The verdict in three sentences
An AI document research assistant (RAG) connected to the knowledge base of a 60-lawyer firm costs EUR 25,000 to 60,000 (about GBP 21,000 to 51,000) to build in 2026, plus EUR 500 to 2,500 per month for compute and hosting. The non-negotiable condition is hosting in the UK or EU with access partitioned by team and matter, to protect legal privilege and handle conflicts. With 3 hours saved per associate per week, the investment pays back within months.
Why RAG rather than a general chatbot
A consumer tool does not know your internal opinions and must never receive them. RAG (retrieval-augmented generation) indexes your own documents, retrieves the relevant passages and drafts an answer with clickable citations to the source. The lawyer checks in one click instead of digging through the DMS.
| Criterion | Consumer chatbot | Publisher case-law database | Internal RAG assistant |
|---|---|---|---|
| Access to your internal opinions | No | No | Yes, 15 years of content |
| Citations to source | Variable | Yes (public sources) | Yes (your documents) |
| Hosting | Often outside UK/EU | UK/EU | UK/EU, tailored |
| Partitioning by team and matter | No | Not applicable | Yes, inherited from DMS |
| Annual cost for 60 lawyers | EUR 15,000 to 25,000 | EUR 30,000 to 80,000 | EUR 6,000 to 30,000 of compute |
| Privilege risk | High | Low | Low if well designed |
The two are complementary: the publisher database remains the reference for public case law, the RAG assistant unlocks the firm's own know-how.
What makes up the EUR 25,000 to 60,000 budget
A typical 60-lawyer firm holds 80,000 to 200,000 documents (opinions, memos, precedents, research notes), often in iManage, NetDocuments or a file server.
| Item | 2026 range (excl. tax) | Comment |
|---|---|---|
| Scoping, content mapping and DPIA | EUR 3,000 to 6,000 | With the DPO and the responsible partner |
| DMS connector and extraction (OCR of scanned PDFs) | EUR 5,000 to 12,000 | More if digitised paper archives |
| Indexing, chunking and optional anonymisation | EUR 4,000 to 10,000 | Metadata: practice area, client, date, author |
| Team rights and ethical walls | EUR 5,000 to 12,000 | Sync with Microsoft Entra directory |
| Search interface and Word or Teams integration | EUR 4,000 to 12,000 | Citations, filters, note export |
| Quality evaluation and training | EUR 4,000 to 8,000 | 200-question test set, 2 sessions |
| Total | EUR 25,000 to 60,000 | 8 to 14 weeks |
Monthly cost depends on query volume and the model chosen. For 60 users asking 20 to 40 questions a week, count EUR 500 to 1,200 with a model hosted in-region through a major cloud provider, and up to EUR 2,500 with a model deployed on dedicated servers. London legaltech vendors with similar scope usually quote GBP 40,000 to 120,000 per year in licences.
Privilege and compliance: the checklist
SRA rules and UK GDPR require strict confidentiality of client information. In practice, we require:
- hosting and processing in the UK or EU, with no reuse of data for model training (written contractual clause);
- encryption at rest and in transit, logging of every query for 12 months;
- inheritance of DMS rights: a lawyer never sees a matter they could not open;
- ethical walls enforced in the index, not only in the interface;
- a signed DPIA before going live.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
James, managing partner of a 60-lawyer commercial firm in the City of London, estimates that his 45 associates spend on average 3 hours a week looking for an opinion or precedent already written.
- Time recovered: 45 × 3 h × 45 weeks = 6,075 hours per year.
- Conservative value at EUR 60 per hour (internal cost, not billing rate): EUR 364,500 per year.
- Assistant cost: EUR 45,000 to build + EUR 1,200 × 12 = EUR 14,400 per year.
- Even if only half of the time saved is truly reused, the project pays back in under 4 months.
FAQ
Are our documents used to train the AI?
No. The contract with the model provider must exclude any reuse, and the index stays in your UK or EU environment. We provide the clause and the GDPR record of processing.
What error rate should we expect?
On a 200-question test set, a well-tuned RAG reaches 85 to 92% correct, sourced answers. Each answer shows its sources: the lawyer always validates before use.
Can we start with one department?
Yes, and we recommend it. A pilot on employment or corporate, with 10,000 to 20,000 documents, costs EUR 15,000 to 25,000 over 6 weeks.
Does it work with iManage or NetDocuments?
Yes, both expose APIs. A Windows file server also works, with an extra EUR 2,000 to 4,000 to rebuild permissions.
Do we need to inform clients?
A mention in the privacy notice and engagement letter is recommended. Since data does not leave the firm's control or region, no specific consent is normally required.
Let's scope your project. Describe the size of your knowledge base, your DMS and pilot teams: we quote a RAG assistant between EUR 25,000 and 60,000, delivered in 8 to 14 weeks. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
