The verdict in three sentences
A high-performing B2B customer support AI chatbot relies on a RAG (Retrieval-Augmented Generation) architecture that queries YOUR knowledge base, not a generic model. In 2026, budget GBP 11,000-32,000 for the build depending on scope, plus an LLM cost of GBP 0.40-2.50 per 1,000 queries and maintenance around GBP 260/month. The payback is measured in ticket deflection: 30-50% of tier-1 requests resolved without a human.
What a custom RAG chatbot costs in 2026
Pricing hinges on the quality of your source documentation and the number of systems to connect (CRM, ticketing, product catalogue).
| Scope | Build (GBP) | Timeline | Maintenance/mo |
|---|---|---|---|
| Enriched FAQ MVP (1 source) | 11,000 - 14,500 | 4 wks | GBP 220 |
| Standard RAG (3-4 sources) | 16,000 - 25,000 | 6 wks | GBP 270 |
| Advanced RAG + human handoff | 25,000 - 32,000 | 8 wks | GBP 360 |
| Multilingual + CRM connector | +5,500 - 9,000 | +2 wks | +GBP 130 |
Maintenance covers re-indexing, hallucination monitoring and prompt tuning. Without it, quality decays within 3-6 months.
LLM running costs, often underestimated
Beyond the build, each conversation consumes tokens. Here is a 2026 order of magnitude for a typical monthly volume.
| Monthly volume | LLM cost (GBP/1000 q) | Est. monthly cost | Hosting/vectors |
|---|---|---|---|
| 5,000 queries | 0.40 - 1.00 | 2 - 5 GBP | 35 GBP |
| 20,000 queries | 0.90 - 1.70 | 18 - 34 GBP | 80 GBP |
| 50,000 queries | 1.30 - 2.20 | 65 - 110 GBP | 130 GBP |
| 100,000 queries | 1.80 - 2.50 | 180 - 250 GBP | 220 GBP |
Marginal cost stays low: the real spend is the build and maintenance, not inference.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Sarah, head of customer support at a B2B SaaS vendor in London, handles 4,200 tickets/month with 6 agents. A tier-1 ticket costs about GBP 8 in internal load. She invests GBP 22,000 in a standard RAG, plus GBP 270/month maintenance and GBP 35 of LLM. With 40% deflection, she saves 1,680 tickets/month, i.e. GBP 13,440/month in avoided cost. The build pays back in under 2 months, and her agents refocus on strategic accounts.
FAQ
Will an AI chatbot hallucinate and make up answers?
With a well-built RAG architecture, the model answers only from your documents and cites its sources. Production hallucination rates fall below 2% after tuning, versus 15-20% for a generic LLM with no guardrails.
How is it different from a classic scripted chatbot?
A decision-tree chatbot costs GBP 2,500-7,000 but cannot handle rephrasing and caps at 10-15% deflection. RAG understands natural language and reaches 30-50%.
How long to go live?
Expect 4-8 weeks depending on the number of sources. The longest phase is cleaning and structuring your knowledge base, often 40% of the project.
Can we keep a human in the loop?
Yes, agent handoff is standard: the bot escalates complex cases with full conversation context, cutting human handling time by 20-30%.
Let's scope your project. Share your ticket volume, documentation sources and target CRM, and we'll price a RAG chatbot in the GBP 11,000-32,000 range by scope. 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.
