The verdict in three sentences
An AI chatbot built on your documents (RAG) costs, in 2026, between 12,000 and 40,000 EUR to build, plus 100 to 1,000 EUR/month in API and hosting. Properly scoped, it deflects 30 to 50 % of tier-1 tickets and pays back in 4 to 9 months for a team handling more than 800 tickets/month. The real trade-off isn't SaaS versus custom, but per-agent subscription versus an integrated asset tied to your data that you own.
What an AI chatbot costs in 2026
A useful chatbot doesn't answer at random: it draws on your documents (FAQ, terms, product sheets, ticket history) through a RAG architecture. The budget depends mostly on the quality of your document base and the depth of integration with your tooling (CRM, knowledge base, ticketing).
| Line item | 2026 range (EUR) | Note |
|---|---|---|
| Scoping + conversation design | 2,000 - 5,000 | Flows, tone, guardrails |
| Ingestion + indexing (RAG) | 4,000 - 12,000 | Cleaning, chunking, vectors |
| CRM / ticketing / site integration | 3,000 - 10,000 | Zendesk, HubSpot, widget |
| Guardrails + testing + go-live | 3,000 - 13,000 | Anti-hallucination, human handoff |
| Total build | 12,000 - 40,000 | 6 to 12 weeks |
On the running side, watch two lines: the API cost (billed per token) and the run (hosting, monitoring, vector store).
| Recurring cost | 2026 order of magnitude | Basis |
|---|---|---|
| Model API (tokens) | 0.002 - 0.02 EUR / 1,000 tokens | By model and length |
| Hosting + vector store | 60 - 400 EUR/month | Volume and traffic |
| Total monthly run | 100 - 1,000 EUR/month | 1,000 to 20,000 chats |
| Maintenance (updates) | 250 - 700 EUR/month | Content + model updates |
Per-agent SaaS or integrated custom
A SaaS widget is often billed per agent or per resolution volume; easy to launch, but the cost climbs with the team and you stay a tenant. Custom requires a heavier upfront spend, but the asset is yours and integrates deeply.
| Criterion | SaaS widget | Integrated custom |
|---|---|---|
| Entry cost | 0 - 3,000 EUR | 12,000 - 40,000 EUR |
| Recurring cost | 40 - 120 EUR/agent/month | 100 - 1,000 EUR/month (run) |
| Integration with your data | Limited | Deep (dedicated RAG) |
| Asset ownership | No | Yes |
| Tone / brand customization | Low | Full |
| Economics at 20 agents | Linear cost | Stable cost |
Mini case study
Sophie runs customer service at an e-commerce SME in London: 1,400 tickets/month, of which 55 % are repetitive (order tracking, returns, sizing). The fully loaded cost of a human-handled ticket is estimated at 4.50 EUR. A RAG chatbot at 24,000 EUR deflects 40 % of the 1,400 tickets, i.e. 560 tickets/month saved: 560 x 4.50 = 2,520 EUR/month. After the run (500 EUR/month) and maintenance (400 EUR/month), the net gain is 1,620 EUR/month. Payback: 24,000 / 1,620 = ~15 months, dropping to ~9 months as volume and deflection rise after three months of tuning.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
FAQ
How long to deploy a reliable AI chatbot?
Budget 6 to 12 weeks: 2 weeks of scoping, 3 to 5 weeks for document ingestion and integration, the rest for testing and guardrails. A pilot on a narrow scope (product returns, say) can ship in 4 weeks.
What ticket share can it actually absorb?
On well-documented tier-1 questions, deflection runs 30 to 50 %. Beyond that you must enrich the base and wire in actions (real-time order status), which raises cost but also value.
How do you avoid hallucinations?
You constrain the model to answer only from indexed documents, show the sources, and force a human handoff when confidence is low. A good guardrail keeps off-topic answers below 2 % of chats.
Can the API cost run away?
Rarely, if you cap answer length and cache frequent questions. For 10,000 chats/month, the API bill typically stays under 300 EUR/month in 2026.
SaaS or custom for a 15-agent SME?
Beyond 10 to 12 agents, custom is often cheaper over 24 months because its cost doesn't scale with headcount, unlike a per-agent subscription.
Let's scope your project. Share your ticket volume, your tools (CRM, ticketing) and your share of repetitive questions: we'll scope a costed pilot with a deflection target. 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.
