The workflow and its real cost
Research briefs and proposals share a pattern. Someone receives a question — a tender, a client request, an internal briefing ask — and has to assemble an answer from what the organisation already knows, add what is specific to this case, and get it checked before the deadline.
Most of the time goes on finding things: the right case study, the current version of the methodology description, the answer the team gave to a similar question last year. Then comes rewriting it to fit, and chasing experts to confirm it is still true.
The real cost includes:
- Expert time spent searching and rewriting rather than thinking.
- Stale claims — certifications, team sizes, client names or capabilities that were true once.
- Leakage risk — reusing a past proposal that contains another client’s confidential details.
- Lost opportunities when there isn’t time to respond well, or to respond at all.
Before estimating value, gather: how many briefs or proposals per month, hours per response split between searching, drafting and review, who reviews, how often content has to be corrected late, and where approved content currently lives.
Options
| Option | When it fits | Limits |
|---|---|---|
| Template and content library (no AI) | Low volume, stable content | Still relies on people finding and adapting the right piece |
| Search over approved content | Content exists but is hard to find | Finds sources; doesn’t draft |
| AI-assisted drafting with citations and review | Regular volume, good approved content, experts available to review | Only as good as the library; needs disciplined review |
| Don’t use AI | Few responses, mostly bespoke, or no approved content | Keeps current cost and pace |
The redesigned workflow, step by step
1. Build an approved content library
Before any drafting, curate the material the model may draw on: standard answers, methodology descriptions, case studies cleared for reuse, team biographies, policies and certifications. Each item has an owner, a review date and a classification. Anything containing another client’s confidential information is either anonymised and approved, or excluded.
Review point: content owners approve items into the library and re-approve them by their review date.
2. Qualify the opportunity — a human decision
Bid/no-bid, scope and pricing stay with people. AI can summarise a long tender document or extract the questions into a checklist, but the decision whether to respond, and on what terms, is a commercial judgement.
3. Retrieve past answers
For each question in the brief, retrieve the most relevant approved items. A retrieval setup over the library — whether a feature in tools you already own or a purpose-built assistant — should return the source document and section, not just text.
4. Draft with citations
The model drafts each answer only from the retrieved items and the specifics the team supplies, and it cites the source for every factual claim. Where the library has no support for something the question asks, the draft says so rather than filling the gap. Gaps are useful: they show where new content is needed.
5. Expert review
Review point: a subject-matter expert checks each section for accuracy, currency, relevance to this client and anything that over-promises. A named reviewer checks the whole response for commitments, pricing consistency and confidential information from other clients.
6. Feed back
Strong new answers go back into the library through the approval step, so the next response starts from better material.
Risks to manage
- Stale claims: enforce review dates in the library, and show each source’s date in the draft.
- Confidential information from other clients: keep it out of the library, not just out of the prompt.
- Over-promising: models tend toward confident, agreeable prose. Reviewers should look specifically for commitments on scope, timelines and outcomes.
- Unsupported claims: any sentence without a citation is checked or removed.
What to check and measure
- Every library item has an owner, a classification and a review date.
- Drafts cite a source for each factual claim, and reviewers can open it.
- Cycle time from receipt to submission, before and after.
- Review effort: time experts spend per response, and how heavily drafts are edited (a rough edit-distance or “light / moderate / rewrite” rating is enough).
- Late corrections and content errors found after submission.
- Win rate, with caution. It depends on price, relationships and competition far more than on drafting speed. Track it, but don’t credit or blame the AI workflow for changes in it without much more evidence.
Worked example (illustrative)
This is a hypothetical example, not a client result.
A consultancy answers around ten tenders a month. Bid managers spend much of each response searching old proposals and asking practice leads whether descriptions are still accurate.
It starts by curating a library of approved answers, with practice leads as owners and a six-month review cycle. Past proposals naming other clients are excluded unless the case study has been cleared. The bid team then uses an assistant that retrieves library items for each tender question and drafts answers with citations, marking questions where no approved content exists.
Practice leads now review drafts rather than writing from scratch, and the gaps list becomes a backlog of content to write. Bid/no-bid decisions and pricing stay with the partners. The team tracks cycle time and review effort monthly, and reports win rate separately without attributing changes to the new process.
Next step
Score your proposal or briefing workflow with the Workflow Opportunity Scorecard to check whether your content and volume justify the change. If you want the content library, drafting practice and review steps set up with your team, that’s what Team Workflow Activation is for.