--- title: "An AI-generated, unbranded executive brief the buyer's champion can circulate internally" type: case-study evidence_type: proof category: "AI & Automation in GTM" publisher: "LeanScale" date_modified: 2026-08-08 word_count: 997 topics: ["ai-in-gtm", "revenue-operations"] canonical_url: https://knowledge.leanscale.team/customers/ai-generated-champion-brief-at-sales-qualified-stage/ source: "LeanScale Knowledge Hub — https://knowledge.leanscale.team" license: "Free to quote and cite with attribution to LeanScale." --- # An AI-generated, unbranded executive brief the buyer's champion can circulate internally **Evidence type:** proof (what happened) A software vendor's deals kept stalling at the sales-qualified stage because the buyer's internal champion had nothing credible to send upward. We built a CRM-triggered generation pipeline that reads the deal's own call recordings and signal data and returns a short, unbranded business case written in the champion's voice, with the finished document linked back on the opportunity record. ## The challenge This company's product is rarely approved on a single person's say-so; the purchase needs cross-functional sign-off. Once a deal reached the sales-qualified stage, the internal champion had to build the case upward to their own leadership, and everything the seller could hand them read like vendor marketing. Deals sat in-stage waiting for internal sponsorship that never got assembled. The GTM technology lead wanted a short, credible document the champion could circulate as if they had done the analysis themselves, generated from what the seller already knew about that specific deal rather than from a generic template. ## The approach Trigger it from the CRM record, not from a new tool Reps never touch the AI platform. A checkbox on the Salesforce opportunity is the trigger; flipping it drops the record into a saved list view that the Clay workflow runs against. The generated document's URL is written back to a field on the same opportunity, alongside a review status and a sent date. The first design used a button firing a webhook, and the customer pointed out that a field toggle plus a list view removes the webhook entirely, so we simplified. This was deliberate and consistent: the engagement had already settled on a pattern of wrapping AI workflows as CRM-surfaced controls, with no direct platform access for sellers. A four-stage prompt chain, not one large prompt Stage one runs a research agent over the prospect's own public material to extract their language: the metrics they publish, the vocabulary they use, what they say they care about. Stage two analyses the recorded sales calls on that deal, pulling a headline plus the specifics behind it. Stage three synthesises the two into a 'why this is worth solving now' argument in the prospect's own terms. Stage four renders the result into a locked output structure. Splitting it meant each stage could be tuned and inspected on its own instead of debugging one opaque prompt. Reuse a grounding set already proven in production Rather than designing new inputs, the workflow mirrors the ingestion of an existing, already-validated pipeline-acceleration workflow: call recordings with topic trackers, notes from the development rep, account executive, solutions engineer and delayed follow-ups, third-party intent fields and competitive-context data, plus a live news agent for current context on the prospect. Critically, it writes to net-new fields on the opportunity rather than into the existing workflow's fields, so a new build could not corrupt a running one. Design the output so it does not sound like the vendor wrote it The customer locked a five-section structure (this problem exists here / it is worth solving now / here is what happens if we do nothing / independent evidence / path forward) and one hard rule: nothing about the vendor appears in the body. The document is unbranded by design and written in the voice of a champion making an internal case. A branded-versus-unbranded A/B test and a PDF variant were both cut to keep a single output path. Reps who want a branded version add the logo themselves after generation. Read the supporting evidence live rather than from a stale copy The brief closes on independent, publicly available evidence. The first plan was to download and pre-categorise that source so the model could ingest it. Because it is publicly hosted, the research agent instead reads it live at its address, so when a refreshed version is published at the same address the generated briefs pick it up with no rebuild. Point it at the deals that actually need it A saved list view isolates the target cohort: opportunities sitting in the sales-qualified stage past the organisation's own days-in-stage red threshold, with renewals excluded. Generation is manual and rep-initiated by design, because the seller is best placed to judge whether a brief will move that specific deal. Automatic generation is reserved for opportunities that stall past a day threshold in the earliest stages. The hard part was delivery, not generation Producing the copy was straightforward. Producing a formatted Word document that lives at a durable web address a CRM field can link to was not. The chain became: a workflow-automation tool receives the generated HTML, converts it to a formatted document, writes it into the company's own document store so it has an address, then writes that address back to the opportunity and notifies the rep. The HTML-to-document conversion step turned into a metered third-party API and a procurement conversation, which forced a genuine design review of whether the file had to live in the document store at all or could simply be delivered to the rep as an attachment. Cut scope twice, in the same direction The concept started as a three-page branded executive brief. It was cut to a one-page unbranded champion-voice document, then settled at roughly two pages after the customer trimmed the template themselves. Each cut removed build surface rather than just page count: dropping the branded variant deleted the A/B reporting tasks, the version tracking, and a separate content-owner review loop from the project. ## Outcomes Built and in rollout Reps trigger generation from the opportunity record; the finished document lands in the company's document store and its link is written back to the opportunity. As of the latest evidence the workflow was being completed end to end and put in reps' hands. This is a shipped-mechanism story, not a measured-results one. Scope shrank as the design got clearer Three project tasks were deleted outright and three closed as delivered when the branded-versus-unbranded test was cancelled. The re-scope landed before any of the affected build work had started, so the change cost no rework. The second use case was pulled by the customer, unprompted Sales development leadership asked to reuse the same call-listening-and-synthesis mechanism at a different point in the cycle, for qualified-but-not-yet-active meetings that need nurturing rather than a business case, and flagged that the prompt would have to be re-voiced because the same document reads differently coming from a development rep than from an account executive. ## Quotes > since we already are building this mechanism of like listening to the call and creating almost an executive summary based off of what was discussed and what we know we can help with, I'm almost thinking the SDRs can leverage this as well > > — Customer, Director, Sales Development > It might sound different coming from like an SDR versus the account executive. > > — Customer, Director, Sales Development > I'm shocked that there's no other path here. There's no other alternative to turn HTML into a Word document > > — Customer, Head of GTM Technology ## Canonical https://knowledge.leanscale.team/customers/ai-generated-champion-brief-at-sales-qualified-stage/