In a legal advisory tender, the technical proposal can count 3 times more than the price in the final score. It is also where, along with reading the consultation file, lawyers' hours that nobody bills disappear: here is how AI hosted in France takes over that work, and what your lawyers keep.
Why public tenders cost a law firm so much
Local authorities, hospitals, public institutions: public buyers purchase legal advice through competitive tendering. The French Public Procurement Code allows them to award these contracts under the adapted procedure (procédure adaptée), whatever the amount (article R2123-1, 3°). Representation in court falls under an even more flexible regime (article L2512-5, 8°).
The result: a large number of tenders, often as framework agreements with purchase orders, split into lots by area of law (employment law, public law, public procurement, company law). One example published in the BOAMP in 2026: 4 lots, including an employment law lot capped at €20,000 excl. VAT per year with 1 single contractor, and a score based 75% on technical merit and 25% on price. The technical proposal is what makes the difference.
Yet every response mobilises a whole chain: the person who monitors tenders and coordinates, the associates who draft, the partners who sign off. These are hours taken away from billable work. On top of that come the supplier panels of large groups, with procurement questionnaires that come back from one tender to the next.
What is specific to legal services tenders
The rules are those of every public contract: a consultation file (rules of the consultation, administrative clauses, technical clauses, price schedule), an application, a bid. The content, however, is specific.
- The technical proposal is about people and a method: the proposed team, the lawyers' specialisation, response times, the format of the advice, the understanding of the buyer's stakes.
- References are regulated: the buyer may ask for named references, provided the clients cited have given their prior and express consent (Conseil d'État, 6 March 2009, no. 314610).
- The practical case: some buyers ask for a note on a sample question. That is where the firm's expertise is judged.
- Pricing is simple: hourly rates per profile and a few fixed fees, to keep consistent from one response to the next.
- Conflicts of interest are checked before responding.
The journey of a tender, step by step
- Detection. A tool reads the notices on your platform every morning and keeps only those that fall within your areas of law.
- First sort. Each notice receives a score from 0 to 100 based on your criteria, written with you: area of law, amount, location, deadline. Above the score you set, the analysis is triggered.
- Full analysis of the file. The consultation file is read and checked in full. The output is a strategy sheet: criteria and weighting, documents to provide, required deadlines, penalties, risks, inconsistencies between documents. The questions to ask the buyer are already drafted.
- Decision. Your lawyers start from the sheet, bring their legal analysis and decide whether to go for it.
- Drafting. The technical proposal is drafted from your winning responses and your way of working, aligned with the buyer's criteria. The application is pre-filled and the price grid prepared.
- Review and expertise. Your lawyers review, adjust and bring the expertise. Every change is attributed to its author.
- File and submission. The file is assembled and checked: no document is missing. Submission is still done by the firm, for security.
- Lessons learned. A winning response enriches the base. A losing response triggers the right questions to the buyer about the scores obtained. The tool keeps what worked and proposes new rules for the next proposals.
What AI does, what your lawyers keep
AI takes everything that repeats: monitoring, sorting, reading the files, assembly, compliance checks, the first draft of the proposal. Your lawyers keep what requires their judgement: the decision to respond, the legal expertise, the practical case, the final sign-off and the submission.
Quality follows, because every response builds on the best the firm has already produced, and every correction improves the next one.
Step by step, what changes:
| Step | Without a tool | With AI |
|---|---|---|
| Detection | Platforms checked by hand | Notices read every morning, only your areas of law come up |
| First sort | Every notice read | Score from 0 to 100 based on your criteria |
| File analysis | Full reading by an associate | Strategy sheet, questions to the buyer already drafted |
| Decision | The lawyers decide | The lawyers decide, from the strategy sheet |
| Drafting | Proposal rewritten from old files | First draft aligned with the criteria, from your winning responses |
| Review | Back-and-forth by email | Every change attributed to its author |
| File and submission | Documents checked by hand | Automatic document check, submission by the firm |
| Lessons learned | Depends on the time available | Base enriched, new rules proposed |
What it represents: a worked example
Take a fictional firm that responds to 40 tenders a year, at 20 hours per response.
- Goal: divide that time by 4. That leaves 5 hours of review and expertise, i.e. 15 hours saved per response.
- In employment law, hourly rates range from €200 to €500 excl. VAT depending on experience and city. At an average of €300 excl. VAT, those 15 hours are worth €4,500 per response.
- Over 40 responses: 600 hours and €180,000 of billable time given back to the firm's practice every year.
Run the numbers with your own figures: responses per year, hours per response, average hourly rate.
Professional secrecy: the non-negotiable conditions
A response file contains sensitive information: the firm's methods, references, sometimes details about clients. Pasting a consultation file and your past proposals into a consumer assistant is out of the question. The framework we apply:
- A French model: analysis and drafting run on Mistral AI's models.
- Hosting in France: the library, the files and the tool are hosted by Scaleway. No data leaves the European Union.
- No training: your documents are not used to train any model and are never reused for another client.
- A confidentiality agreement before any exchange, with return or destruction of the documents on request.
- Named access and a trace of every change.
To go further on this framework: AI and Local Government: Why the Public Sector Demands a Sovereign Stack and our Security & Sovereignty page.
Where to start
- Measure: tenders per year, hours spent per response, success rate.
- Gather your material: templates, CVs, methodologies, won and lost responses, scores obtained.
- Test on a live tender, under a confidentiality agreement: analysis of the file and an outline of the proposal.
- Compare the time spent before and after on your next 3 responses.
The tool is built with your teams as part of our AI Partnership, from €499 excl. VAT per month with no commitment.
Conclusion
Legal services tenders are won on the quality of the technical proposal, and they cost lawyers' hours. A well-framed AI, hosted in France, handles the reading, the sorting and the first draft; your lawyers focus on the decision and the expertise. The right starting point: measure your current time and test on a live tender.
Sources
- French Public Procurement Code, articles R2123-1 and L2512-5
- BOAMP, notice no. 26-12221 (framework agreement for legal advisory services, 2026 to 2030): https://www.boamp.fr/pages/avis/?q=idweb:26-12221
- Conseil d'État, 6 March 2009, no. 314610 (named references in legal services contracts)
- Hourly rates of employment lawyers: https://www.swim.legal/blog/honoraires-avocat-droit-du-travail-budget-dossier-social-entreprise
