The problem
The cost of bidding was quietly deciding their growth strategy.
Each tender was two weeks and ~30 hours of senior time: read every requirement in a 40-200 page document, build the compliance checklist by hand, cost the delivery from scattered spreadsheets, assemble the pack.
The visible problem was speed, whoever answers first frames the deal. The hidden one was self-selection: because each bid cost two weeks, they only entered tenders they were already sure of. The paperwork was capping the pipeline.
The approach
Automate the retrieval, the costing and the formatting, not the judgement.
Phase 1: intake and requirement extraction from PDF and portal tenders into a structured compliance checklist, tuned to procurement language.
Phase 2: automated cost estimation for delivering the tender, pulling from the agency's own rate cards and past projects to price materials, labour and timeline.
Phase 3: an auto-assembled document pack in their template, compliance matrix, costed proposal, required attachments, with every figure traceable back to its source.
We used to pick which tenders to enter by how much work the bid would be. We didn't notice we were letting paperwork set our growth.
What was messy
Where the line between machine and human had to hold.
The temptation was to let the system write the win themes too. It can't, and shouldn't, the strategy for why this bid beats the field is exactly the part that's human. The system does retrieval, costing and formatting; a partner still owns the pitch.
Scanned and portal-only tenders needed OCR and per-portal handling before extraction was reliable, that was setup work, not a model limit.
Cost estimates were only as good as the rate cards behind them. Getting their historical pricing structured was the real first phase.
The outcome
More bids entered, without diluting the win rate.
- Tenders won (first 4 months live)$1.2M+client-reported
- Bid turnaround~2 weeks → ~1 dayfirst costed draft
- Bids entered, same team→ ~2.5×directional
- Win rate as volume grewheldthe real test
How we measured it
What's solid, and what's directional.
Tenders won: over $1.2M in the first four months live, client-reported. This is the figure the agency stands behind.
Turnaround: first costed draft in about a day versus roughly two weeks before, from the agency's own process timings.
Volume: roughly 2.5× more tenders entered with the same team. Average tender value ($250-350k) is a range, not an audited mean.
The real test was whether entering more bids dropped the win rate, it held. Treat that as directional over a four-month window, not a controlled study.
What we did not automate
The bid is still theirs.
Win strategy and final sign-off stay with a partner. The system drafts a costed, compliant bid; a human decides how to win it.
Pricing judgement stays human, the estimate is a starting point from their own data, not an auto-submitted number.
Nothing is submitted to a tender portal without a person reviewing and sending it.
What's next
Proof the engine travels beyond marketing.
For agencies reading this: the same machine, research → costing → first-draft proposal, is what we point at your new business or build into your stack for your clients. The vertical changes; the shape doesn't.
The tender agency is now extending the same pipeline to post-award project documentation.