advisory firms scope fresh for every deal, under time pressure, in a document that doubles as a sales pitch. That produces three predictable failure modes:
Optimism gets written into the boundary. The proposal is written to win, so ambiguous items get phrased generously. "We'll review the current process" costs one workshop or three weeks depending on who reads it — and the client always reads it the expensive way.
Effort is estimated from the best prior engagement, not the median one. The partner remembers the client whose team was responsive and whose data was clean. The estimate quietly assumes this client is that client.
Exclusions are left implicit. What you are not doing rarely makes it into the document, because listing exclusions feels adversarial during a sale. Then delivery starts, and everything unlisted is negotiable — in one direction.
None of these are fixed by adding a contingency percentage on top. A 15% buffer on a scope that is 40% ambiguous is just a slower way to lose.
Scope from signals, not from the ask
The client's stated request is the starting prompt, not the scope. Before estimating anything, a scoping conversation needs to extract the signals that actually drive effort:
- The decision behind the ask. What will they do differently when this engagement ends? Work that feeds a real decision has a natural boundary. Work commissioned "for visibility" expands forever.
- The state of their inputs. Who holds the data or access you need, and how many hands does it pass through? Every additional owner of an input is a delay you are about to buy.
- The approval path. Who says "done"? If the person paying and the person accepting are different people, scope to the stricter of the two.
- Prior attempts. If they tried this before — internally or with another firm — the failure of that attempt is the single most predictive fact about your effort. Ask what happened.
Forty-five minutes of structured discovery against these four signals beats a week of internal estimation debate, because the estimate's biggest error terms live on the client's side of the table.
Write the assumption ledger before the number
Between discovery and estimation, force one artifact: a plain list of every assumption the estimate depends on. Not prose — a ledger.
- Client provides read access to the delivery tracker by day 3. - One consolidated feedback round per deliverable, five business days max. - Workshop attendees have decision authority; no re-runs for absentees. - Historical project data exists in exportable form.
Each line does double duty. Internally, it converts a vague feeling of risk into countable items — an estimate resting on twelve fragile assumptions is a different bid than one resting on three solid ones. Externally, the ledger goes into the proposal verbatim, which changes the delivery conversation from "that's not in scope" (adversarial, after the fact) to "assumption four didn't hold, so here's the adjustment" (mechanical, pre-agreed).
The ledger is also where underbidding becomes visible before it happens. If you cannot write the assumptions down, you do not have an estimate — you have a hope with a currency symbol.
Estimate in ranges, price the boundary
For bespoke work, a single-point estimate is fiction, and sophisticated clients know it. Estimate honestly in a range, then make the range earn its keep:
- Anchor the low end to the ledger holding. "If the assumptions hold: 120 hours."
- Anchor the high end to the likely breaks. "If data access slips or feedback rounds double: 160."
- Price the commitment you can bound, phase what you can't. Fixed-fee the parts with decidable boundaries; run genuinely open-ended parts as a paid discovery phase whose deliverable is the scope for phase two.
That last move is the strongest anti-underbidding tool available. When a client will not fund discovery, they are showing you how the whole engagement will go — and a small paid diagnostic that produces a real scope is an easier first yes than a large bet on an ambiguous one.
Close the loop or repeat the error
Firms that underbid once usually underbid the same way for years, because nothing feeds back. The fix is unglamorous: at the end of every engagement, record estimated versus actual effort per assumption — which lines in the ledger held and which broke. Ten engagements of that data tells you your systematic bias (almost everyone underestimates approval latency and data cleanliness, roughly never the analysis itself), and your next range tightens for real reasons instead of vibes.
This is exactly the kind of synthesis that is tedious for humans and trivial for AI: turning discovery notes into a draft scope and ledger, checking a proposal against the ledger before it ships, and reading past engagement actuals into the next estimate. The judgment — where the boundary sits, what the risk is worth — stays with the partner. The bookkeeping that makes the judgment accurate is the machine's job.
Scope is the product in advisory work. Firms that treat it as a formality subsidize their clients; firms that treat it as an engineered artifact get paid for what they actually do.
Isn't underbidding just a pricing problem?+
No. By the time you set a price, the damage is usually done. Underbidding happens when the scope you priced and the scope the client heard are different documents. Fixing the number without fixing the boundary just changes how much you lose.
How detailed should a scope be for bespoke advisory work?+
Detailed enough that a third party could tell whether a given request falls inside or outside it. The test is not word count — it is whether the boundary is decidable. Deliverables, exclusions, and assumptions each need to be explicit.
What if the client refuses to give discovery time before the proposal?+
Price the uncertainty. A client who will not spend forty-five minutes clarifying the problem is telling you the engagement will run the same way. Either widen the range, phase the engagement so discovery is the first paid step, or decline.
Do effort ranges make proposals look unconfident?+
The opposite, when anchored to named assumptions. "120–160 hours, and here is exactly what moves it within that range" reads as a firm that has done this before. A single suspiciously round number reads as a guess — because it usually is.
Where does AI fit into scoping?+
In the synthesis, not the judgment. AI is good at turning discovery notes into a structured scope draft, checking a proposal against your assumption ledger, and flagging what past engagements say about similar work. The decision of where the boundary sits stays with you.
Related service
Want us to deploy this workflow for you?