AI Chatbot ROI: What Resolution Rate Should You Actually Expect?

What resolution rate should an AI chatbot actually hit? It depends entirely on what kind of chatbot it is. A basic FAQ bot resolves roughly 20-40% of conversations on its own. A chatbot with real business logic behind it (checking order status, pulling account data) reaches 40-60%. A true agentic system wired into your actual systems — one that can take action, not just answer — routinely resolves 70-90%. If a vendor quotes one number without asking which tier you're buying, that number is marketing, not a real estimate.

I get asked this constantly by people evaluating whether AI support is worth building, and the honest answer starts with which of those three things they're actually picturing — because the gap between them is the whole ballgame, not a rounding error.

Why Does Resolution Rate Vary So Much Between "AI Chatbot" Products?

Because "chatbot" gets used for three genuinely different things. A FAQ chatbot answers from a fixed script or a document — it can't check your systems, so anything outside its script gets escalated. A business-logic assistant can query real data (order status, account details) but still can't take action on its own. A true agent can actually do something — issue a refund, update a record, reschedule an appointment — which is why it clears so much more of the queue without a human. See what actually separates an agent from a chatbot if you're not sure which one a given vendor is really selling you.

Most of the disappointment I hear about "our AI chatbot doesn't work" traces back to buying tier one and expecting tier three's numbers.

Is AI Customer Support Actually Worth the Investment?

The market growth says yes — the AI agent segment of this market is growing roughly twice as fast as the plain-chatbot segment, and the majority of enterprises now use some form of AI agent in their support stack. Businesses that implement it correctly report strong returns, and small businesses specifically have reported very high first-year returns on a well-scoped rollout.

But here's the number that matters more than any of those: realistic net cost reduction lands around 20-35% in the first year — not the 60-80% some vendor pitches imply. That's still a genuinely good return. It's just not the number on the sales deck, and setting expectations at the real figure is what keeps a rollout from getting judged a failure six months in.

Why Do So Many Companies Pull Back Their AI Support Rollout?

Because a large share of companies that launched an AI customer service agent have already rolled it back — and it's rarely because the underlying technology doesn't work. It's almost always one of two things: the tier didn't match the promise (a FAQ-level bot deployed with agent-level expectations), or there was no fallback path for the cases it couldn't handle, so a bad experience in the 10-30% it can't resolve overshadowed the majority it handled fine.

A rollout that's honest about its resolution ceiling from day one, with a clean handoff to a human for the rest, is the version that survives past the first quarter.

What Should You Actually Budget For an AI Chatbot Build?

Cost scales with the tier, same as resolution rate does. A proof-of-concept build runs in the low five figures. A single production workflow — one real use case, done properly, with a human fallback — runs into the tens of thousands. A full multi-channel rollout across several use cases costs more still. Most support teams that build the right tier for their actual volume see it pay for itself inside a year — the mistake is buying tier three's price tag for tier one's rollout, or the reverse.

Pricing here is scoped per project, not a fixed menu — contact me with your support volume and what you'd actually want it to handle, and I'll tell you honestly which tier fits and what it should cost.

Where This Fits With What I Build

If what you need is accurate answers from your own docs or product data, that's RAG & Knowledge Base Chatbots — tier two territory, and often enough on its own. If you need it to actually take action — process a return, update a booking, escalate only the genuinely hard cases — that's AI Agent Development, and it's worth reading what hiring an AI agent developer actually involves before you scope it. I'd rather size the build to what you actually need resolved than sell the bigger tier by default.