AI chatbots that actually convert.

The bar for AI chatbots has risen sharply since GPT-4 went mainstream. Customers expect a bot that understands their question, answers from your actual content (not generic boilerplate), and either resolves the query or hands off cleanly to a human. The old "press 1 for sales, press 2 for support" decision trees are dead — and rightly so. Here's what modern AI chatbot development actually involves.

What a good 2026 chatbot looks like

It's grounded in your business — not Wikipedia. It uses retrieval-augmented generation (RAG) to pull relevant context from your knowledge base, product catalogue or documentation before answering. It can call tools — check stock, book appointments, generate payment links, escalate to humans. It speaks in your brand voice. It handles 40–70% of conversations end-to-end and escalates the rest cleanly. And it gets smarter every month because someone (us) is reviewing transcripts and refining prompts and content.

Use cases that actually pay back

Support deflection: 40–70% of repeat questions answered without a human, freeing your team for complex cases. Lead qualification: bot collects requirements, scores fit, hands warm leads to sales with a transcript. Appointment booking: bot checks calendar availability and creates bookings in Calendly, Cal.com or your custom system. Onboarding: walks new users through setup, answers questions in context. Internal knowledge: staff query company SOPs, HR policies and product docs in natural language.

Models we work with

OpenAI GPT-4o and GPT-4o-mini: the workhorses. GPT-4o-mini at ~£0.15 per 1m input tokens is the cost/quality sweet spot for most chatbots. Anthropic Claude Sonnet/Haiku: stronger on careful reasoning, long context (200k tokens) and tone control. Google Gemini: good for multimodal and Google Workspace integrations. Open source (Llama, Mistral) via Groq or Together AI: useful for high-volume use cases where token costs matter. We pick per project; we're not locked to a vendor.

The build process

Week 1: scope, content audit, intent mapping. Week 2–3: ingestion pipeline (your docs, FAQs, product data) into a vector store like Pinecone or Supabase pgvector. Week 3–4: prompt engineering, tool integration with your CRM and calendar. Week 5: UI build (web widget, WhatsApp, voice as required). Week 6: testing, evaluation against a golden dataset, soft launch. Most SMB chatbots are live within 6 weeks.

Pricing

Starter chatbot (web widget, FAQ answers, basic lead capture): £1,200–£2,500. Standard (RAG, CRM integration, booking, custom UI): £3,500–£8,500. Pro (multi-channel, voice, advanced tooling, fine-tuning): £8,500–£25,000. Ongoing care plans from £200/month for prompt tuning, content updates and performance monitoring.

What I bring

I've built chatbots for ecommerce, professional services, SaaS and hospitality. I work with the tooling that's mature in 2026 — LangChain or LlamaIndex, OpenAI/Anthropic APIs, Pinecone or Supabase vector, Make/n8n for orchestration — and I integrate them properly with your existing stack rather than bolting on a chat widget that lives in isolation. Every project ships with a transcript review dashboard so you can see what people are asking and improve the bot over time.