The verdict in three sentences
An internal RAG assistant (retrieval-augmented generation) on the knowledge base of a 60-consultant firm costs EUR 20,000 to 45,000 excl. VAT to build and EUR 300 to 1,200 per month to run in 2026. The real challenge is not the language model but indexing quality and per-engagement access control, because a consultant must never see deliverables from a client that is not theirs. With 2 to 4 hours saved per consultant per week, the project usually pays for itself in under six months.
What a RAG assistant actually does in a consultancy
A partner at a Berlin consultancy holds 15 years of proposals, methodologies, steering committee minutes and lessons learned, scattered across SharePoint, network drives and mailboxes. A consultant preparing a proposal often spends half a day looking for "that similar engagement from 2021". The RAG assistant indexes these documents, splits them into passages, then answers questions while citing its sources: "What savings assumptions did we use in our procurement engagements for automotive suppliers?" The answer comes back with links to the three relevant deliverables.
| Component | Role | 2026 range (EUR excl. VAT) |
|---|---|---|
| Scoping and document audit | Source inventory, typology, confidentiality rules | 2,500 to 5,000 |
| Connectors (SharePoint, Drive, DMS) | Extraction and incremental sync | 3,000 to 8,000 |
| Indexing and chunking | OCR of scanned PDFs, client, engagement and date metadata | 4,000 to 9,000 |
| Hybrid search engine | Vector and full-text search, re-ranking | 3,500 to 7,000 |
| Per-engagement access rights | Permission inheritance, filtering before generation | 3,000 to 8,000 |
| Interface (Teams or web) | Chat, clickable citations, history | 2,500 to 5,000 |
| Evaluation and acceptance | Set of 150 to 300 test questions, accuracy measurement | 1,500 to 3,000 |
The low end, EUR 20,000, covers a single clean document source and a simple web interface. The high end, EUR 45,000, includes several sources, scans requiring OCR and a Microsoft Teams integration with single sign-on.
Monthly running cost, EU hosting and the AI Act
Recurring cost depends on question volume, the model chosen and the hosting mode. For a firm handling industrial client data, EU hosting is almost always a contractual requirement, German clients frequently add GDPR data processing clauses, and the AI Act requires minimal documentation of the use case.
| Monthly item | Light use (20 active users) | Heavy use (60 users) |
|---|---|---|
| Language model calls (EU API) | EUR 120 to 250 | EUR 400 to 700 |
| Vector database and storage | EUR 50 to 100 | EUR 100 to 200 |
| Application hosting (Frankfurt or EU region) | EUR 60 to 120 | EUR 120 to 200 |
| Monitoring and updates | EUR 70 to 130 | EUR 100 |
| Indicative total | EUR 300 to 600 | EUR 720 to 1,200 |
Three hosting options exist: a major provider's API with an EU data residency commitment, an open-source model hosted by a European provider, or a model deployed on your own servers. The last option adds EUR 8,000 to 15,000 to the build and is only justified for classified data or defence clients.
Timeline and pitfalls
| Phase | Duration | Deliverable |
|---|---|---|
| Scoping and access rules | 2 to 3 weeks | Source map, rights matrix |
| Connectors and indexing | 3 to 5 weeks | Indexed corpus, verified metadata |
| Prototype and internal tests | 2 to 3 weeks | Pilot for 10 consultants |
| Adjustments and rollout | 2 to 4 weeks | Firm-wide launch, training |
| Total | 2 to 4 months | Assistant in production |
The three most expensive mistakes: indexing without cleaning (duplicates and outdated versions produce contradictory answers), forgetting access rights (a direct professional ethics risk), and not measuring accuracy before launch. A reference question set lets you target more than 85% of answers rated correct and sourced before rollout.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Lena, partner at an operations consultancy in Berlin, has 60 consultants with an average internal hourly cost of EUR 95. The project is quoted at EUR 34,000 excl. VAT, plus EUR 800 per month. Conservative assumption: 2 hours saved per consultant per week, over 44 working weeks. Annual gain: 60 × 2 × 44 × 95 = EUR 501,600 of freed time. Even if only 15% of that time turns into billable work or won proposals, that is EUR 75,240 per year, against a first-year cost of 34,000 + 9,600 = EUR 43,600. Break-even lands around month seven, and later years only cost usage and maintenance.
FAQ
How much does a RAG AI assistant cost for a consultancy in 2026?
Expect EUR 20,000 to 45,000 excl. VAT to build, depending on the number of sources and the complexity of access rights. Monthly usage runs between EUR 300 and 1,200 for 20 to 60 active users.
Do our client documents leave the European Union?
No, if the architecture is designed for it: an API with EU data residency or a model hosted in Germany or elsewhere in the EU. Serious providers contractually guarantee no training on your data, which you should check in the contract.
How do we stop a consultant seeing another client's deliverables?
SharePoint or DMS permissions are inherited by the index, and filtering applies before the model reads a single passage. This module costs EUR 3,000 to 8,000 and should never be cut.
How long until a first pilot?
A pilot on one document source with 10 consultants is realistic in 6 to 8 weeks. Full rollout takes 2 to 4 months.
Does the AI Act apply to an internal assistant?
An internal knowledge assistant generally falls under limited-risk uses, with transparency and usage governance obligations. An internal note of 3 to 5 pages describing purposes, data and controls is usually enough.
Let's scope your project. Send us your list of document sources and your consultant headcount: we will price a RAG scope between EUR 20,000 and 45,000 with a pilot in 6 to 8 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.
