Short answer. A professional services firm's best answers are already written, in its past proposals and policies. Nobody can find them fast. An AI assistant over those documents finds the closest past proposals, drafts from them and links every source. A partner reviews the draft and decides what goes out. The assistant runs in the EU and reads only the documents the firm approves.
This post covers how such an assistant works, what it must not do, and how to tell whether it is worth building. The numbers in the worked example are illustrative, not from a client project.
Why does every proposal start from a blank page?
For each new proposal, someone has to remember the closest earlier one, find it in old folders and check what still applies. The firm's knowledge sits in documents, but the index sits in a few people's heads.
So the search and the first draft take time, and the partner review starts late. When the person who remembers is busy or away, the firm writes from scratch what it has already written well.
How does an AI proposal assistant work?
Every request goes through the same four steps:
- Question. Someone asks about a past proposal, in the tool they already work in.
- Read. The assistant finds the relevant past proposals and policies and summarises them.
- Draft with sources. It drafts an answer with a link to every source it used.
- Review. The draft lands in the proposal tool. A partner reviews it and decides what goes out.
The assistant answers only from the firm's documents. Each statement links to its source, so a reviewer can check it in one click.
What must it not do?
- It drafts proposals but never sends them. A partner reviews every one.
- It reads only the documents the firm has approved.
- Price and scope stay a human decision.
These limits are not caution for its own sake. They are what lets partners trust the draft enough to use it.
What could the result look like? An illustrative example
Take a firm where finding the right material and writing a first draft takes about two days. This figure is illustrative.
With an assistant, the first draft of a proposal can be ready in about three hours, with every answer linked to its source. Partners spend their time on the parts that need experience. Finding and reassembling old text stops being their job. The firm chooses which documents the assistant may use and can see the sources behind every answer.
How do you know it is worth building?
- Count proposals per month and time the first draft of five recent ones.
- Check the archive. Are past proposals and policies in one place, in readable files? If not, that is the first job.
- Decide what is in scope. Which documents may the assistant read, and which must it never see?
- Name the reviewer. Who signs off every draft, and how fast?
- Settle hosting and GDPR before the build, not after.
The automation blueprint helps you write these answers down for your own process.
How would we approach it?
We time one proposal as the team writes it, from the first question to a draft ready for a partner. A 2-week fixed-price sprint maps how proposals get written and agrees which documents the assistant may read, where it runs and what "done" means, in hours and days. A 4-week pilot builds the assistant over the firm's past proposals and policies, hosted in the EU for GDPR, run next to the manual process with the drafts compared. Rollout opens it to the whole team inside their usual tools, with the partner review as the last step.
See AI integration and AI chatbots and voice agents.
FAQ
Will the assistant make things up?
It answers only from the documents the firm has approved, and every statement links to its source. If no source supports an answer, it should say so instead of guessing. The reviewer checks the original document before trusting the summary.
Is our client data safe?
Host the assistant in the EU, limit it to an approved set of documents and keep access rights the same as in your file system. Settle the GDPR basics first: your legal basis, where data is processed, and how long inputs and outputs are kept. Ask your data protection officer or counsel to confirm the setup.
Does it send proposals to clients?
No. It drafts. A partner reviews every draft and decides what goes out. Price and scope stay a human decision.
Do we need a new tool for the team?
No. The assistant works inside the tools the team already uses, so people ask it questions where they already write proposals.
Illustrative example. Company type and figures are illustrative, not from a client project.