Stop Guessing: Use Topics Explorer to Upgrade Fin Fast | Ep. 4 with Jen Weaver

Knowing where to focus when improving Fin's performance is one of the most common challenges for support leaders. Without clear data, it is easy to spend time on content that does not move the needle — and miss the topics that are quietly dragging your resolution rates down.

In this episode of Support Stack, Conor Pendergrast is joined by Jen Weaver, a customer support specialist at Supportman working with Tettra, to explore Intercom's Topics Explorer — a built-in beta feature inside the Fin AI Agent that transforms vague instincts about Fin performance into specific, actionable priorities.

The episode centres on a treemap-style view that automatically clusters conversations into topic groups. Larger blocks represent higher conversation volume, while the shading indicates resolution rate: darker colours signal lower resolution, lighter colours signal stronger performance. Viewing Tettra's year-to-date data, Conor and Jen quickly spot GuideMaker browser usage sitting at just 33.3% resolution — a standout gap against Tettra's overall Fin resolution rate of around 65%.

From there, the conversation moves into Intercom's Fin testing suite, where Jen creates a new test group scoped specifically to that underperforming topic. Intercom generates a set of representative questions drawn from real inbox conversations, and Fin attempts to answer them in real time. Conor walks through how to rate each response as Good, Acceptable, or Poor using simple keyboard shortcuts, and explains how adding internal notes makes the process useful for product and content teams — not just support specialists.

The episode closes with a key insight: poor resolution on a topic like GuideMaker often points to a gap in content, data, or available actions. In this case, Fin was drawing from general Tettra documentation rather than GuideMaker-specific content. The fix is straightforward — update the documentation, then retest — but the Topics Explorer is what makes it possible to find that gap in the first place. If you are working with Intercom, Fin, or exploring AI in customer support, this episode provides a practical framework for turning resolution rate data into targeted, confident improvement actions.

Join my daily email list to get emails about future episodes.

Episode transcript

Conor (0:00)
Hi, I'm Conor Pendergrast from CustomerSuccess.cx.

Jen (0:03)
And I'm Jen from Tettra.

Conor (0:05)
Welcome to episode 4 of Support Stack.

Conor (0:16)
So Jen, we're talking about quite a fun topic today. Topics — we're talking about topics. The topic of today is topics. So I don't know about you, but one of the things I found with Intercom is, you know, there are opportunities out there to improve how Fin is performing and to improve how the whole support experience is going. But the challenge is where do you focus? Like, how do you find those needles in the haystack? How do you find those opportunities? So what I had suggested is we look at Topics as a way of doing that. Are you happy for us to do that, Jen?

Jen (0:53)
Absolutely. I have it up on my screen.

Conor (0:56)
That's super. Okay. So for those who are not familiar, Topics Explorer is an Intercom built-in feature. The pathway to it — thanks for the reminder, Jen — is through the Fin AI Agent. And then you're going to go into Topics Explorer, which is a beta feature at the moment, but I think it's pretty cool and it's working very well for me and my clients. And so what it does is Intercom defines these topics. So that's the first thing to know. These are not topics that you are setting in stone. They are — I've seen some of their recent webinars — they're going to allow a little bit more input into how topics are generated and defined, but that's not today. That will be in the future. It basically looks through and has a bit of a guess based on the content of conversations. So what we've done here with your Intercom setup, Jen, is we've looked from the beginning of the year to today — the year-to-date option — because we want to get a nice big sample. As many conversations as you can. And in this case, I suggested we look at the resolution rate — that's Fin's resolution rate. You can also look at CX score, which is the customer experience score. It's a measure of how Intercom thinks the customer support experience is handled. It's not rated by the customer, so it has a much higher coverage rate than CSAT surveys would. But it is still Intercom's guess based on their AI of how the customer experience was. Today we're looking at resolution rates. So you've got this — I can't remember the name of this type of chart, but I think it's pretty. It's colour-coded and it's got boxes that go bigger for more volume topics and smaller for lower volume topics. And then the colours: in light mode, the darker the colour, the lower the resolution rates; the lighter the colour, the higher the resolution rate. So there's one, exactly as you're looking at, Jen — GuideMaker browser usage has a 33.3% resolution rate. Like, Tettra's overall resolution rate is very high for Fin — it's a solid, what, 65%-ish? What's going on, Jen? How is it doing so terribly?

Jen (3:28)
This is one of our newest tools that we have brought into the family of Tettra tools, and so GuideMaker — we probably don't have as much documentation about it.

Conor (3:39)
Aha! Well, you know what that sounds like. That sounds like an opportunity. But the question is, what would you do? So now you know there's a gap here, but how do you then do something about it? What would you do, Jen? Where would you go next?

Jen (3:55)
Good question. So if we go over to Fin — where we are now — and then down to Test. So this is where we can play around with what Fin does.

Conor (4:07)
Yeah, so we've got — as you were telling me earlier — in the centre we have these questions and then we have the answers where we can rate the responses. But there's more to it than that. I'm sure you're going to tell me.

Jen (4:19)
I, well, I'm going to make a suggestion. So we just saw that terribly performing topic earlier with the 33% resolution rate. What is going on? So here's the thing, dear watcher — you can do something about that. And the best thing to do is you can test against that group. So they will give you questions and Fin will try its best to answer them based on the content, data, and actions it has available.

Conor (4:46)
So Jen, let's click on the top — let's do a new group. Let's go and set up a new group for this. And you can do create new group. GuideMaker, why not? And so there are options — Intercom gives you three sets of options here. You can generate from the inbox, the inbox being customer support conversations. Or if you're maybe setting up Intercom for the first time and you don't have a lot of customer support conversations, you can add them manually either by typing them in or via CSV. CSV would be useful if you had a bunch of data somewhere and you asked your favourite GPT of choice to generate questions for you. You could get your help site, paste it into Claude or ChatGPT or Gemini, and then have it create questions in a customer's voice. In this case though, we're just going to use generate from inbox, by topic — specifically the GuideMaker browser use topic. So Jen, you can click create there. What it does is it goes in, looks through past conversations where it worked out what the topic was and puts them into this bucket. And so you get a bunch of different variations on different questions that customers will have asked. And then Fin will try and answer those based on the content, data, and actions available. This is kind of one of my favourite parts of Intercom and Fin — it can very quickly give you an idea about the quality of the interactions and also allows you to do something about it and then retest. So Jen, you can see Fin has generated 10 questions in the middle here, and answers on the right-hand side. You can very quickly go through these — G for good, A for acceptable, P for poor. And importantly, you can add internal notes. So for someone like me — I come into a company, my expertise is not in a client's product. My expertise is in Intercom and Fin, driving up resolution rates, customer satisfaction, CX scores. I don't know the product. So when I get someone to go through this, I have one of their product experts review these responses and rate them and add notes for what was wrong or what was right.

Conor (7:51)
So now you can see — if you scroll up slightly — it has the question and also this detailed response from Fin. And so this is a good way of seeing what actually happens when Fin is interacting with these questions. And if you scroll right down, it's a pretty long answer. So it also has the content that it drew from and the guidance that it used.

Jen (8:14)
Hmm. Interesting. Yeah.

Conor (8:18)
Yeah, it tells you which part of the guidance it used. It's using a neutral tone. It's comprehensive, which is why it's getting longer answers. And then click on Other there as well — Other is the data connector. So that's going and checking your Tettra status page to see if there are any incidents it could talk about. So content — you can click into any of those. The content snippets are the middle one and the help articles are the first and third. And so you can review this and decide: is this a good answer? Is it an acceptable answer? Is it a poor answer? And then when you've reviewed all of these, you can filter and say, okay, let's focus on poor first. What would Fin need to know? What would Fin need to be able to do to fully answer this question and push it from poor to acceptable or from poor to good? And that's where it comes back to: what's the content? What's the action? What's the data that Fin needs? Could be editing content, removing duplicate content, moving out-of-date content, adding data that gets pushed from your systems. Or it could be actions — using a data connector and a Fin task to go off and grab information. Tell me about your thoughts on this, Jen.

Jen (10:03)
Um, my thoughts are I really need to go back into documentation, because all of this content is regarding Tettra and not regarding GuideMaker — the specific tool within our suite of tools. And so I would rate this poor because it's answering for the wrong tool, but I would just make a note here. And then I can go through and look at each of these and rate them — and kind of see, if I went through all of them, I'd probably see that at least hopefully some of them are pulling from GuideMaker content. But then I probably would ask someone to update that and then test it again.

Conor (10:43)
Yep. Yep. And that could be an example. So given you have multiple products, that's a good example of where data is going to help — give Fin a better ability to answer a question, right? So by pushing into Intercom the fact that someone is a GuideMaker user, Fin can use that and you can set up guidance that says, this is the field that says they're a GuideMaker user. If you see this, then interpret their question through GuideMaker and give them the right information. So that's exactly what I mean when I say content — because the content for GuideMaker might be there already, but if it's not there, Fin definitely can't do the work. But the same with data — if the data is not there, then it's going to be a lot less likely to provide a spot-on answer.

Jen (11:38)
Hmm. Yeah, this is cool. We uncovered something I want to dig into here.

Conor (11:44)
So thanks for that. Ah — live on the episode. Well, thank you very much, Jen Weaver, for being here on episode number four of Support Stack. If people were delighted by your presence here — which I think they should be — how would they find out more about you and how could they connect with you?

Jen (12:05)
Sure, LinkedIn — connecting with me on LinkedIn is my preferred way for you to connect. But also when you do that, I have a newsletter just for support leaders that's available on LinkedIn. It has both my podcast, Live Chat with Jen, which is for support leaders, and also my Friday newsletter, which is a roll-up of all the podcasts in support during the week. Every Friday, yeah.

Conor (12:32)
That's amazing. Every Friday. I would recommend both. I would recommend connecting on LinkedIn and subscribing to that. And if you, dear...

Jen (12:39)
Go on.

Conor (12:41)
Jen. And if you, dear, dear, dear watcher would like to get future episodes, you can subscribe to my email list at CustomerSuccess.cx/daily, where you will get a weekdaily email from me. It's all about Intercom and some customer support leaders and very rarely about anything else.

Jen (13:06)
Thank you. Bye.

Conor (13:08)
That's awesome.