Every rep priced deals a little differently, and all of them priced under pressure. Quotes were built by hand in spreadsheets, cloned from whatever similar deal someone remembered, and then held for days in an approval queue while a manager tried to reconstruct the reasoning. Margin leaked in the gaps.
The problem, precisely
A pricing copilot is easy to demo and hard to trust. If it produces a number the rep can’t defend, they’ll quietly ignore it and go back to the spreadsheet — and now you’ve spent six figures on a tool nobody opens. The constraint here wasn’t accuracy in the abstract; it was legibility: could a rep see why the price was what it was, and could a manager approve it in one glance?
The approach
We split the problem in two. Retrieval surfaces genuinely comparable historical deals and their outcomes; a deterministic guardrail layer enforces the pricing rules that are non-negotiable. The model proposes and explains — it never gets to silently override a floor.
A recommendation a salesperson can’t explain to a customer is worse than no recommendation at all.
What we built
# quote flow
match → retrieve comparable historical deals
propose → suggested price + margin, with rationale
guardrail → clamp to policy floors and ceilings
explain → side-by-side comps shown in the CRM
approve → one-click manager sign-off + audit trail
The whole thing lives inside the CRM they already use. No new tab, no new login — the copilot meets reps where the work already happens.
How we knew it worked
Adoption is the only honest metric for a tool like this, so we watched it weekly and read the ignore cases. Above 80% of reps were using it inside six weeks, approval cycles roughly halved, and blended margin recovered 2.3 points — with a full audit trail behind every number.