The situation
A 160-person architecture-engineering firm with a heavy federal book — GSA design contracts, USACE district work, a growing NAVFAC backlog. They were winning enough to grow, but each proposal felt brittle: a 14-day scramble that pulled five senior staff for the final 72 hours, with last-minute formatting errors and a non-zero failure rate on compliance review.
The firm's proposal lead was openly worried about burnout. They'd already lost one senior PM the year prior, partly attributable to the proposal cycle. Leadership knew the talent risk was real.
What we built
One core tool, two supporting workflows:
- Proposal writer trained on the firm's library — 200+ past submissions ingested with the metadata that mattered (project type, contract vehicle, evaluation factors). The system drafts sections in the firm's voice, with citations that point back to the source projects.
- Compliance checker — runs every draft against the solicitation's L/M sections and surfaces gaps before principal review.
- Resume formatter — pulls staff records and assembles section H bios in 5 minutes instead of an afternoon.
How the work ran
We started with a one-week data sprint — getting the firm's past submissions into a clean, queryable shape. This was where 60% of the value got created. The proposal lead had been the human index for years; we replaced the index with structured metadata and a search system anyone on the team could use.
The build itself took four weeks, with two sprints. Sprint one shipped the proposal writer; sprint two added the compliance checker and the resume formatter. We ran one live federal pursuit in week six as a supervised pilot — the team submitted in 4 days with zero compliance flags.
We've been told the new submissions read more like our best work, not our average work. That's the part I wasn't expecting. — VP, Federal Business Development
Outcomes after four months
- Average cycle dropped from 14 days to 3 days.
- Compliance pass rate rose from 71% to 92% on first internal review.
- Senior staff hours per proposal dropped 64%; the team voluntarily reduced overtime.
- Proposals submitted rose 41% over the same period, without adding head-count.
What we'd say to a similar firm
Federal proposals look like a content problem and turn out to be a metadata problem. The firms that get fast first invest in the boring layer: structured tagging of past projects, named evaluation factors, normalized resume data. The AI work is comparatively cheap on top.
If you're scoping something similar, plan to spend the first three weeks on data and one week on prompts — not the other way around.