Podcast appearance
Mastering Online Review Insights and Marketing Analytics
How the review ecosystem connects conversion, personalization, sentiment, and action.
How the review ecosystem connects conversion, personalization, sentiment, and action.
Alex Sofronas recognizes that the value is in the whole review ecosystem, then presses on how review acquisition, public responses, granular sentiment, and competitive evidence can be measured and used together.
“I love that you really focused on the review ecosystem, the whole thing, the whole vertical.”
Terminology note: this conversation uses the earlier word “facts.” RightResponse now calls the verified business knowledge used in a response “messages.” The transcript preserves the language used in the original recording.
The review ecosystem produces both public marketing evidence and private operating evidence. The episode connects what customers write, what the business says in response, which topics are changing, and how marketing teams can use that evidence to improve decisions.
It identifies the topics and phrases customers discuss, determines sentiment in context, and tracks patterns by location, competitor, or time period.
The average hides the reasons behind the score, including which products, services, employees, and experiences are improving or declining.
Responses are public conversion content, while their performance and the source reviews can be analyzed alongside acquisition, location, and competitive evidence.
It uses appropriate customer or transaction context to help the customer remember the specific experience, rather than adding a first name to a generic template.
The original Marketing x Analytics episode and YouTube video are linked from this page.
Explore the Review Sentiment Analysis Platform
This transcript was generated from the original episode audio and lightly formatted for readability. Minor speaker or wording errors may remain.
[00:00:00] Hello, and welcome to the Marketing Times Analytics podcast. I'm your host Alex Sofronas, and today we're on with George Swetlitz. George, would you like to introduce yourself? Sure. Yeah, so it's great to be here. My background is quite varied. I everything from working at McKinsey to starting a division of Fortune 50 in Poland after the fall of the Berlin Wall in 1992 to being the CEO of a private equity backroll up in the healthcare space, which led to the current things that I'm doing around RightResponse AI. I had a co-founding that with some others to bring a modern reputation management to the industry. Wow. Okay, tell us more, response AI. Tell us what value it provides and a little bit about the business itself. Yeah, the genesis of RightResponse AI goes back to an IRCEO of a company called Alpaca, which was this roll-up. And we had 220 locations and one of the things that, you know, that we learned, like one of the key observations that we had was that that our clinics that had better reputation for forward better with paid social and paid ads. Right. And so that led us to try to understand how do we deal with reputation at scale? So this is back in 2021, which is a lot of worlds away from where we are today. But in 2021, very difficult to do sentiment analysis at scale in a cost effective way. Platforms were using templates for review response, right, which sounded very repetitive and robotic. Right. So we were very challenged with how do we do this at scale? We never really quite figured out. And so when ChatGPT came out, I got my team together and said, I think we could, I think we could actually solve this problem. I think we could develop a product that solves the problem that we faced and bring it to the market. And so that's what we did. We spent about a year, a post, the launch of ChatGPT figuring this out, building, building a product and then launched it in 23. So how does the product work? So essentially what it does is it downloads your reviews, whether it be from Google or any platform source.
[00:02:19] And then we do a ton of things with those reviews. We analyze them for sentiment. And we do that in a very specific way. We, we, we develop a set of topics for the business based on the Google business profile and the description of the business. So everybody set of topics is different. It's unique to that business. And then the business owner can change it. They can modify those topics to whatever makes sense for them. So we analyze and all the reviews. So you have a very kind of bespoke view of your sentiment. We analyze the last thousand reviews. And we identify the things that people talk about. And then we say, OK, if that's what they're talking about, how can we have a very rich response to them? And so we make one up. There might have been an answer in a prior review response pair, in which case we use that. If there wasn't, we make one up and then we tell the business owner who fix it. And so then what happens is when reviews come in, we analyze the review against all of those facts. We find the facts that are relevant. And we incorporate that into the response. And we do that across a whole series of customization so that every response sounds like it was written for that review. So that's the core. OK, yeah. And as I guess as a baseline, why should entrepreneurs and businesses be responding to reviews?
[00:03:55] Yeah, no, exactly. That's what we found back at Alpaca. What we found was that reputation impacted the take up of an offer. What's going on with the background? You said, there's an ad somebody sees the ad they come to your site. That's interest. So why is there a difference between our clinics, the ability of them to convert that into a visit? That's not the last step for the customer. The customer then does their own research. How do they do research? They read reviews. More people read reviews than visit websites. So if you're trying to capture that customer, you have to capture them where they are. And where they are is reading the reviews. They're doing their research. They're reading the reviews. They're making that decision as to whether they want to choose you. And so the kind of the learning is that the epiphany was the review, the people reading the reviews are the closest thing you have to a customer to a perspective customer.
[00:05:05] So when you respond to a review, you're not really responding to the customer who left the review. You're creating a response that will engage the next perspective customer. That's the shift. So most people think I have to respond to reviews. That's not the point. But point is somebody that your next customer is reading those reviews. And that's the person you're talking to. So in our business, which was an audiology business, if an elderly person said I had a hard time finding parking, oh my, that's horrible. Because then every other 70 year old that's reading that is going to say, I'm not going there. I'm going to have to walk. I can't. I don't want to do that. So the answer is, oh, the parking space is behind the building and we have a special entrance. So that perspective customer reads that they, oh, yeah, okay, that's fine. I can go.
[00:05:59] So it's thinking about reviews as part of that overall kind of marketing perspective. That really is a key to driving. But I would say the key to driving organic growth. Yeah. And that gave me an idea. Does that extend to online comments about a business? Let's say on Twitter or on any social media where somebody might criticize the business. Would there be a value to also for the business to respond to that? Yes. Absolutely. It's everywhere where people are doing the research. Now we don't do social yet in our business. But there's just, it's just a huge, there's such a huge, it's such a huge space, all the things that you can do.
[00:06:42] But you're exactly right. That's on our plan is eventually to engage with social because that's it's just a comment. But in a sense, it's a review. And you have to engage with those as well. Absolutely. And should the, this might be getting to into the weeds, but should the person that's responding to the review be the business itself? Or given the option should it be a particular person within the business? Like the owner or in a particular, is there a particular vertical of people within the business that would be good to be responders? Like who should respond to the review? So it's whoever, whoever is the guardian of the unique attributes of the business.
[00:07:27] The owner knows what, so the way I like to describe it to people, customers, and I talk to them is if somebody walked into your location and asked you a question, what would you say? And who would know that if your recipe came, if your lasagna recipe came from your great grandmother, who knows that? And what we're trying to do is use AI to personalize its skin. So I talk about this for a second, because I think it's important. People are frustrated with AI often because it's a fake personalization. Like I'll get inbound email, which I don't mind getting. But it'll be like, oh, I saw that you went to Penn.
[00:08:13] Did you like this restaurant? Of course, I went to Penn 35 years ago. So that restaurant didn't even exist when I went to Penn. AI fake personalization. What we're trying to do is AI real personalization. That story about the grandmother, where's the parking? What are your unique differentiators that if you're a busy business owner, you don't have the time to go in and write that for every review. But AI can take that knowledge and personalize its scale.
[00:08:48] It's a very real personalization. And so I try to draw that distinction. But it's really to answer your question. It's really that the individuals in a business might be in the marketing department in a larger organization, typically in marketing, that know what are the brand values. And how should we extend those brand values into the review itself or into the review response? Yeah. So you mentioned that you do like a topic clustering. How does that work? So you're talking about for sentiment analysis.
[00:09:22] Yeah. Yeah. So essentially, it's a multi-step process. And so when we talk about AI, one of the things that I've learned building this is that the smaller you can make each individual ask the better off you are. Yeah. Because AI, when you overwhelm it, it gets lazy. It's just, it's a very weird phenomenon.
[00:09:50] And so what we do is we do the smallest bite-sized piece that we can in an economical way. And where we need to do something large, we use the best model available. So if we're trying to develop topics, we'll use something like 103, very expensive. But we only have to do it one time. We ingest 600 reviews. And we have this multi-step with checks and all these things to develop a set of topics that are unique to that business. And then we have other agents that are checking them to make sure they're right.
[00:10:33] But when we're trying to process a review, we do that at the smallest level possible. And we'll use a cheaper model that will be more cost effective because we're doing that on every review that comes in. Yeah. So really this balance between the complexity of the ask and the model that you're using and who's checking behind it and whether you need to check behind it and all of those things. But that's the approach that we take. That's really interesting. So you're using AI both for topic clustering and analysis and review answering.
[00:11:12] And request writing. So another feature that we're just launching now is AI personalized review requests. Everyone gets you get them every day. You go somewhere and you get a request. Hey, you know, you're just here. Could you leave a review? You get more and more of those. And so now there's more competition for your time. Because you're not going to write everyone that you receive.
[00:11:40] So what we're trying to do is use personalization. To make that request more personalized and more emotional, right? So, for example, you're an auto dealer selling a car. You know what car they want. You know why they bought the car. They told you. Right?
[00:12:01] Like you can take a picture at the closing. You can do all of these things and you can then put that into our software. And it writes this. Hey, this is Jim. It was really great to see you. We know this is the second car you bought from us as we appreciate your loyalty. And we took this picture when you park for the car. We hope you can keep this as a memento of whatever.
[00:12:25] A very emotional. Now you look at that and you're like. And then they say, look, I would really appreciate it if you leave a review. It makes a real difference. And you'll let other people know that this is such a great dealership. And then we even have an SEO piece that you can put at the bottom if you want that says. Hey, if you're looking for inspiration about what to write.
[00:12:45] And then we populate that with. Questions that kind of implicate certain keywords that might be valuable to have in the review itself. Now you look at that and you say. Yeah, this is actually meaningful to this guy. Really help me out. Thanks for the picture, whatever. And you write this review. So the idea is how can you use the real things right those fields that we talked about the real things.
[00:13:10] How can you use real things. To have personalization at scale. Yeah, and that's the trick. That's really interesting. I love that you really focused on the review ecosystem. The whole thing like the whole vertical. I'm curious like you you mentioned that social is next.
[00:13:31] What else are you imagining is next in the development of this. Yeah, no, yeah. So one thing that we're working on is. So let's say you get that request. And you don't answer. So normally a couple of days later, you're making a follow up. So on our path is you get a follow up.
[00:13:52] But it takes you to a slightly different place. It's a web. And it says, would you like us to. Would you like us to write this review for you if you're busy pressed for time. Just type some things in doesn't have to be full sentences. Just whatever. You like love gym, great dealership, wonderful experience.
[00:14:12] And then we'll go write a review. And bring it back to you. And then say, hey, edit this, whatever. And then you can copy and paste this into Google. Make it easier for people. Who. A lot of people they went in the research we've done.
[00:14:29] They've just said, look, I just don't have. I don't have the time and mental energy, right. For this. So that's one thing. Yeah, that we're pretty excited about. It's like review AI talking to AI. But from our perspective, it's all grounded in real things.
[00:14:47] They have the type of phrase. There has to be something about the individual, right? It's it's actually real. So that's that's one of the big things that's next we've been playing with computer use. So that people. When they're trying to when a company is trying to. Respond to review with a review platform that doesn't have an API to take the review response.
[00:15:14] They have to copy paste that review response that takes a lot of time. So we're trying to use computer use. So that AI will actually copy paste. And that's very complicated. Yeah, computer use quite isn't quite there. Yeah. That's it.
[00:15:34] So you're using review APIs to post the reviews. Yeah, to Google, for example, Google has a very powerful Google business profile API. And so we're using we use that to pull and push. Oh, interesting. Yeah, is that available to everyone? Do you need like enhanced permissions or anything? Nope.
[00:15:59] No, anyone with the Google business profile can. When they log into our system, they give us authority to let to use their API or use their key. And then we pull and push to Google. Okay. Oh, interesting. So they put in their API key. Yeah, it's probably it's like part of a Google authentication, right?
[00:16:20] They sign in and they give us permission on a particular location. Okay. Or a set of locations. And then they can pull that permission whenever they want. It's very Google is a very secure way of doing it. Interesting. But not many, not many review platforms have APIs like that.
[00:16:40] So it makes it more complicated. Gotcha. The other thing that we're doing a very interesting little tiny thing is when you go to Google, they don't show you the reviews based on newest first. They show them based on their view of relevance. And that relevant review might be quite old. So one of the next things in our pipeline is to we will go in and look at the top 20 or 30 relevant reviews.
[00:17:12] Mark those in our platform. And then analyze the response and say if these are the ones that many people are looking at first. These are the ones that you should improve. And this is how you should improve it. So that the reviews that people are seeing give you the best possible appearance. Yeah, it's almost like how FAQs work. It's a very similar system like FAQs are just almost like preemptive answers to reviews where people are saying,
[00:17:49] this is a challenge I have is just sitting a challenge instead of asking a question, but it's a similar concept that's where people go to figure out answers to questions they have about the business. And every one of the FAQs can be a fact in our system. So if somebody raises the topic of that FAQ, it automatically gets placed back into the response 100% 100%. And that's another area. So for example, Q&A, Google Q&A. People can ask questions.
[00:18:21] And then there can be answers. Often those questions have been answered before or they're already somewhere. It's just somebody asked them the form of a Q&A. And so again, we can look at those and decide whether or not we have enough information to answer that Q&A. Yeah. That's very interesting. How are you marketing your business?
[00:18:46] For us, it's entirely SEO and outbound email. We don't do any paid for the same reason that my old job paid is just very difficult, especially at the price points that we're at. It's very difficult to get return. Yeah. So we're all we're all about SEO. We're all about outbound trying to find people at the right moment.
[00:19:08] Yeah. That makes sense. That makes sense. And kudos. It's definitely the harder path, right? Like the easy option is I'll just pay for traffic and yes, it'll lead into margins, but it's easier. So that's the that's the harder and I think better path in the long run is really investing in SEO.
[00:19:27] One of the best things we did is we put a free review responder on our site. We use like the least expensive model. And we give people some options to tailor things and they can use it as much as we want. We do thousands and thousands of free review responses every week. But Google looks at that and says, man, these guys are important. There's so many people coming to this site. And we rank very highly for review responder and review responses and all of that because.
[00:20:01] Because of that free review responder. Interesting. So that's a almost like a hook of some kind before the user even creates an account. I'm assuming. Yeah, yeah. So we have we get a lot of customers paying customers for more free site. They come.
[00:20:19] They use it. We pop up something that says, why are you doing it the hard way that kind of thing to try to get them to look. But a lot of people they just don't want to pay a penny and they'll just come back. A week after week month after month to get these free responses. Which is fine because it's just websites. Authority from our perspective. But some people come there and after a while they're like, you know what I am doing it the hard way.
[00:20:44] And so then they'll sign up and they'll create a paid account. Yeah. That's very interesting. Are you using AI to actually code the app or is it more just using the API. Integrations for us. So our dev team does use AI to code. So we we actually have seen a real increase in the productivity of our team using AI.
[00:21:11] It's it's interesting. We talk about this a lot like which is it the juniors that had benefits similar as it's the seniors that benefit more. But there's a lot of discussion about the right way to use AI in the development process. But I'll give you another example. Our development team is really busy. We were I was very curious about how operator would work. And in particular how much it would cost.
[00:21:38] And I don't know whether if you have a pro version of chat you can use operator. You can go in and use operator. But it doesn't tell you what it costs. The only way you can understand what it costs is if you use the API. So I didn't want to bother the dev team. I've never written the line of code in my life. I sat down and I brought up.
[00:21:59] I said explain that. And I said I want you to talk me through everything that we need to do. So that I can run operator from my desktop through the API. I'm downloading Python, Git, play right all of these tools installing. Then I'd run it and it would fail. And I would give that to chat CPT and it would change it. And now today having never in my life written a line of code.
[00:22:27] I can run operator from my desktop. And after every interchange I can find out the input tokens, the output tokens, the number of screenshots. Everything that I need to know to understand costs. I did myself. Wow. I just find that.
[00:22:45] Just amazing. Now when I think about how long it took me to do that, how many times when I tried to correct an error failed and I had to go back. Sometimes it would be six or seven times before it would finally get something that an experienced person may have seen that much faster. But I eventually got there. That to me is a power.
[00:23:09] It's just this tremendous power that that AI gives you. Yeah, there's no longer a moat around development. Like development used to be this very guarded, almost like it's a topic, a subject area where if you weren't a very technical person, you're just not breaking in. You're not getting. There's this very large moat around that you just need in many cases years of
[00:23:39] research and studying to get through. And now it's like a few days. And you can get anything off the ground. The skill you need now is the patience and clear writing. You just need to be really clear with what you're asking for. And you need to be patient with. There's going to be a lot like the AIs right now.
[00:24:06] Most of the AIs are like very technically smart, but not very wise people. Like they can they know how to do this stuff, but they'll miss the most obvious things. Like I've been by putting an app and I'll ask it to make a change. And then it's, you know what, the file is too long. I'm going to rewrite it.
[00:24:25] And then it takes out half of the functionality. And I'm like, that's not no. We want to add to the way all the work. That's exactly this whole idea of it gets lazy at the worst time. Yeah. Like I gave I'm writing a blog article yesterday. And I gave it this very detailed.
[00:24:47] Bullet points every age to had a series of bullet points. And then we talked about what's the right structure. And there's our overlap and did a great job of doing that. And I said, I'm going to assign every one of those bullet points to what we talked about. And give me an output. And it gave me like this paragraph. What started with a five page document turned into a single page paragraph.
[00:25:11] So lazy. And then I realized I used the wrong model. I should have used O3 for that. Because it would have listened to what I said about assigning the bullet points. And really understanding you have to have some experience in what models do I use when. And even in a single chat, you're like constantly switching between models because you're trying to accomplish different things throughout.
[00:25:38] You can't use O3 if you're at the point where you're trying to write a paragraph because it's terrible at writing. So you need to go to something else. Yeah. Is that the model determination for me is typically all think about that when I'm determining the length of the conversation and how much I need to process because these like super powerful models. They're good at like solving complex problems.
[00:26:05] But then they'll hit limits so fast if you keep going. So then like the more efficient models typically they can run for a lot longer. For those kinds of use cases. So yeah, I totally agree. I think it's critical to pick the right model for the workload. And then I think based on what you said before, the real. As a, as you're developing something, you have to be thinking down the road someday.
[00:26:29] Hey, I was going to be able to do this functionality for somebody. So what's my competitive? What's the reason that they would buy the software? What will we be able to do that other people will be able to do on their own? And so you constantly have to be thinking about where's your moat in a year or two years? I think it will be moving slower than we think, but still every day there are some advancements. And you have to think about how do you build a moat into your software?
[00:27:03] Yeah, that's interesting. I can see a future where the AIs. They almost work together. And the user is not just using one AI, but it's like a multi AI approach that will dynamically select the best. AI for that use case. Similar to when I started my career, I worked for IBM. And everybody was coming out with their cloud solution.
[00:27:37] And IBM's strategy was to go for hybrid cloud, understanding that businesses are complex. And most of the biggest companies are not going to be able to standardize to a single cloud. And so IBM was going to be the leader in hybrid cloud when you have multiple clouds working together. They will be the best at identifying how those work together and integrating them. And I think that is going to be, we're going to see a similar development in AI where there is just not like businesses won't be able to just use one AI. In the case of my day job where I work in marketing analytics for a software startup, we don't just have one AI. We have two. And oftentimes for my coding tasks, I'll need to use both.
[00:28:23] Not because sometimes because we have Gemini that uses a larger context window and like I do need that sometimes. But also because I hit my limit like every day on with these AIs, especially for coding tasks where you have hundreds of lines of code, they can't sustain a conversation for hours. So I'll need to bounce between the two. And so that hopefully will be solved eventually, but the idea that we need multiple AIs, there's many reasons for it. So I definitely see that evolving in the future. Right now, so single AI focus, they don't really interplay together. Yeah, no, you have to, you have to do that programmatically. Your code has to call the right model for the right use case.
[00:29:13] And that's a similar. So for example, we have areas in our software where we, we randomly inject various things just so that it's not always the same AI is terrible. If you say, if you say to, if you have a prompt and then you say use one of these three things, it will always use the first one because it's, it doesn't know that the last time it used the first one. So we'll always, so you have to inject the one you want. And you could have another AI agent that's making that decision, or you could just program. And programing it is just much cheaper than having an AI just decide between these three things. And so there are still a lot of cases where it's the mix of programming an AI that is the most cost effective solution. And that I seek down the road as unless AI gets super, super cheap, there will always be an advantage to this kind of combination of programming an AI. Yeah, totally agreed, totally agreed.
[00:30:16] Thank you so much, George. And this has been an excellent conversation. It's been great to learn about your business. I think you're in a very unique place. And yeah, really interesting. And thank you so much for the chat. Yeah, really enjoyed it. Thank you. Awesome. Thanks everyone for listening. We'll talk to you soon.