Almost every business site has a chat bubble in the corner now. The part that actually matters — which one resolves a real question without a human, what it costs once you're past the free tier, and where your visitors' conversations physically end up — varies a lot more than the bubble itself suggests. We looked at the four platforms businesses actually shortlist, plus the option most vendors won't mention: building your own.
What actually differs between these
Every platform here claims to "resolve conversations with AI." The real differences show up in four places:
- Resolution rate — the share of conversations the AI actually closes without a human, not just replies to.
- Pricing model — flat monthly seats, per-resolution billing, or a usage tier that quietly jumps at scale.
- Data residency — whether conversations sit on a vendor's shared infrastructure or somewhere you control.
- Customization depth — whether you can change the retrieval/knowledge-base logic, or you're stuck with what the vendor ships.
1. Drift — best for high-ticket B2B sales teams
Drift is built around conversational marketing, not support deflection — account-based routing for named accounts, native chat-to-meeting booking, and playbook templates for sales teams. Entry pricing starts around $2,500/month, which prices out most small businesses on day one.
What Drift is actually built for
- Account-based routing — recognizes named target accounts and routes them differently than anonymous traffic.
- Native chat-to-meeting flows — books a sales call directly out of the chat window.
- Resolution rate averages 40–55% — it's optimized for qualifying and routing leads, not closing support tickets.
Best for: enterprise SaaS and B2B teams selling $50K+ deals who are already running account-based marketing.
2. Intercom Fin — best resolution rate, if you're already on Intercom
Intercom Fin posts the highest resolution rate in the category — 50–70% of conversations closed without a human — by pulling directly from Intercom's own Help Center knowledge base. Pricing is per-resolution ($0.99 each) on top of an Essential plan starting around $74/month, and typical mid-market spend on Fin alone lands at $500–2,000/month.
Why teams pick Fin
- Highest resolution rate in the category (50–70%), because it's tightly coupled to your existing help docs.
- Per-resolution billing aligns cost directly with successful deflections instead of flat seats.
- Native handoff to a human agent inside the same Intercom inbox, no context lost.
Best for: mid-to-large SaaS companies already running Intercom for support who want the AI layered on top, not bolted onto something new.
3. Tidio — best value for small business and e-commerce
Tidio is the most accessible AI chatbot for small teams — $29/month to start, with its Lyro AI add-on from $39/month, scaling to $499/month at the top tier. Native Shopify and WooCommerce integration makes it a natural fit for stores handling order status and abandoned-cart questions.
Resolution rate averages 35–55% — solid for FAQ and order-status volume, less suited to nuanced support. Best for: SMB e-commerce and cost-conscious teams that want AI under $200/month without an implementation project.
4. Crisp — best for EU teams and multi-channel support
Crisp has a genuinely usable free tier and a Pro tier ($25–95/month) that adds its MagicReply AI. Its real strength is a unified inbox across chat, email, SMS and social — AI depth is the trade-off for that breadth, with resolution rates averaging 30–45%.
Best for: budget-constrained teams, and especially EU-based businesses — Crisp's GDPR-default posture is stronger out of the box than most US-first competitors.
Side-by-side
| Platform | Starting Price | Resolution Rate | Best For | Data Control |
|---|---|---|---|---|
| Drift | $2,500+/mo | 40–55% | Enterprise B2B sales | Vendor-hosted |
| Intercom Fin | ~$74/mo + $0.99/resolution | 50–70% | Existing Intercom users | Vendor-hosted |
| Tidio | $29–499/mo | 35–55% | SMB & e-commerce | Vendor-hosted |
| Crisp | Free–$95/mo | 30–45% | EU teams, multi-channel | Vendor-hosted (EU-based) |
| Self-hosted RAG | Build + infra cost | Depends on your data | Compliance-bound or custom needs | Fully yours |
When a pre-built SaaS chatbot isn't enough
Every platform above hands you a fixed chunking strategy, one vendor-controlled embedding model, and no visibility into how it's actually retrieving answers from your content. That's fine until it isn't — usually around a compliance requirement, a knowledge base too specific for a generic integration to answer well, or a volume where per-resolution or per-seat pricing stops making sense.
Where self-hosting genuinely wins
- Data sovereignty — conversations never leave infrastructure you control, which matters the moment PHI, PII, or a signed DPA is involved.
- Full retrieval control — you choose the vector database, the embedding model, and can tune retrieval scoring instead of trusting a black box.
- Cost crosses over at scale — the breakeven against SaaS pricing typically lands around 80,000–100,000 conversations/month, per published self-hosting cost analyses.
- No vendor lock-in on the LLM — swap providers without renegotiating a contract.
The honest trade-off: self-hosting needs a minimum of roughly 0.3 FTE of ongoing engineering time for updates and operations — below that, operational debt piles up faster than the setup pays for itself. It's the right call when compliance is non-negotiable or your knowledge base is too specific for an off-the-shelf integration, not a default choice for every site.
This is exactly the kind of build we do for clients whose support content, compliance requirements, or scale don't fit neatly into a SaaS platform's fixed pipeline — including the assistant on this site, which runs on rag-chatbot-plugin, an MIT-licensed, drop-in RAG chatbot we built and open-sourced: an embeddable widget plus a self-hosted RAG backend (point it at a folder of PDFs) with enquiry/booking email notifications, for Next.js, Vite, Express, or standalone.
How to choose
- 1Start with volume: under a few thousand conversations a month, any SaaS tier is cheaper than the engineering time self-hosting needs.
- 2Check what you're already paying for: if you're on Intercom for support, Fin is the lowest-friction add — you're not integrating a second system.
- 3Match the platform to the job: Drift for sales-qualification, Tidio or Crisp for support/FAQ deflection — don't buy a sales tool to do support.
- 4If compliance requires it, skip straight to self-hosted: no SaaS DPA fully substitutes for data that never leaves your own infrastructure.
- 5Pilot before you commit to per-resolution billing: Fin's $0.99/resolution is efficient at moderate volume but can surprise you at scale — model it against your actual traffic first.
FAQ
What's the best AI chatbot for a small business website?
For most small businesses, Tidio ($29–79/month) offers the best balance of price and capability, with native Shopify/WooCommerce support if you're running a store. Crisp is the stronger pick specifically for EU-based teams that want GDPR-default compliance without extra setup.
How much does an AI chatbot for a website actually cost?
Anywhere from free (Crisp's free tier) to $2,500+/month (Drift's enterprise tier). Most small-to-mid businesses land between $30–200/month on Tidio or Crisp; mid-market teams using Intercom Fin typically spend $500–2,000/month once resolution volume is factored in.
Is a self-hosted chatbot cheaper than a SaaS one?
Not at low volume — SaaS wins until roughly 80,000–100,000 conversations a month, since self-hosting needs about 0.3 FTE of ongoing engineering time plus infrastructure cost. Past that volume, or wherever data sovereignty is compliance-required, self-hosted becomes the more economical and defensible choice.
Do AI chatbots actually resolve customer questions without a human?
Yes, at meaningfully different rates by platform — from 30–45% (Crisp) up to 50–70% (Intercom Fin), largely depending on how tightly the AI is coupled to your actual knowledge base rather than a generic model alone.
When does a business need a custom-built chatbot instead of an off-the-shelf one?
When compliance requires data to stay on infrastructure you control, when your knowledge base is too specific or proprietary for a generic integration to answer well, or once conversation volume is high enough that per-resolution or per-seat SaaS pricing stops making financial sense.
Conclusion
There's no single "best" chatbot — there's the one that matches your volume, your compliance requirements, and what you're already running. Most businesses starting out should pick a SaaS tier with a free plan, learn what visitors actually ask, and only look at a custom or self-hosted build once that volume or a compliance requirement makes the case on its own.
Weighing a custom build? Get in touch to talk through your actual traffic and compliance needs, or explore our AI development services if you're evaluating a self-hosted RAG chatbot for your own site.
