What are your best practices for chatbot deflection? | Community
Skip to main content
Question

What are your best practices for chatbot deflection?

  • June 1, 2026
  • 1 reply
  • 34 views
jenmr

Hi everyone! I'm exploring best practices around chatbot deflection and would love to hear from this community's experience. Even though you may be using Fin, the underlying strategies likely apply across platforms.

We're currently using AI chatbot tools for customer support and are looking to improve our deflection strategy. A few questions I'd love your input on:

  • What deflection rates are you seeing, and what's considered "good" in your industry?
  • What types of content (articles, FAQs, flows) drive the most successful deflections?
  • How do you measure true deflection success vs. a user who just gave up?
  • What's the biggest mistake you made early on that others should avoid?
  • How do you handle graceful handoff to a human agent when the bot can't resolve an issue?

Appreciate any tips, benchmarks, or lessons learned, even if your setup looks different from ours!

1 reply

Tonje Ness Meinhardt
Forum|alt.badge.img+2

Hi, ​@jenmr 👋 — great questions, really relevant discussion.

We’ve been using Intercom and Fin since 2024 and have gathered some practical learnings along the way:

  • We’ve improved our resolution/deflection rate from around 54% to 60%. It’s hard to say what’s “good,” though, since our setup is somewhat limited — we don’t have APIs available or much automation in place, meaning Fin can’t take many actions through data connectors. Everything on our end is based on generic information. Also, like you mentioned, it’s difficult to know whether these are true resolutions or cases where the customer simply gave up, or jumped to another channel, calling us instead.

  • In terms of content, our approach has been to continuously fill knowledge gaps based on actual customer questions. We’ve essentially kept adding content (FAQ, snippets, PDFs and synced websites) until Intercom stopped suggesting new areas to improve. That said, we clearly see data and action gaps, largely due to missing APIs toward our core customer systems.

  • We also find it challenging to distinguish between true deflection success and user drop-off. Our assumption is that Pro Add-On features could help us better quantify this, especially with clearer resolution tracking using CX score. There is also a lot of other useful tools in the Pro Add-On feature, that can help keep track of things.

  • The biggest mistake we made early on—before becoming an Intercom customer—was uploading all our content at once, not written for AI. In hindsight, we should have taken a more incremental approach, allowing us to observe how the chatbot behaves and improves over time. That’s exactly the approach we adopted with Intercom, and it delivered significantly better results.

  • For human handoff, we currently rely on escalation guidance and procedures. Procedures have been particularly valuable when we need to collect structured information from the customer before performing an action. This allows our support agents to simply input the data directly into backend systems without chasing missing details — Fin has already done that work. Even without strong API integrations, this has proven very effective for us.

That’s some of our experience so far. 

    Terms & ConditionsAccessibility statement