Reusable Case Study Drafter

The useful job is a first draft of an anonymized case, pulled from the close pack, so BD is not writing from memory six months later. Speed is the point. The cost of that speed is a paragraph that sounds like a win and cites a result nobody measured.

Do not invent outcomes or metrics. If the steering pack, the final report, or the client's own close letter never stated a figure, the case cannot invent one. "Cycle time improved" is allowable only when that phrase, or a measured equivalent, already exists in approved outcome language. A generated percentage you cannot source is a claim you should not put in front of a buyer, a journalist, or your own partners.

Nothing leaves the team until two things are true: the client has consented under the engagement terms, and a human has stripped identifiers a sector peer would use. Anonymization is not a synonym for deleting the logo. A sector, size, and geography combination can name the account as surely as the legal entity.

Treat the model output as an internal draft. Proposal platforms in the AutogenAI and Responsive class, and knowledge stores in Notion or SharePoint, are where a cleared case lives later. They are not a reason to skip consent.

Draft from artifacts, not from the team's favorite story of the work. The source pack at close should be short and dated:

• The signed SOW or equivalent, so the problem statement matches what was actually sold.
• Final deliverables the client received: report, playbook, operating model, implementation pack.
• Outcome language the client already approved: steering readout, benefits tracker, or a written close letter. If they never signed off on a number, there is no number.
• Method artifacts that are yours to reuse, after a reusable IP extraction agent has separated firm templates from client-owned maps.

Do not feed the model the lessons learned synthesizer dump as if it were a case. A postmortem is an internal conversation. Blame, named misses, and "what we would not do again" do not belong in BD copy. Hints from a follow-on opportunity signal detector are a partner conversation, not a proof point.

Prompt the draft to a structure proposals actually reuse: situation, what you were asked to do, what you did, what changed, and what is still client-confidential. Map each outcome sentence to a source file and a page or slide. If the model cannot point at a source, delete the sentence.

The failure mode here is invention. Models complete a benefits paragraph the way they complete any paragraph. "Reduced working capital," "cut pick time," and "unlocked capacity" appear because those phrases live in other cases in the library, not because this engagement measured them. A partner who was on the job should kill any claim they could not defend in an oral.

This is a worked example with made-up firms, written to show the cuts, not a measured result.

Pellam Advisory, a 180-person operations boutique, has just closed a 14-week slotting and labor-standards engagement for Harborline Logistics, a regional 3PL with a dense Newark campus. The engagement manager dumps the final report, the week-12 steering pack, and a SharePoint folder of workshop photos into a drafting prompt. BD wants a case on the site by month-end.

What the first cut claimed

The draft read clean. Situation: a multi-site 3PL with aging slotting rules. Approach: ABC slotting, labor standards, and a "control tower" for exception management. Outcome: pick time "down 34 percent in 12 weeks," plus a color piece about "the night shift in Building 7, where associates had been walking past the same dead slots for years."

What the close pack actually said

The steering pack tracked lines per hour on two aisles in one building, for six weeks, with no baseline the client would stand behind in writing. Nobody had measured pick time across the campus. "Control tower" was an internal nickname for a daily huddle; the client's close letter said "stand-up meetings." Building 7 and Newark are how a competitor names Harborline. The workshop photos still showed Harborline vests and dock numbers.

The team loved the control-tower story. It was the slide they presented at the internal win lunch. A later buyer interview on a lost 3PL bid, the kind of debrief win/loss pattern synthesis is for, never mentioned control towers. The evaluator had asked for a named diagnostic in the proposal and for labor-standards proof from a client who would take a reference call.

What survived

They rewrote from the close letter: problem (slotting rules that had not been refreshed), approach (ABC plus labor standards, named methods already in the report), and outcome (the two-aisle lines-per-hour observation, labeled as limited and not a campus result). They dropped the invented percentage, Building 7, Newark, the night shift, the photos, and the control-tower brand. The draft stayed inside the team until client legal answered the publicity clause.

That is the pattern. The model is fast at narrative. It is not a witness.

Anonymization is a partner job with a checklist, not a find-and-replace on the client name.

Strip, then have someone who was not on the job try to identify the account:

• Legal names, logos, people, email domains, and "as we did for [company]."
• Sites, building numbers, city-plus-sector combinations, and unique facility layouts.
• Org charts, role titles that only one buyer uses, and process names that are internal jargon.
• Exhibits, photos, and screenshots. A redacted slide that still shows a unique KPI dashboard is still theirs.
• The combination test: "regional 3PL, Northeast, PE-backed, three sites" can be one company. Broaden or drop attributes until a practice lead cannot guess.

Then stop. Client legal review is not optional because the draft is anonymized. Many engagement letters require written consent for any publicity, named or not. Send the stripped draft, the intended use (website, proposal appendix, oral credentials), and the list of remaining attributes. If they refuse a number or a method detail, it is out. If they refuse the case, it does not go to BD. Do not anonymize harder and publish anyway.

A partner who worked the job should sign that they would say every remaining sentence in an oral. A second partner who did not work the job should sign that they cannot name the client. Either no is a halt.

BD reuse starts after consent, not after the first fluent draft. File the approved version with as-of date, industry, problem type, and a note on what the client allowed: named, anonymized, referenceable, or proposal-only. Knowledge bases in Notion or SharePoint are the usual library. Proposal platforms in the AutogenAI and Responsive class can retrieve from that library when a questionnaire asks for credentials. None of these tools decides what is true. They will happily retrieve a beloved story that a buyer never treated as proof.

Gate reuse with win/loss pattern synthesis. A case the team loves and the buyer never mentioned is not evidence. If interviews in that segment keep citing "named diagnostic" and "labor-standards proof," those are the cases that should surface. A control-tower narrative that only the delivery team retells should stay off the RAG proposal draft generator index, even if it is a beautifully written page.

Keep losses, drafts, and uncleared internals out of the credentials index. A generator that mixes an uncleared Harborline anecdote into the next 3PL bid is a leak with a cover sheet.

Success on this use case is elapsed time from close to a cleared, reusable case, plus zero invented metrics and zero identifiable leftovers in anything that left the team. Failure is a fluent page that a client, or a buyer's counsel, can still recognize.