Deal desk

A deal desk is the cross-functional function that reviews, structures, and approves deals falling outside standard terms — discounts past the rep’s authority, custom payment schedules, ramped or multi-year commitments, non-standard contract language, and scope the price book does not cover. It answers one question on a clock: can we sell it this way, at this price, and who says yes? Every company already makes those calls. A deal desk is the same call made by a named owner against a written policy with a turnaround SLA, instead of made in a Slack thread by whoever replies first.

A deal desk is not CPQ, and conflating the two is the costly mistake in this category. CPQ is software that encodes and enforces rules — valid configurations, price books, discount bands, approval routing. The deal desk is the human layer that writes those rules and decides the exceptions software cannot decide. It is also not a deal review: a deal review coaches the rep on how to win the deal, a deal desk rules on the commercial terms. And it is not headcount. Most companies should run a deal desk as a process for a year or more before anyone’s job title contains the words.

What it governs

A desk that tries to govern everything becomes a queue nobody respects. Scope it to the levers where a bad answer costs real money:

• Price and discount — anything past the band the rep can approve alone.
• Payment and billing terms — net-60 and beyond, annual-upfront exceptions, quarterly billing, ramps.
• Term structure — multi-year commitments, opt-outs, co-terminated add-ons, auto-renewal changes.
• Services and delivery commitments — implementation hours given away, dated go-live promises, custom SLAs.
• Contract risk — liability caps, indemnity, security and privacy commitments, anything that pulls in Legal.

Everything else routes to the standard quote path. A request touching none of these five should never reach the desk.

When you need one

The trigger is not company headcount — it is the rate of non-standard requests and who currently absorbs them. Two tests, both observable this quarter:

1. Frequency. Non-standard requests arrive weekly rather than monthly. Weekly means someone already does this job part-time without a title, and the work stays invisible until it slips.
2. Concentration. One person — the RevOps lead, the CFO, the VP Sales — is the single approval path, and their calendar is the constraint on quote turnaround.

Hit both and you need the process now. You do not need a hire. Run it as a standing 30-minute slot twice a week with RevOps, Finance, and the sales leader in the room, plus a written matrix so anything inside policy never needs the meeting. Dedicated headcount is justified when intake volume no longer fits that slot, or when the matrix has enough branches that “who approves this” is itself the question costing a day.

Waiting for headcount before installing the process is the common failure. The process is what makes the eventual hire productive on day one; without it, the first deal desk analyst spends a quarter reverse-engineering decisions from closed-won records.

The approval matrix

A matrix names what triggers review, who approves in each band, what inputs are required, and what happens when concessions stack. Vendor-published trigger examples cluster around deal value in the $50K-$500K range and discounts past 20 percent (DealHub), but those are illustrative bands, not your bands.

Calibrate the bands off your own closed-won data, not this table. Pull the last two quarters of non-standard deals, plot discount against win rate and against realized margin, and set the first band where win rate stops improving. If discounting past a threshold has not moved win rate, that threshold is where approval should start costing something.

The rule mattering more than the bands: concessions compound. A 15 percent discount is one decision; 15 percent plus a ramp plus 40 free implementation hours is a different deal, and a matrix scoring each lever independently will approve it one lever at a time. Score the package.

The turnaround SLA

A desk without a clock becomes the thing sales routes around. Publish response and decision windows separately, and lane them by complexity — same-business-day decisions for the fast lane, one to two business days for complex packages, and a named interim checkpoint for executive exceptions rather than silence. Enforce aging: at 80 percent of the SLA elapsed, the request escalates to the approver’s delegate automatically.

Measure first-pass yield — the share of submissions arriving with every required input — and report the median alongside the 90th percentile. Averages hide the deals that sat for a week, and those are the ones reps remember at quarter end.

Where it reports

Reporting lines vary across CFO, CRO, VP Sales, and RevOps. RevOps is the defensible default because the desk carries two mandates pulling against each other — velocity and governance — and a desk owned by Sales approves too much while a desk owned by Finance approves too slowly. RevOps ownership keeps the escalation path to both.

What AI changed, and what it did not

Quote assembly has genuinely moved. Salesforce shipped Agentforce for Revenue inside Revenue Cloud in July 2025, where a rep describes a quote in natural language and an agent generates it with the correct products, pricing, and terms; Salesforce reports a 75 percent drop in quoting time and an 87 percent reduction in clicks on its own internal sales team (vendor self-reported, 16 July 2025). Deloitte data cited in the same announcement puts 71 percent of B2B executives struggling with manual, fragmented sales processes and 13 percent of deals lost to disconnected tools.

What has not moved is the exception. An agent can assemble the quote, check it against the price book, route it to the right approver, and draft the margin summary. Deciding whether to trade 5 points of margin for a three-year term on a logo you want in the case study is a judgment about strategy, and no production deal desk in 2026 delegates it. Budget AI against intake quality and cycle time — first-pass yield is where it shows up — not against the approval decision itself.

Common pitfalls

• The desk becomes a rubber stamp. An approval rate near 100 percent means thresholds sit too high to catch anything. Guard: review approval rate by band quarterly; a band approving everything gets a lower trigger or gets deleted.
• No documented precedent. The same customer request gets a different answer twice, and reps learn to shop approvers. Guard: log every exception with its rationale on the CRM opportunity, and check precedent before deciding.
• Intake with no required fields. Reps submit “can we do 30 percent?” with no margin model, and the desk spends its cycle time chasing inputs. Guard: a submission form rejecting incomplete packages, so the SLA clock starts only on a complete request.
• Scope creep into deal coaching. The desk starts advising on how to win, the meeting doubles in length, and terms decisions queue behind strategy talk. Guard: keep deal reviews a separate forum with a separate agenda.
• Thresholds set once and never revisited. Pricing changes, segment mix shifts, and last year’s bands route the wrong deals. Guard: re-calibrate against closed-won data every two quarters.

Related

• CPQ (Configure, Price, Quote) — the software layer enforcing what the desk decides
• Revenue leakage — what unmanaged exceptions cost downstream
• Deal stage definitions — the forecast data a governed desk protects
• Deal desk pricing skill — a Claude Skill for structuring the margin model a submission needs