This is a transcript from the AI and the Future of Work podcast episode featuring Mahesh Ram, CEO of Solvvy (acquired by Zoom), discusses the future of conversational AI for customer service

Dan Turchin (00:25):
Good morning, good afternoon, or good evening, depending on where you’re listening. Welcome back to AI in the future of work. Thanks again for making this one of the most downloaded podcasts about the future of work. If you enjoy what we do, please like comment rate and share in your favorite podcast app, and we’ll keep sharing great conversations like the one we have today. I’m your host, Dan Chen advisor insight finder, the system of intelligence for it, operations and CEO of people reign the AI platform for it and HR employee service. We now come to expect that whether it’s our credit card company, our airline, our wireless vendor, they won’t require us to sit on hold, waiting to speak to a live agent, just to get basic questions, answered common requests, like, you know, password resets, check my balance, upgrade me. What’s my billing cycle. Those things can almost always be handled by virtual agents today.

Dan Turchin (01:24):
We’d almost always prefer resolving routine issues with a virtual agent rather than dealing with a live agent, but it is absolutely the case that that represents 180 degree switch from how our attitudes were toward virtual agents. Even a few years ago, the technology behind those solutions is really complicated and it requires a combination of dialogue editors and workflows and third party integrations, and often conversational AI trained on historical data from customer sessions. Well, today’s guest is an expert in the field having founded. So a leader in virtual agents for customer service back in August, 2015, Mahesh Ram has grown the team to nearly 80 employees in raised more than 16 million from some amazing investors, including one of my favorites and a previous guest on this podcast, Rio Driscoll from scale venture partners before Sovi mesh spent more than a decade at global English before it’s acquisition by Pearson in 2012. And without further ado, it’s my pleasure to welcome Mahesh to AI in the future of work Mahesh let’s let’s get started by having you maybe share a little bit more about your background and how you stumbled into this space.

Mahesh Ram (02:45):
Hey, Dan, first of all, it’s great to meet you. I’ve been a fan of the podcast and as I mentioned to you earlier, I enjoyed it while riding the bike and now it’s my distinct pleasure to be on with you. And I really appreciate the opportunity. Talk to your great audience about solving and about the future of conversational AI and the support automation space. So I I’ll brief, I’ll be brief about my background. My, my background has been very much geared around applying technology to automation and improving business process across a wide range of disciplines. So you know, the prior venture to this where I was CEO was a company focused on business, English, learning and acquisition, using automation for billion and a half people around the world who were trying to improve language skills, but didn’t have access to a human teacher in most cases.

Mahesh Ram (03:39):
And in the business context, we were focused on global companies and we applied a wide range of very leading edge technologies, such as speech recognition understanding of, of language inflections and patterns to be able to actually help people dramatically improve their business English communication skills. And that we ended up with about 450 of the global 2000 as customers in a, in a and of users in over 120 countries. And it was very gratifying that we were able to use automation to deliver something that could never have been delivered, you know, a few decades ago, which is to make someone fluent using technology in a business context. So that was, that was the last company global English. And prior to that, I had actually started and worked on automation for legal tax professionals in the automation space, again, automating extremely complex tasks using technology that could do things like doing a filing for a corporation that would normally have taken 20 days in the past, down to a few hours and thereby speeding up a whole lot of business process. So my entire career has been in these B2B SAS subscription companies that are all about automation and improving business efficiency. I like to say that my entire background is about time. You know, the most precious, I always say time is the most precious commodity all of us have. And I’m kind of a, I personally a little obsessed with how do we give everybody back more of their time? And that’s always been a driving common factor in everything I’ve done.

Dan Turchin (05:12):
So I read some of the very impressive case studies on the soy website, take us through a customer that you’re most proud of and how to use the platform.

Mahesh Ram (05:20):
Yeah, so, so I think the first thing I’d say before I go into customer, you know, customers themselves is I think we’re, I think in this, I think so’s leading kind of the third wave really of chatbot technology. I think the first wave of chatbots is the one that everybody despised probably the most, which is the idea that, and you still see a lot of this, unfortunately out there where it, I don’t understand you. I don’t know. You know, I don’t know when to, it, it, it, I guess the classic answer is a lot of deflection and not a lot of satisfaction. And so CU and companies that adopted these U usually had to make that false choice, the false premise of either I more people and keep them out of my contact center or out of my agent’s hand and save a lot of money that way, but lower CSAT, or I let more people in to talk to my agents and I increase my cost.

Mahesh Ram (06:08):
And to me, that was always a false dichotomy. So when we thought about creating Sovi, our, our first vision remains our vision today, which is how do we give consumers back their time, deliver a better differentiated experience that is personalized, that is consistent with what their expectation is. That gets out of the way when they actually don’t want to talk to the chat bot and they want to talk to a human, are they need to, and we talked, we started thinking about this as concierge, like experiences. And I think in order to deliver that, what we’ve really built is a conversational AI automation platform. That includes a chatbot front end, but that allows brands to, to configure experiences. So I think of as a concierge level experiences, I understand what you’re asking. I determine the best path forward. Maybe I give you the answer right away.

Mahesh Ram (06:57):
You asked me, what are the store hours in Madison, Wisconsin. I should be able to answer you that question without waiting for a human. You tell me that you feel threatened in a, in a, in a ride share. I probably don’t wanna selfer you. I wanna understand your intent and get you to a human being faster. That’s not, self-service, that’s not what you’d expect out of a chat bot in most cases. And in other cases, you’re telling me that you don’t know you, you tell me that you are having challenges with you know, your iPhone and troubleshooting your new device that you’re trying to connect via a Bluetooth. I ought to be able to understand that specific issue for the product you’re asking about and guide you through a troubleshooting. And if I still can’t help you get you to a human agent who can then help you faster and better, or maybe get you to the right agent.

Mahesh Ram (07:39):
What I’ve just described is a set of experiences. That would be very similar to if you went to the, to the lobby of your favorite hotel in a five star resort and said to a concierge helped me and it would under the concierge would understand you would have knowledge would have access to some data and would actually be able to configure an experience for you. And that’s what we think that a really conversational support automation platform should do. So what’s a customer, what’s a typical customer example. Well, let’s use an example of, of a company that I admire greatly, a calm, which is in the mindfulness and wellbeing and health and wellbeing space, and com provides, you know, a mobile app for meditation and wellness to, to hundreds of millions or tens of millions of people around the world. And again, these people, when, when we started working with comm many years ago, they weren’t very big.

Mahesh Ram (08:31):
They didn’t have a contact center. I think they were, you know, 20 people or something, and they’ve grown massively since then, but they always had a vision for saying, Hey, the people who use us are, you know, would like to be more self-sufficient they wanna be able to use the app where they need it. They have a question about how do I do X or Y they don’t wanna wait for a human agent. And so we set about really building those concierge level experiences in multiple ways for calm that the persona of the user that understood the need that could interrogate the knowledge bases and FAQ content that they had, and actually deliver question answers, which is the original PhD work of my co-founders at Carnegie Mellon that spawned this. But in other cases, guide users through a specific experience of they, maybe they wanna refund, or they need a, they need money back.

Mahesh Ram (09:14):
And then it understands that Dan is, is, is a user who bought through it iTunes. And so the refund policy is different. Then if Dan bought through the Chrome store or via, you know, direct from the retailer, right? So all of these are differentiated experiences, but the system can actually deliver that experience without you having to talk to an agent in many cases, or in other cases, get you to an agent if that’s what it is. And now com is doing amazing extensions with corporate partners like Kaiser. And so that’s a differentiated experience again. So, so what we’ve been able to do is essentially help them scale from, you know, single digit thousands, a a hundred thousands of customers to, to tens of millions of users around the world, and to do that incredibly scalably, improving the CSAT while actually reducing their cost, therefore improving their profitability. So that’s one example. We have many, many others of course, that that are equally compelling. But that’s a, that’s I think a good example.

Dan Turchin (10:13):
You gave an example that is, let’s say unambiguous, what are the store hours for Madison, Wisconsin, but you, and I know that the challenge of doing true NLP high quality NLP is that there are a thousand ways to express the same intent, right? in the field that you’re in using, you know, an example like com or I’m sure other customers as well. How do you manage the complexity of kind of dis utterances that aren’t always as unambiguous as the one that you gave?

Mahesh Ram (10:48):
It’s a great question. And it’s one of the, it’s one of our core strengths in why we’re a true AI company, as opposed to a scripted decision tree company was a lot of the players in the chatbot space have become, this goes back to the core PhD work that my co-founders did, which was groundbreaking work that was featured in New York times and scientific American and others around this idea of semantic similarity to be able to take data that that’s already available to us. And, you know, we work with companies, so they give us their prior conversation datas an example, we can pull it right outta their CRMs and other system. And we’re able to, to, to use the proprietary AI algorithms, to be able to understand core intense and actually predict intent from that the million different ways in which people express an issue.

Mahesh Ram (11:35):
And I’ll give a real example of one of our large customers that we probably do 25 million conversations a year for them, there are home food delivery business, and they operate in nine or 10 languages, 11 or 12 countries, massive amounts of volume coming conversation, data, but users express issues in ways that have nothing to do with the typical keywords that you might expect for a spoiled ingredient. Let’s use spoiled ingredient. People say things like the chicken is spoiled. The broccoli is wilted. The, you know, the ice has melted or, you know, you, you can go on and on, right. And what we’re able to, the semantics similarity training that we do, the proprietary ad that we use is able to instantly understand that all of those are intense around a bad ingredient or a spoiled ingredient. Then we can then walk users down the path for a spoil ingredient, which might be asking a second question about it, understanding what is it, a primary ingredient or a secondary, if it’s a primary ingredient like the steak, I might give you a full refund.

Mahesh Ram (12:41):
If it’s a secondary ingredient, I might give you a partial refund as an example. Right. but the ability to predict that intent better than just anything else in the world is one of our core strengths. Now, let me also say that the predicting intent can do a couple of things. One is it can get you into the right journey. The journey to report is spoil ingredient two. It can actually be used to do topic based categorization for the brand. And we had one of the LA largest consumer electronics companies in the world within two hours of releasing a brand new product. We started detecting spikes in power issues that related to the firmware release they had made in the, in the electronics product within two hours before their product team knew about it or their engineering team knew about it because users were coming in and asking about this issue.

Mahesh Ram (13:30):
They weren’t saying firmware is broken. They weren’t saying even the power. They were just saying, what blue lights flashing? I’m not seeing the stream on the device. And the system automatically recognize the topic based categorization and gave them a spike alert. And they were able to go in and see these questions and immediately go to product and engineering. Now that they couldn’t self serve those issues. They couldn’t solve them, right. It was a defect, but they were immediately able to push the firmware upgrade and solve the problem before it became a big issue before it flared up on Twitter. So, so the idea of detecting the semantics similarity in the topic base has multiple use cases. The separate AI algorithm that we use, which is again, very proprietary and powerful is the ability to then do the question answering out of the unstructured knowledge based sources.

Mahesh Ram (14:15):
So typically most companies have big FAQs. They have KBS, they have multiple KB. They have lots of places where the store hours for medicine, Wisconsin might be our ability to understand that question that you’re asking about, you know, how late are you open? And then, you know, in Madison and extract that specific paragraph or sentence from a huge Corpus of text is another massive differentiator. So, so there’s really two AI, big technology capabilities here, the question answering and the intent understanding they, but for the user, the user doesn’t care, like they just they’re just expressing their issue the way they would, they were talking to you or I, so it was a long answer, but I hope it answered your question.

Dan Turchin (14:56):
Now, the good news about using NLP for customer service is that the way a customer can express their intent is unbounded versus say a traditional IVR tree. But the bad news is even though we, as consumers have come a long way in terms of accepting virtual agents, I would go so far as to say our tolerance for, you know, let’s say a gibberish answer provided by the virtual agent is still very low. How do you think about mitigating the impact of potentially responding with Gibb versus the great experience that you generate by responding with the right answer? How do you apply kind of a confidence threshold or some way to know whether or not to respond?

Mahesh Ram (15:41):
So I think, I think first of all, I think there are two answers to this question. One is about the confidence threshold, but I’d say there’s a bigger overarching capability. That’s a hard capability to build. And I think we have it. And it’s quite unique. It’s the ability for brands to orchestrate the experiences by persona segment or personalization and even by issue type. So, so the example I gave earlier, you know, if you’re in a, if you’re in a ride share and you feel threatened the right orchestration for that should be that the brand should be able to say, when people are threatened immediately skip past self-service, you know, get it to a trust and safety agent, give them a phone number, let them call, right. If I’m a consumer, that’s the experience I want and expect. I don’t want the chat bot to come back with a gibbery answer.

Mahesh Ram (16:23):
I just wanted to get outta the way. But in order to do that, you have to have powerful technology to be able to do that. And you have to make it possible for the brands to control the orchestration layer, which we do. That’s number one, number two is com brands often wanna segment and personalize by a lot of different criteria. Dan might be a V I P whereas Mahesh might be a free trial user. We both ask the same question. They might, you know, they might actually wanna take Dan down a different journey, ask him few different questions. Maybe give him an offer even, and take him down that journey. Dan will appreciate that because it knows that it’s Dan, right? So that’s the second level. There’s a persona level. The third level is specific individual persona. So I don’t know that you’re Dan you’ve asked me a question.

Mahesh Ram (17:05):
You tell, asking me what status of your order. That’s been delayed by a week. I don’t know that who you are. I ask you to give me some identifying information. Our chat bot actually hooks up to a whole range of data sources in the enterprise and says, Dan, your orders was shipped a week ago and it has, and you know, it’s here. It is in FedEx. And here’s what the NetSuite system says. You know, where it is in treatment. And you still might say that that doesn’t help me. I haven’t seen it. Where’s where’s the package. and at that point, the brand can create an orchestrated experience to get out of the way and say, let me get you to a human being who can help you, but now they know who you are. And so that agent is able to help you three to 10 times faster because they know you’re calling about order 1 0 1.

Mahesh Ram (17:45):
It should have been there last week and you didn’t get it. So what I’ve illustrated for you is a range of capabilities, but it comes down to one principle, which we have observed here at solving, which is if you put the user first, if you put the consumer first and you are not a deflection tool, but you’re really about getting the user to the right place at the right time, as fast as you can, you’ll always succeed. So what you never see with our chat bot experiences is hiding the support options to us. That’s an athema because it doesn’t make sense. It’s not what I want as a user. And so I think if you do that, well, what you do is you get better and better over time and users are more willing to engage with you, cuz they know they have the backup option. It’s when you hide the backup option and you keep saying, I didn’t understand. You try again, try again that they get, they wanna throw the computer out the window or they wanna get very upset with the brand. Hopefully that answers your question

Dan Turchin (18:39):
Spot on. Absolutely. So we’ve talked mostly about intent detection, but the other big part of the interaction is getting the right answer. So discovery or retrieval now in a simple case where I’m asking, let’s say, you know, for, you know, a fact and an FAQ, something like that, it’s usually gonna be a one shot interaction. You ask a question, I understand it. I give you the answer we’re done, right? Or you, you go and you ask another question, but what about the case where you need to gather additional information? You know, what’s my balance. Well, you have multiple balances, which one, you know, things where it’s interactive or where there’s kind of a multi turn dialogue that requires maybe querying a remote system to solve the system. Does Soly also build out those kind of more complex dialogues beyond just a kind of a one shot answer?

Mahesh Ram (19:33):
Yeah, absolutely. I mean, we’re solving issues that would that in that the most experienced agents would’ve otherwise have to solve and that involves a few different aspects, right? It involves first this what we call dis Ambi. I need to understand what the nature of your issue is. I might need to ask you two or three questions. I need to, to do that, but also data infused experiences. So I might need to know that you are in Montana and you’re asking about the sales tax in Montanand you’re asking for an electronics product, that’s over $250 and that you’ve already purchased a product. I’m just giving four variables, right? Each of those is a, is a potential variable that I might have to get from multiple systems. But in order to do that, I first need to ask you something I need to G I need to come back and forth with you.

Mahesh Ram (20:17):
And then based on your answer, I might take very different paths, right? You come back, the product comes back and it’s a hundred dollars product. I might give you a different path. I’ll give you an example of a very large subscription company that we work with. That’s in the subscription space, they were having a hard time dealing with a lot of very small refunds for like $2 and $3 where people were doing a simple two week freetrial and putting a credit card in. It was more expensive for them to have a human being refund that $2 than it was for solving to be able to understand from the user of who they were look up a system, determine that the amount was less than X. Ask a couple of questions, make sure that that person has not repeatedly asked for a refund and actually say, great refund will be on your way by the end of the day.

Mahesh Ram (21:03):
And they never had to speak to a human being. So we’re doing that kind of stuff all the time and we’re, and it’s getting even more increasingly interesting because brands are actually exposing their data, their variables, their APIs to us. And why does this all possible? It’s possible because we built the best workflow, builder automation studio that in the market that allows any subject matter expert to do this since we’ve started solving and we have 600 and we have now about 700 million end users of the brands we serve. I think our top 10 customers are 4 trillion in market cap. We’ve not had a single customer, right. A line of code. So this is all subject matter experiences who understand the optimal customer experience. They’re building it in our workflow, automation studio and connecting to the data that matters for their brand Shopify for an e-commerce NetSuite, for shipping, you know, a segment which is a CDP for understanding your personall that’s done in the builder.

Mahesh Ram (21:57):
And so then once it’s exposed there, any, any expert in the company can build an experience and deploy it in a minute, two minutes and deploy it on every channel, by the way. So you build the business logic. Once you can have a Facebook messenger bot, a WhatsApp bot, soon, a voice bot web mobile mobile app, you name it. And if the business policy changes, you only have to go into solving and change it once. It’s wonderful. If you refund policy change from 45 days to 30 days, would you, you don’t wanna go through six places and change that. So that has really unleashed enormous power for our customer and better experience for the user. I mean, as a consumer, that’s what I want, know who I take, the actions that are most relevant for me.

Dan Turchin (22:40):
Now in this show, we focus mostly on AI and automated decisions. And I’m curious to know, you mentioned some of the work that your your co-founders have done on semantics similarity. Where does the data come from that you used to train these models that detect semantics similarity.

Mahesh Ram (22:55):
So there really are three, three sources of data and some of them commingle, but one is past resolution data, which we can get prior conversations, transcripts, you know, what questions have been asked in the past? How did agents answer them? That’s that’s available to us, right? The brands have to give us access, but they do so readily. Cause you know, that that’s how you train on one of the Mo the models. There’s also all the Corpus of knowledge that the brand has the FAQs, the KBS. And in fact, that has a lot of information semantics and understanding of how the brand talks about their products and services. So you can ingest that, right? And then the third piece is user interaction with, with our platform. And we can see if a user’s how they all the myriad of ways in which they’ve asked questions.

Mahesh Ram (23:40):
And in real time, we presented them with solutions or things, and they’ve interacted with that, right? There’s interaction data that you can collect at the moment of truth, that you can also then use to train and say, this answer was consistently shown to people asking this question, but it, but nobody ever was able to resolve the issue using it. Right. What does that tell us that tells us either the knowledge is wrong is possible. We got the, we got the intent wrong possible, or there’s, there’s a gap in the knowledge that could easily be filled because, and I’ll give you a real example. We had a double sided marketplace that started doing business in Asia and in that market, they didn’t support PayPal. And so a lot of users in that market were asking questions about PayPal. And so we were immediately able to say, there’s a gap in your knowledge base for PayPal.

Mahesh Ram (24:28):
And so then they were able to, to immediately write a piece of content in their knowledge base, again, single source of truth. This is a big differentiator for us versus everybody else is most chatbot companies are asking you to write scripted knowledge for the bot. We don’t we say, leave it in your source of truth, zenes knowledge base, or Salesforce knowledge or any webpage for that matter. So if you’re gonna change your policy, change it in a single source of truth, we’ll find it. And so this is an example of that exam example. I just gave with the PayPal. They just wrote a new article about PayPal that had five or six paragraphs, but we’re extracting the specific paragraph on that answer. So if they say I’m in Korea and I wanna use PayPal, that could be paragraph five, we’ll show you that answer. So hopefully that answer your question, but that’s how, that’s how we do it.

Dan Turchin (25:16):
So the problem with using any training data in any field is that there’s often bias inherently built into the data and in your business, I’m just gonna spitball, but I can envision a time when solves recommending a retail location or a part to purchase or a plan to use. And based on the data that it was trained on, there may be certain retail stores or certain plans or certain parts that are over or underrepresented, and Sovi may unintentionally be proposing solutions that could potentially not send customers to a store owned by an underrepresented minority or doing things that introduce unintended bias, even though the solving algorithms have no bias, the underlying data might, what kind of, first of all, do you think of that as a problem we’re solving and to the extent you do, how do you think about mitigating the impact of inherent bias in the data?

Mahesh Ram (26:23):
Yeah. and, you know, I would just say customers care about that. So by extension, we have to so, but we would care about it anyway, cuz it’s an important point. I think what you need to do with any AI based system is you need to be able to have these subject matter experts provide some overlay of, of additional data or, you know, votes if you will, or some, some mechanism to overcome that. Right? But the first thing you need to do is identify that that’s happening. So one of the things we do really well is provide a lot of analytic data about what those trend lines are. So you can start to see in any journey, let’s just, let’s take a journey, right? Subscription management, you know, all of us as consumers wanna sometimes cancel a subscription. Sometimes we wanna pause it cuz we’re gonna be on a vacation.

Mahesh Ram (27:14):
Sometimes we wanna skip it because we’re tired of, you know, we have too much food at home and we don’t want, we don’t want that meal. All of that falls under a broad description of subscription management right in 10 category. But then what we may wanna do is to, is to the brand, may wanna encourage people to do the pause before they more so than they wanna do the skip as an example. So what they can do is to start to understand the trend lines with how users go through those journeys and immediately then be able to optimize for particular outcomes and provide enough data or change the journey just slightly to be able to take users, but it comes from data, right? And so they’re able to go at at a node level, analytic level and see what’s going on. Like where are the branches?

Mahesh Ram (27:57):
You know, what are people, what are people doing in those, in those decisions, if you will. and then they can do that. They can also do that for the train, for the knowledge content. They can see that particular things are being under, under, you know, underutilized content. And they can go in and explore. Why is that? Oftentimes though companies and knowledge administrators think that all content is created equal. And in reality, there’s a little bit of an 80 20 rule. And so they wonder why nobody’s using this answer, that there’s brilliant, you know, KB answer that I wrote. And in reality people aren’t asking that question. And so the data will tell you that. And so, but if, but if in fact they are asking the question, if not showing up, you can, you can actually provide some overlay and make sure that it’s surfacing more. You can do that. We, we tend to not encourage people to do too much of that because then they’re overfitting and it, and it will break it. So, you know, we don’t want that, but sometimes it’s appropriate.

Dan Turchin (28:53):
Sue and I are both pretty excited about the future of virtual agents. And to your point earlier, the potential of a virtual agent to yield higher CSAT, better customer satisfaction than having to call a live agent and wait on hold. But just a thought experiment here let’s fast forward, 10 years or 10 years. Plus do you envision a time when a virtual agent will be capable of fully replacing the need for that fallback to a live agent?

Mahesh Ram (29:22):
My, my view on this is a little bit more nuanced and replace or simply augment. I think what you’re going to get into is the role of the agent is going to change into more of an advisor then, and it’s gonna free them up and liberate them to be more of an advisor. How do you get the most out of the product or service you just purchased versus telling you how to troubleshoot the I, the iOS version of your app. So if you think about that as the ultimate goal, what you really wanna do is to attack the things that are repetitive, but, but complex, you know, repetitive doesn’t, you know, people tend to think of repetitive as being simple, repetitive is, can be extremely complex. There might be 65 use cases within that cancellation subscription that all depend on the data that about Dan and what Dan needs and what Dan wants.

Mahesh Ram (30:17):
And Dan’s history. The reality is training a thousand agents in a call center to be good at that is, is not, is not easy, especially with the area of the era of the great resignation. And there’s enormous turnover of the likelihood. You’re gonna get somebody on day one to be able to do that as well as agent who’s been there a thousand days is very low, but if we can learn from a lot of that journey data, what questions do they ask? What data do we have to look up and what decisions do they take? We can in fact predict for any new issue that comes in, what questions I should ask, what data I should collect. And based on that data, what steps should I take? There’s actually nothing stopping us from doing that. You might still have a human overlay to supervise that, nothing wrong with that.

Mahesh Ram (31:08):
And, but you may also wanna think about when do we wanna introduce humans right into this experience. If someone is got a, I don’t know, $2,000 in their shopping card, and they’re asking a question about the finance financing, you might wanna get them to a human being there and not have them go through the agent. The, but the system should learn that because what it can say is people with $2,000 in their card are, have a propensity to buy that’s X, so let’s get them out of it. Right? So when you start thinking about it in the advanced level, I don’t think we’re replacing agents. I think what we’re doing is, is just allowing us to use humans at they’re most optimal, which is exercising judgment, giving advice, giving guidance, giving counsel, and freeing them from road tasks. I mean, we hear all the time from our customers that the agent attrition has gone down after implementing solving, cause agents are more empowered they’re they just feel better. They’re not doing the road stuff. That’s a, that’s a great thing to hear. And I think that’s the future

Dan Turchin (32:01):
And mesh, I gotta get you off the hot seat, but not without as answering one last question for me. So I mentioned at top of the show, so you’re going on almost seven years. It’s obvious since 2015, when you founded the company. And before that you had along that distinguished career as an entrepreneur as, and as a tech exec, what does surprise you most about your entrepreneurial journey and the process of founding a successful high tech venture.

Mahesh Ram (32:29):
So I think that one of the surprises is that each time I’ve done it is the third time now has been fundamentally diff different. And I think part of that is the eras have changed and the way employees expect to be treated and, you know, has changed. There’s a, the era today is, you know, and it was a change for me is after we started solving, we committed to radical transparency with our employees. And that was not easy. You know, in the prior era, you’d say, well, you know, need to know basis, tell people what you need to know. Most of the people in our company know just about everything, about the company, good, bad, and different. They might not know somebody else’s salary. They might not know, you know, compensation or, or those kind of, they certainly don’t know personal details about their lives, but we try to be as open and honest with people as possible.

Mahesh Ram (33:18):
But that took honestly took a little bit of migration for me mentally, because it’s just, you’re sharing information trust with enormous amount of trust without any legal boundaries. You know, it’s not like somebody could not go out and tell somebody what we just said in an all hands meeting. So I think that’s number one. Number two is I think the available technology, I think the power of what we have now with the technology, the AI and the understanding blows me away. And so there’s a lot more applications for this there ever were. And that’s always exciting cuz that reinvigorates you, that energizes final thing I’d say is just, you know, as you get a little older and you’ve been doing this a while, I think one thing you learn is my ex-chairman of the board Reese. Stka, who’s an amazing man once said, remember one thing, sadly, but truly many of you are gonna spend as much time with the people you work with as your family members.

Mahesh Ram (34:14):
It’s an unfortunate fact of society. You’re gonna spend eight hours, nine hours in some cases. And you might not even spend that with your family members. Remember that the people you, you work with have to be the kind of people you’d want. They don’t have to be family. I think that’s a false, that’s a false paradox, but it’s a false statement, but they have to be people you really feel make you smarter, better collaborate with you. And I think we’ve really applied that stringently and I’ll leave, I’ll close this by saying one thing, Dan, the key moment in an entrepreneur with any company is when they know that a new hire could be made by any five people in the company, without the entrepreneur having to be involved in the process or blessing the, the higher, their judgment’s better than mine. And even if I said no, and those five people said, yes, I would hire them because I know that they, the culture will take care of this. And I think that to me is like, you don’t learn that lesson until you’ve been doing this a while. You know, you, you have to give up, you giving up control, giving people autonomy and giving up, you know, giving up your own control is a, is, is actually a makes you better.

Dan Turchin (35:19):
It’s so intense. And you learn so much about yourself and you have so many opportunities to change lives in the process and hopefully leave the world in a little better way than it was when you when you started out. And I think that’s why we do it even though it’s oftentimes not a rational decision.

Mahesh Ram (35:37):
Exactly. I think you, you know it well, and I think you said it well,

Dan Turchin (35:42):
Well, Mahesh before I let you go, where can the audience go to learn more about yourself and some of what you discussed about solving?

Mahesh Ram (35:49):
So pretty straightforward, www.solve.com. That’s S O L VV y.com solve.com. That’s probably the best place to go to learn what we do and how we do it. And you know, we’d be happy to, anybody’s interested in getting a demo of the product and your, you know, your company or your needs, needs, support automation, and wants to delight its customers and end users. It’ll also saving a lot of time and money for your business. Please do visit us. Let us show you. We’re we’re very much a show, not tell company. So we’d actually prefer to show you. So if you’re interested, feel free to, to check in with us. And of course, social media channel as well. Sol V Y you can look us on all the channels.

Dan Turchin (36:34):
My thanks for hanging out. This has been a lot of fun. Great work.

Mahesh Ram (36:37):
Yeah. Dan, my pleasure. Thank you. Thanks for your audience as well.

Dan Turchin (36:40):
Good stuff. Best of luck to you. And gosh, that’s a wrap for this week. He’s your host of AI in the future of work, Dan Turin signing off, but we’re back next week with another fascinating guest.