- ai-chatbot
- pricing
- customer-support
- roi
- smb
Ask three vendors what an AI chatbot costs and you will get three wildly different answers — because "chatbot" now covers everything from a £30/month website widget to a six-figure custom build. The honest answer is that pricing splits into two layers (what you pay to set it up, and what you pay every month to run it) and that the right number depends almost entirely on volume and how much you customise. Here is what those layers actually cost in 2026, and how to work out when a bot pays for itself.
The two cost layers
Every chatbot has a build/setup cost and an ongoing monthly cost. Off-the-shelf tools front-load almost nothing and charge you monthly; custom builds do the opposite. Knowing which layer dominates is the whole game.
- Off-the-shelf (SaaS) — you configure a hosted product, point it at your help docs, and pay a subscription. Setup is measured in hours, not weeks.
- Custom build — an agency or developer builds a bot around your systems, data and brand. You pay a project fee up front, then a smaller monthly bill for hosting and model usage.
Off-the-shelf pricing
For most SMBs this is the sensible starting point. Entry-level AI chat widgets have free or near-free tiers up to around USD 20/month, and the first serious paid tier clusters at USD 20–50/month. A typical small business running a real support or lead bot lands in the USD 30–150/month range depending on features and message volume (sources: Quickchat AI, Kaily, Crescendo, Tidio).
The important 2026 shift is that many platforms have moved away from per-seat pricing towards per-resolution (outcome-based) billing — you pay only when the bot actually settles a conversation. Two reference points from published pricing pages:
- Intercom's Fin charges USD 0.99 per outcome (a resolution, procedure handoff or disqualification), with a 50-outcome monthly minimum; lead "qualification" outcomes cost USD 9.99 (source: fin.ai pricing).
- Zendesk's AI agent runs roughly USD 1.50 per resolution on committed volume, about USD 2.00 pay-as-you-go, on top of an Advanced AI add-on of about USD 50/agent/month (source: 2026 Zendesk pricing analyses).
Per-resolution pricing sounds cheap until volume climbs — at 1,000 resolved conversations a month, USD 0.99 each is nearly USD 1,000. That is the point at which owning a bot starts to look attractive.
Custom build pricing
Custom pricing spans a huge range because "custom" covers very different things:
- Basic FAQ bot (answers from your documents): roughly USD 3,000–15,000 to build.
- Lead-qualifying bot (asks questions, scores intent, hands off to your CRM): roughly USD 15,000–40,000.
- NLP-heavy or deeply integrated bot: USD 75,000–500,000+, with complex generative systems exceeding six figures.
After the build, a custom bot typically runs USD 200–800/month in API calls, hosting, vector database and maintenance for a straightforward deployment — though heavy-traffic systems can reach several thousand a month (sources: Metageeks, thecrunch.io). The economics against a subscription usually flip somewhere around year two: the up-front build is sunk, and your monthly running cost is a fraction of a per-resolution bill at scale.
The hidden costs
The sticker price is rarely the real price. One 2026 analysis found that the true 12-month total cost of ownership averaged 2.3x the listed subscription once you added everything in (source: thecrunch.io). The usual culprits:
- Per-resolution or per-conversation fees layered on top of the base subscription.
- Onboarding and implementation charges.
- Integration work for any CRM or helpdesk that is not natively supported.
- Team training time.
- Overage charges when you blow past a monthly message limit.
Budget for these up front and the numbers stop surprising you later.
When a chatbot pays off
The ROI case for a support bot is deflection — every question it resolves is a question your team does not touch. The rough sum:
Monthly saving = (conversations handled) × (share the bot resolves) × (cost of a human handling one)
If a human-handled ticket costs you €5 in staff time and the bot resolves 45% of 2,000 monthly conversations, that is 900 deflected tickets — around €4,500 a month in freed-up time. Published resolution rates for the better tools sit at 42–50% (sources: fin.ai, Gleap), so that 45% figure is realistic rather than optimistic. Against a per-resolution bill of roughly €890 (900 × €0.99), the bot pays for itself several times over — and that is the exact moment a fixed-cost custom build starts beating usage-based pricing.
Rather than eyeball it, put your own numbers into our chatbot ROI calculator — ticket volume, resolution rate and hourly cost — and see where the break-even lands for your business. This is the same deflection logic behind how much automation can save across the rest of your operations, not just support.
Quality and hallucination control
A cheap bot that invents answers is more expensive than no bot at all — a confidently wrong reply about refunds, pricing or safety can cost you a customer or a complaint. When you compare quotes, weight these controls as heavily as price:
- Retrieval grounding — the bot answers only from your approved documents (RAG), not from the model's general memory.
- Confidence thresholds and human handoff — when the bot is unsure, it escalates to a person instead of guessing.
- Guardrails and scope limits — it declines questions outside its remit rather than improvising.
- Logging and review — every conversation is reviewable so you can catch and correct drift.
Grounding and handoff are the difference between a bot that deflects tickets and one that quietly creates them. This is a big reason cheap widgets and serious builds price so differently — the guardrails, not the chat window, are where the engineering goes.
How this fits the bigger picture
A chatbot is one line in a wider digital budget. If you are also weighing up the site it lives on, our guide to how much a website costs in 2026 covers the same build-versus-buy trade-off for your web presence, and the two decisions are best made together.
Getting it right for your business
Start off-the-shelf if you are testing the idea and your volume is modest; move to a custom build once per-resolution costs or integration needs justify owning the system. Either way, the guardrails matter as much as the price tag. If you would like help sizing the right option — and making sure hallucination control is built in from day one — take a look at our AI tools service or book a free consultation and we will walk through the numbers with you.