Making Unstructured Feedback Actionable with AI | INBOUND 2025
Robby introduces HubSpot's Feedback Topics feature, which uses AI to automatically organize scattered customer feedback and flag actionable signals like competitor mentions, pricing concerns, and churn risks. The tool tags feedback automatically and triggers actions across your team.
Chapters
All right, inbound. What's happening? This is making unstructured feedback actionable with AI. Before we get to me, and before I say the word unstructured about 50 times, let's read the room with a show of hands. I will only do this once. How many of you have heard the phrase unstructured feedback before? Okay, not bad. Now, of those who raised their hands, who among you has the guts to give me an example of unstructured feedback? Anyone? Yes, Anything. Where would you find unstructured feedback? We request it from our clients to quit after their experience. Right. So you'd request it after something, like in an email. Okay. Any other examples? That's one. One more. Yes. How did you hear about us? Right. So you ask them, and then they would tell you. They'd write it, or they would say it to you. Unstructured feedback is if you know what unstructured data is. Unstructured data, any text, anything that isn't a number. Right. A date, chat, email, audio. Unstructured feedback is feedback in those forms, whether or not the customer intended to give it to you. So this is about making use of that. I'm Robby. That's a big. That's a really big picture. I joined HubSpot as a staff PM from a company called Frame AI, which HubSpot acquired right at the beginning of the year. I've been working on this problem for a long time directly with customers who were desperate to extract meaning from unstructured feedback really just means to read between the lines to hear what their customers were already saying to them. And they did this because the usual suspects weren't doing the job. Surveys are useful, but they weren't telling them what they needed to know. And they wanted to understand what was happening in the conversations they already had. One CEO told me the biggest growing pain he had in going from a small company of 20 people in one office to a much bigger company spread across the country was losing the customer intuition he had when he just knew exactly what the problems were. He was in product planning meetings. He was talking directly to customers. He was directly in sales meetings. And as they got bigger, he lost that. They had Personas. He was in slide meetings, getting the quarterly NPS summaries and the customer interview summaries. And they were looking for a way to get back to that feeling and something closer than just secretly reading support tickets, which CEOs sometimes do. And it's a very unhealthy habit. But it's not just that CEO. Every CEO wants that customer intuition. Yamini. Wants that intuition. You really want to feel that connection about what customers are actually trying to say to them. And every company says we listen to customers not just to make customers feel good, but because they want to know how to get better, how to make their products better, how to make more money. So today I want to talk about. Why feedback is hard, why this is such a hard problem, even for some of the most advanced technology companies, advanced public and private companies. And it's such an, such a deceptively hard problem. I'm going to spend more time than usual dragging all of us through the glass on this so that when we get to the next step, which is showing you today's announcement, we'll all have a full appreciation for how powerful it is. Because the solution is also deceptively simple in HubSpot. And the third thing I'm going to impress upon you why this is for all of you. It's for everyone. It's not just a support tool. It's not just a success tool. It's for revenue leaders. It's for content creators. It's for everyone. All right, let's do it. So I'm going to set the stage for you. Imagine it's 9am on a Monday. CEO drops a note like this in Slack. You're on your first cup of coffee and asks a very simple question. How many customers complained about pricing last quarter? It's a simple question. And if you're like this young woman here, you open your magic spreadsheet without all the customer feedback in it and you click a declack your way to the answer and you answer right. Well, for the rest of us, we don't have that. And if you're actually going to try to answer this question and you don't already have a solution set up, you're going to start asking things like is it in the gong calls? Okay, we'll go get those. Did it come up in NPS survey comments? Is it in the QBRs? Did we tag the support tickets? And like on and on, you're going to think about where's all this data? Did it come up? Did we note it? Did all the reps write it down? Did we collect it, et cetera? It's natural that the data would be scattered. So just one question is a full on data project. Assuming you're going to actually answer the question, you might also ask the CEO, what are you going to do with this information? Like what action are you going to take? What form do you need it in? Simple question and I have Two predictions about this question. First, this sounds like an important question the CEO is probably going to ask you again. This won't be the only time. Second, they'll probably ask you another hard question some other time. That will become another full on data project. So the CEO should not have to keep asking you this question. If they ask an important question, it should become an important question that you're always able to answer, which means you shouldn't have to keep asking. You should also have a way to just have that answer. So we should have systems that work for us even when we don't, even when we don't ask them. We should not have to keep logging into a system and asking again, what's the answer here? What's the answer here? Systems should get better than that. Okay, so we know that the data's scattered everywhere. This is common. We've all talked about this before and it's natural because everyone works in their own tools, they work on their own systems and that's where the data lives. So those teams own that data. And as a result, if you want to actually build a system to make use of this and answer that question and answer any similar question and get that feedback, you're going to need a few things. You need ongoing unification of data. And how many people this week have already talked about that? You got to unify the data. You have to get it out of silos. Karen talked about that in the kickoff. You need alignment on what all this data means, how do the schemas talk to each other, what do we do when some of the data is missing, blah, blah, blah. And even some of the biggest, most advanced teams struggle with this. I've worked with like public technology companies, fully staffed, fully using the latest technology, who still had the same kind of constraints that any company does. They don't have time. They couldn't get pull it together. So this is just, it's a real challenge even when you know exactly what to do and, and you have all the tooling and people. Okay, so I've described to you what may feel like a big enterprise problem and just getting data together. And I know a bunch of you right now are thinking, robbie, doesn't matter. I have a bunch of AI tools that I love and many of, many of you have showed them to me over at the booth this week. Look at this awesome note taker. And I believe you and I agree with you. There are, they're sick. They take awesome notes or they do awesome call coaching. And as Karen also said, they're really good at what they do. Their job is to nail a space, to nail a task, to nail a job function. If you're in success, there's a tool for that. If you're in support, there's a tool for that, et cetera. There's also some. There are some tools that do things that are fairly trivial or standard. So one example is if you turn on some support tools, and I say support because that's where a lot of conversations exist that are meaningful. They will tell you a bunch of your conversations have a negative sentiment. No kidding. That's what support is. They will also tell you that a lot of your conversations are about your products and about billing. No kidding. I know that. Thanks for the autoclustering. Okay. At the same time as you are experiencing churn and trying to figure out why people are frustrated with your products, your support team will tell you the RCSAT score is between 95 and 97%. So there's no issue. We don't need to look at it any further. And they're doing a great job, which is probably true. So I guess you start doing more surveys. You're trying to figure it out. But the point is that those tools that you use to do amazing call recording, for example, are not built to help you understand all of the data together. In order to answer that very simple question, you can't ask it what happened last month. You can ask it what happened on that one call with Bill. And they're definitely not equipped, it's not even their job to help you figure out what to do with that. Even if you were able to answer what happened, what should I do with that? What action do I take? It's just not their mandate. So you need something that will help you get the answer and then actually figure out what to do with it. So. So now we're at the stage where companies have amazing tools and they're building up their AI team. It's been called a hybrid team. And if you're at this stage, this is becoming your AI strategy and you're giving your team tools, and you're also probably giving them, or you're personally using GPT or an LLM or notebook. I'll bet about 30 of you are staring at it right now. And this is amazing. You get superhuman employees. They can take documents and they can put them into some sort of tool and they can get answers from it. Hopefully it's the right answer. It's hard to verify. And this gives you some people who are awesome. Unfortunately, business runs all the time. Not only when you put documents into a tool. So while you get answers and then they come out, that's where it ends. The rest of the team did not benefit from it. The company did not as a whole get better from it. And it doesn't actually feed back into your normal systems. Systems don't run on text, they run on numbers, they run on currency, they run all sorts of things. If you want to actually turn into an AI company, you need AI that's actually built into your systems. It cannot just be talking to a robot. It cannot just be analyzing documents one at a time. It's a really helpful start, but it's. It cannot be the only way your business functions. If you want to be an AI company, I posit that to be an AI company, you need three things. You need the data unification that about 10 of us have talked about so far. You just have to get the data together so AI has a fighting chance of helping you. You need tailored AI, not generic. So system prompt is one way to get there. But you need an AI that like knows your business, knows your data, knows your customers, your products, your voice, your conversations so you don't have to tell it every time. It's. It's a brand. Voice is a start. But what happened in that deal, what happened last quarter? And finally, you need to be able to take action on it. An insight is not actionable just because you called it that. Okay? You need to trigger something, you need to send something, you need to do something with it. You cannot just be a report. No one knows what to do with that. The good news is, if you use HubSpot, you are an AI company. HubSpot already unifies the data for you. If you run your business through HubSpot, it's unified. Thank you, CRM. If you integrate your calls and your emails and your chats and your social into HubSpot, we've done the work for you. You're welcome. It's very boring work, I promise you. Gmail, Outlook, any sort of service system. It's easier if you use HubSpot, but you don't have to if you want tailored AI. Great Breeze is continually getting better at using the data you put into it. You can enter data into it with system prompts, but it's also learning how to use the data that's already sitting there. Not the structured data, the data that's sitting in the CRM fields and the conversations that are running through it. And HubSpot is where action takes place. Ta da. Right next to where the reports Are is where the workflows are, it's where the APIs are, it's where the ecosystem is. See, something happens and you can just set up a trigger right there. That's why HubSpot's already equipped to help you with all these. So all of this sets the stage for today's launch, today's announcement. It's already launched until now. The feedback product in HubSpot is equivalent to surveys and most feedback is equivalent to surveys. Think about every phone call you have with a company. Please stay on the line after this call to answer a brief survey. They do that because surveys are useful, but also because they cannot figure out what happened in the call for all the reasons I just described. Even chat widgets will need you to give a smiley at the end of it. It's very difficult to pull exactly what happened in that chat. But today Feedback Topics completely changes that. Isn't that a beautiful box? This is the ability to ask any question of your data and have it continually answered for you. This is the ability to get anything you want auto tagged in your data. All you have to do is tell it what you want. It doesn't matter what part of the company you work in, it doesn't matter where the data comes from, doesn't matter whether you are in content or revenue or support. If you have a question that you would like to get asked all the time, every time someone talks to customers, you should put it here as a topic. Breeze will monitor new conversations as they come in. That's great. And it will apply topics to old conversations. So every time you edit a topic, it'll give you a preview of what that topic looks like on existing conversations. You don't like it, change the topic. And if you're thinking that sounds kind of like a lot, like that's not quite a prompt, but that's a lot of horsepower. You are right. And that's why it will help you write a good description. You can't save it until you get we get a shot at helping you write a good description because you are about to unleash a lot of firepower on could be tens of thousands of conversations every month being evaluated for what you asked to get looked at. A name and a description. That's it. Now there are some guidelines for good descriptions. What are you looking for? What's an example of it? What's not an example of it? Today HubSpot will listen to survey comments and all transcripts, meetings, calls, whatever you send in. Other data sources are coming imminently. All unstructured conversation data should be able to be scanned and answer your questions. When a topic is monitored, it comes up at the lowest level as a mention. So every time we find something, every time Breeze finds something, it's a mention. So you'll see what exactly was it that someone said, who said it, and when do they say it? And what was the nature of the mention? Was it positive, was it negative, was it. We can't tell. So it's not just did they say something about towels and locker rooms, but was it a complaint about towels and locker rooms or were they talking about how amazing your towels are in locker rooms? You also get, every time you load the page, a summary of that topic. What are the mentions talking about here? Even without doing a bunch of data analysis, you can get a. You can get a vibe check. Okay, I see what's going on here. Okay, you can do the usual filtering, etc. One level up, you're looking at all of your topics. You get a sense for what dimensions are, what the overall sentiment is, and you can page through them and do the usual filtering. So this is using a lot of the same things. If you used feedback before, that exists for surveys, except that now it's sneakily looking at all conversations. So if you work in support or success, even a lot of this, hopefully this sounds pretty standard, like I need to tag things, I need to listen to things, I need to report on what things are saying, what things people are saying. And some of you other leaders might be grumbling saying this kind of sounds like a support thing. And I'm here to tell you or remind you that it's not. And I want to tell you one more thing that might not be obvious. This kind of solution is very difficult to purchase. You know why? Because listening to other people's data at your company is difficult to to get budget for. If you want to listen to support data but you are not the support team, good luck. It is hard. It's hard to get the support team to buy something that isn't for them. It's hard to get the sales team to buy something that isn't for them if they already have a tool that works for them. Getting tools to read data for your purpose if you're not on that team is just difficult unless you're talking about a really big ETL system and data warehouse, et cetera. I've spent years working on this. HubSpot brings it all to you in one big hub. But if you try to do it on your own, it is a difficult problem just as a human problem, even if all the parties want it to happen, it is hard to do. Marketing can listen to sales not can listen to sales, but can pick out things from sales conversations. What were they excited about in discovery calls? What kind of things are clicking from the landing pages? I want to know that. Product always wants to know what's happening in support conversations. Now they can just kind of pick out the things that they care about. Leadership always wants to know a ton of things, including things that are going well, not just things that are going wrong. Feedback is company wide intelligence. It's not just about knowing what's going on at support, it's about every conversation. Sales support survey comments Anytime you're talking with customers, everyone who cares should have a shot at knowing what's going on and getting their questions answered. So just to recap, what is feedback topics first? It's always on answering your questions, listening 247 all the conversations that are coming in 2 it is tailored. Breeze is tailored, but this is very tailored. It's exactly what you want to listen to. And three it's action ready. When topics come in submissions, they are ready to be triggering other things. You want to alert the account owner when something happens, alert them something negative just got mentioned. Go look at might get might be a lot of emails, but you can do it. You can do a little alerts. So the old world was, you know, all these I really don't want to use the word data silo, but data silo, separate owners of data and asking for permission to look at it. Trying to get through all the red tape, et cetera. And with feedback topics at HubSpot, those walls just don't exist. You don't need to worry about it. The data's already there. Feedback topics kind of blast through all of those old barriers. There's also something interesting about it. Usually if you want to give access to those data sources, you run into permissions issues. I don't know if I want everyone to see my sales calls. I don't really know if I want everyone to see the support calls. You don't have to. Feedback topics will only pull out dimensions. You don't have to change your permissions at all. Also a very difficult problem to solve if you're going to do it some other way. So it's just the important parts, just the mentions, just the questions you want answered and you don't have to worry about giving anyone else access to it. Okay, so if you're wondering how to start this first connect your conversations to HubSpot. Ideally you're having them in HubSpot, but if you're not, that's okay. Just sync them over. If you're not recording your calls in zoom, start transcribing them. AI feeds on data. If you are not recording what's happening in your business, you do not give it a chance. Sync your emails, record your calls, type your notes up, create deals, close deals. Create the data that gives a lets AI help you and put it in HubSpot so it can all come together. 2. Join feedback topics Beta what that means is if you are not on Service hub already, trial it. Trial one seat and then join the beta that gives you access to feedback topics, to asking any question you want in your data and then three Define the topics you care about. Hopefully you don't run out of topics. You'll run into the limits. Breeze will handle the heavy lifting for you. So now what? What kind of topics do people create? I'll give you a preview. One of the first things companies tend to do is want to know what the competitor mentions are. Who talked about our competitors? What do they say? Just enter your competitor names. We'll tell you when they said it and what the sentiment was in every deal, on leads, in success calls, whatever. Why that deal churn? Well, there was a competitor talked about pricing discussions like we talked about it earlier in this session. When do they talk about pricing and what is it like? Upsell signals, Churn signals Service praise. When are they really happy with us? I want to know. I want to give kudos. It's very difficult for me to track that. When are people really happy with our sales folks? I want to know that anyway. So the the sky's the limit. I'm happy to answer more questions about what kind of topics people create, but it sort of depends on what your role is. So if you've never had to had the opportunity to create topics, sometimes it can be hard to know what to do for the first time. So think about it. So that's it. So as a wrap, I invite your questions. If you think this would serve you well, if you think it would fall over in the real world, I want to hear your challenges. Find me afterward. We can dance to some of this great conference music and we'll talk about it. That's it. Thank you.
