The verdict in three sentences
A RAG assistant (retrieval-augmented generation) plugged into your precedents, opinions and internal notes pays off from 20 lawyers, provided the knowledge base is well organised. Legal privilege requires controlled hosting, zero-retention APIs and matter-level access control. Market tools such as Harvey are quick to roll out, while custom builds win when you want to exploit your own know-how and keep full control of the data.
What document search costs today
In a commercial firm in Paris or London, an associate spends on average 3 to 6 hours a week finding a clause precedent, a past opinion or an internal research note. That time is not always billable, and partners' know-how stays locked in matter files.
| Indicator (40-lawyer firm, 2026 order of magnitude) | Without assistant | With RAG assistant |
|---|---|---|
| Search time per lawyer per week | 3 to 6 h | 1 to 2 h |
| Hours recovered per year (40 lawyers, 45 weeks) | 0 | 3,600 to 7,200 h |
| Value at EUR 150 an hour (average internal rate) | 0 | EUR 540,000 to 1,080,000 |
| Time to find a precedent | 15 to 40 min | under 1 min |
| Reuse of existing opinions | low, memory-dependent | systematic, with cited sources |
| Onboarding a new associate | 3 to 6 months | 1 to 2 months |
Even if only 25% of these hours are actually converted into billable work, the gain far exceeds the project cost.
Market tool or custom assistant
| Criterion | Market legal AI tool (Harvey type) | Custom RAG assistant |
|---|---|---|
| Upfront investment | EUR 0 to 10,000 rollout | EUR 20,000 to 50,000 |
| Recurring cost | EUR 80 to 250 per lawyer per month | EUR 300 to 1,500 API + EUR 150 to 400 hosting per month |
| Annual cost for 40 lawyers | EUR 38,400 to 120,000 | EUR 5,400 to 22,800 |
| Use of your internal documents | partial, depends on connectors | full (DMS, shares, archived email) |
| Hosting | often multi-tenant or offshore | dedicated instance in your region |
| Matter-level access control | varies | mirrors your DMS permissions |
| Customisation (house style, templates) | low | full |
| Time to go live | 2 to 6 weeks | 8 to 14 weeks |
For 40 lawyers, a custom assistant usually pays back against per-seat subscriptions from year two, while keeping data under your control. London firms often combine both: a market tool for public law research, a custom assistant for internal know-how.
Confidentiality: the non-negotiables
| Requirement | Recommended implementation |
|---|---|
| Legal privilege and professional secrecy | no data shared with a third party without contractual safeguards |
| Hosting | servers in the EU or UK, encryption at rest and in transit |
| AI model APIs | contract with zero retention and no training on your data |
| Access rights | replicate DMS permissions, segregation by client and matter |
| Logging | audit trail of every query and document accessed |
| GDPR / UK GDPR | DPIA and updated record of processing |
| Conflicts of interest | automatic exclusion of matters behind ethical walls |
Mini case study
Laurent, innovation partner at a 40-lawyer firm in Paris, has 25,000 documents indexed (precedents, opinions, research notes). Custom project: EUR 38,000, then EUR 900 a month in API and EUR 250 in hosting, or EUR 13,800 a year. After 3 months, lawyers save 2.5 hours a week on average, 4,500 hours a year. Valuing only 20% of that time at EUR 150 an hour, the gain reaches EUR 135,000 a year. First-year cost: EUR 51,800. Payback in under 5 months.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
FAQ
Can the assistant invent case law?
The RAG approach limits that risk: every answer cites the internal documents used, with a direct link. Unsourced answers drop below 2% with strict settings, and the lawyer always checks the source.
Are our documents used to train the model?
No: the API contracts used exclude retention and training. Documents stay in your vector database hosted in your region.
Which DMS can we connect?
NetDocuments, iManage, SharePoint or a plain network share. A connector costs EUR 2,000 to 6,000 depending on the available API.
How long to index 25,000 documents?
Technical indexing takes 1 to 3 days; clean-up (duplicates, outdated versions) takes 2 to 4 weeks with a firm sponsor.
Should clients be informed?
A clause in the engagement letter or privacy notice is recommended, especially if client documents are indexed.
Let's scope your project. Tell us the number of lawyers, your DMS and document volume, and we will price the RAG assistant, its indicative budget, hosting and go-live timeline. 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.
