The verdict in three sentences
An internal RAG assistant connected to your customer service documentation costs, as a 2026 order of magnitude, between EUR 15,000 and 35,000 excl. VAT to build, plus EUR 300 to 1,200 per month in API and hosting costs. Properly scoped, it cuts average handling time by 25 to 35 % because agents stop hunting for the right page across 2,500 pages of documentation. Accuracy depends less on the model you pick than on document quality, source citations and access rights management.
What a customer service RAG assistant really includes
RAG (retrieval-augmented generation) first queries your knowledge base, then has a language model write the answer using only the retrieved passages. For a software publisher in Amsterdam with 40 agents, that means indexing help articles, release notes, internal procedures and resolved tickets, then showing a suggested answer with links to sources inside the ticketing tool.
Three building blocks separate a prototype from an adopted tool: smart document chunking (by section, with headings and versions), rights-based filtering (a tier 1 agent does not see tier 3 procedures) and continuous evaluation on a set of 150 to 300 real questions.
| Item | 2026 range (EUR excl. VAT) | Comment |
|---|---|---|
| Scoping and documentation audit | 2,000 - 4,000 | Source inventory, duplicates, outdated pages |
| Indexing pipeline (connectors, chunking, vectors) | 4,000 - 9,000 | Confluence, Zendesk, Notion, PDF |
| Interface inside the ticketing tool | 3,000 - 8,000 | Side panel, clickable citations |
| Rights management and logging | 2,500 - 6,000 | SSO, roles, GDPR traceability |
| Evaluation set and tuning | 2,000 - 5,000 | 150 to 300 scored questions |
| Training and rollout | 1,500 - 3,000 | 2 to 3 sessions, internal guide |
| Total project | 15,000 - 35,000 | 6 to 8 weeks |
Running costs and accuracy: the numbers to watch
Running costs depend on question volume and the model used. With 40 agents asking about 25 questions a day each, you get roughly 20,000 requests a month. Depending on the model and passage length, the API bill sits between EUR 150 and 700 a month, plus the vector database, hosting and monitoring.
| Indicator | Before assistant | After 3 months (order of magnitude) |
|---|---|---|
| Average ticket handling time | 11 min | 7.5 - 8.5 min |
| Documentation search time per ticket | 4 min | 1 - 1.5 min |
| Share of answers with a cited source | not measured | 85 - 95 % |
| Share of answers rated correct (evaluation set) | n/a | 80 - 90 % |
| Ramp-up time for a new agent | 8 weeks | 5 - 6 weeks |
| Monthly API + hosting cost | EUR 0 | EUR 300 - 1,200 excl. VAT |
Accuracy can be managed: an assistant that says "I cannot find this information in the documentation" is better than one that makes things up. You set a relevance threshold below which the tool routes to an expert. Agents rate each answer (thumbs up or down), which feeds corrections to the documentation itself.
Custom build, ticketing add-on or specialised SaaS
Ticketing vendors now offer built-in AI modules, billed per agent or per resolution. They work if all your documentation already lives in the tool. As soon as sources are scattered (internal wiki, technical PDFs, release notes in Git), a custom solution regains the edge.
| Option | Indicative 2026 cost | Strengths | Limits |
|---|---|---|---|
| Ticketing AI add-on | EUR 30 - 60 per agent/month or EUR 0.80 - 1.50 per resolution | Fast setup | Sources limited to the tool |
| Specialised RAG SaaS | EUR 500 - 2,500/month | Ready-made connectors | Data outside your control, rising cost |
| Custom build | EUR 15,000 - 35,000 + 300 - 1,200/month | Fine-grained rights, multiple sources, reversibility | Upfront investment |
For 40 agents, the add-on at EUR 45 per agent comes to EUR 21,600 a year, with no control over external sources. A custom build typically pays back in 12 to 18 months and remains your property.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mini case study
Femke, head of customer service at a software publisher in Amsterdam, manages 40 agents handling 9,000 tickets a month. Average handling time drops from 11 to 8 minutes after rollout, saving 3 minutes per ticket: 9,000 x 3 = 27,000 minutes, about 450 hours a month. At a loaded cost of EUR 32 an hour, the theoretical saving reaches EUR 14,400 a month. Even counting only half of it (part of the time is reinvested in quality), EUR 7,200, a EUR 28,000 project plus EUR 800 a month in running costs pays back in just over 4 months.
FAQ
How long does it take to deploy a customer service RAG assistant?
Allow 6 to 8 weeks for a base of 2,000 to 3,000 pages, including 1 to 2 weeks of documentation audit. A pilot with 5 to 10 agents over 2 weeks comes before full rollout.
Can the assistant make up answers?
The risk exists, but it drops sharply with mandatory citations and a relevance threshold. On a well-built evaluation set, the target is 80 to 90 % correct answers and an explicit refusal for the rest.
Does our customer data leave the company?
Passages sent to the model can be limited to documentation, with no personal data. You pick a provider with EU hosting and a no-training commitment, and log 100 % of requests for GDPR audits.
What monthly budget should we plan after go-live?
Between EUR 300 and 1,200 excl. VAT a month for 10,000 to 40,000 requests, including hosting and monitoring. Also plan 2 to 4 days per quarter to refresh the index and the evaluation set.
Should we clean up the documentation before the project?
Yes, it often accounts for 20 to 30 % of the final gain. Removing outdated pages and duplicates improves accuracy more reliably than switching models.
Let's scope your project. Send us your ticket volume, documentation sources and tools: we will price the scope, give an indicative budget and a 6 to 8 week plan. 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.
