Eighty tender submissions a year, each drafted from their own library of past wins
A marine engineering and contracting firm bids for work on wharves, jetties, bridges and pontoons. To win it, clients put them through two kinds of forms: an EOI (expression of interest) first, to prove they're qualified and get shortlisted, then a full RFT (request for tender) with pricing, methodology and compliance schedules once shortlisted. That's roughly 80 of these a year, around 150 pages each, all carried by one person, and most of what goes into them, they have already written down somewhere in a past submission. We built them a Claude project that drafts each new one from their own library of past wins.
The problem
Every RFT or EOI comes on the client's own form, 80 a year at around 150 pages each, all on one person. Most of it is not new writing. It is company history, project references and methodology they have already won with, reformatted by hand into whatever structure this client uses. The states make it worse, each with its own forms, licensing and safety rules.
Past submissions
Every previous RFT and EOI response they have already written and won with.
Their existing libraryNew form arrives
The client's own RFT or EOI for the next opportunity.
Uploaded as-is- Automated
One new submission
Each section drafted by pulling the matching pieces from past documents.
Claude Projects
The process
They did not want a tool that owned their content or locked them in, so instead of building software we set them up in Claude, on their own documents. The real work was the library. Claude can only reuse what it is given, and their folders were full of blank forms, duplicates and dead files. We curated it down and renamed everything, because that is the difference between reusing a winning submission and pulling from the wrong file.
- A dedicated Claude project
- Set up to read a new RFT or EOI form and return copy-ready text for each section, drafted from their past wins, with the source named for every answer.
- A curated library
- Their proven past submissions, cleaned of duplicates and dead files, so Claude finds the right precedent instead of pulling from the wrong file.
- Built state-aware
- Instructions written in once that know how QLD, NSW and WA differ in forms, licensing and the safety regulator.
- Built to flag gaps
- Says so when there's no precedent for a section, rather than inventing something plausible and wrong.
The result
The repetitive 80 percent of a submission now arrives as a draft from proven, winning content, in the client's structure and for the right state. And because it is just their own documents plus plain-language instructions, it stays theirs, and ports with them if they ever move to Copilot.
Built with
- Claude
- Claude Projects
- Project knowledge
- Project instructions
- Project memory
- Artifacts
- Extended thinking
Questions we get asked about this
Why not just buy a dedicated tender tool?
Because a proposal platform owns your content and charges you for it, and a custom build leaves you depending on the developer. Here the knowledge base is your own documents and the behaviour is instructions you can read and edit. All of it ports if you switch assistants later.
Does it write and submit the tender for us?
No, and it should not. It drafts the repetitive sections from your proven past submissions, and names the documents each answer came from so you can check it. Pricing, judgement and final sign-off stay with you.
We bid across different states. Can it handle that?
Yes. One instruction set names what changes between states, the legislation, licensing and safety regulator, and matches those and your project references to where the work will be performed.
What about a section we have never answered before?
It flags it. Where there is no precedent, the draft carries a clear needs-input marker with the reason, so the gap is visible rather than filled with something plausible and wrong.
Do we have to be a big firm for this to pay off?
No. This one runs on a single person doing 80 submissions a year. It comes down to repetition, and headcount does not matter. If the same content keeps being rewritten into different formats, it is worth setting up.
