Episode 6: AI Powered CRM 

In this episode of AI at Work, host Rob May has a conversation with Justin Kao, Co-founder and VP of Growth at Spiro.ai. Tune in to learn more about Spiro.ai and their AI powered CRM. 


Subscribe to AI at Work on iTunes or Google Play

Rob Circle Headshot

Rob May, CEO, Talla 

Justin Kao headshot circle

Justin Kao, Co-founder and VP of Growth, Spiro.ai



Episode Transcription 

Rob May: Welcome to the latest edition of the "AI at Work" podcast. I'm Rob May, co-founder and CEO of Talla. We are an intelligent, knowledge-base / chatbot / machine learning, everything mixed together. If you think about information and automation coming together in a really interesting tool, that's what we do.

This show is about AI people and the stuff that they're working on, and if you are trying to deploy an AI project somewhere or you're thinking about having AI at work in some way, shape, or form, we're here to help you learn and make better decisions. Today our guest is Justin Kao, the co-founder and VP of growth at Spiro.ai. Welcome.

Justin Kao: Thanks for having me.

RM: To start, tell us a little bit about your background first, and then what Spiro does.

JK: At Spiro right now I manage the growth. That oversees all sales and implementation. Before that I was in CRM consulting. I would actually go to a lot of enterprise customers and I would help build their CRM for them. I would listen in to the requirements, talk to managers, talk to the sales people, and I would, whether with Siebel, Salesforce, Microsoft Dynamics, just go in there and build enterprise CRM systems for these folks.

As far as what Spiro is up to right now, we're building an AI-powered CRM. What that means is that we're baking AI into the core functions of CRM. The idea is that Spiro can read all your emails and automatically log them to the right spot. It can automatically know who you're talking to and create deals for you and give you proactive recommendations to make sure that you're not dropping anything. It’s this concept of a sales assistant that's actually just doing stuff for you in the background.

RM: It's interesting, there's a lot of parallels to what we do at Talla. Much of the work is sort of maintaining a knowledge base and making sure that your support queries and your sales queries are answered, and organizing that automatically, and doing a bunch of cool stuff with it.

JK: Yeah.

RM: You did this CRM implementation in the pre-AI world, and now you're doing it in the post-AI world. What have you seen as the differences there, and what have you learned? Where do you think this is going?

JK: I have a very pointed view of how CRM adoption is going. I've been in CRM pretty much my whole career, and I don't want to throw any past clients under the bus, but we had a past client who spent millions of dollars on the CRM implementation and, at the end of a one-year implementation they had 10% usage. So 10% of their users were logging in once a month.

RM: Wow.

JK: They came back to us and they were like, “Hey, this is terrible. Let's order some change management consulting”, and we got the number up to 20% after another half a million dollars of consulting for them. They were like, “Yeah, that's a huge win.” For us that was like, you're up to 20% after you're two and a half million dollars down the hole and 20% of your team is using it.

RM: How big is this team?

JK: Hundreds.

RM: What did they use before? Did they have a different CRM that they stuck with?

JK: They had a legacy ERP kind of thing that they had a CRM add-on. They were going to Salesforce. The idea here is that, as much as we try to automate workflows for these folks and really get them going on CRM, we try to do the best we can with what we think the data they need to capture and all that kind of thing. At the end of the day they still didn't care. They still didn't really want to use it.

The big difference for us is that we want to build a system that's in the background, not in the way, collecting the data that they need to collect without really needing them to even log in in the first place. Seeing the difference between pre-AI and post-AI, it's really: how do we take the information that you're actually using-- your email and your calendar, your phone-- and give that to you, and present it in a useful way where it can really give you good recommendations and make sure that we're collecting the right data for you?

RM: In some sense you're hoping you can actually make more of the CRM vanish.

JK: Exactly. Our dream is like, if you've seen the movie Her, Scarlett Johansson plays a computer, an operating system in that movie and Joaquin Phoenix was just talking to her. Right? And he's just asking, “hey, what's on my calendar today?” And she responds, “Oh, you've got this thing, this thing, this thing, and also tomorrow you have to call your mom.” That's our dream. It's just something that you can have a conversation with. In the future it's something you're interacting with on a really personal level.

RM: We just had a Talla board meeting, actually, and that concept came up of, “how do you market and present the power of these AI tools when you message?” They're very different than the SaaS tools that came before us in terms of the benefits that they provide and everything else.

JK: Right.

RM: One of our investors was talking about, “Think of it as if you have the entire power of the organization, everything that everybody knows, at your fingertips behind this basic conversational interface that you can interact with at any given time, 24/7, on demand.” It sounds very similar to the stuff that you guys do. How do you think about product management in an AI world? CRMs do lots of things, how do you decide whether you're going to do something with voice next or email next? Do you have a consistent framework where you say, “Here's the next place that we're going to apply AI and here's why”?

JK: We've kind of followed our users in that sense. Originally we had an app. We still have an app, obviously, and it's reading your data. You go to the app for the recommendations, but adoption wasn't good enough for us. We wanted something really sticky. Our user base tends to be field-based sales folks that are on the road, and they're addicted to their phones, they're living in their phones all the time. We're like, well, we've got to give it to them. So we're pushing things towards email. We still know that no matter what they're going to be living out of their Outlook apps or their Gmail apps.

That's where they're living, let's just be there. That really pushed us to have a whole component that's an email assistant that's just conversational where you can actually do pretty much anything you need to in a CRM via the email assistant.

RM: Has there been anything that you've tried to pull off-- well, let me back-up. I ask this question because one of the challenges of data science in product versus traditional sort of engineering in product management is that with engineering you kind of know what you're getting most of the time. You can make rough estimates on stuff. With data science it's very different because you may say, "Oh, we should build a model to do this thing, automate this process, predict this thing". In order to be good enough for our users it needs to be 96% accurate. You have no idea if you can achieve 96% accuracy. So sometimes what happens in data science is you're at your 87% accuracy and you go to your data science team and say, "This needs to go to 96%". And they go, "Great. How long do you want us to work on this?" There aren't a lot of best practices. It's not as well understood as engineering yet.

Have you run into anything that you tried to automate or predict or add some AI to that you felt like you had to back off of for a while because you didn't have enough data and you had to try again later, or the models really didn't work or it turned out that it just didn't work like you thought? Do you have any good stories like that?

JK: A hundred percent. When we started we wanted a feedback loop to understand how good our recommendations were. At its early beginnings all Spiro really was, was a recommendation engine. It looked at your opportunities, looked at your activities, looked at the things you were doing. We had a model to predict when you should be reaching out next based on your sales process, your average opportunity length, your sales cycle, your past performance. So we took all that stuff into account and we'd give you kind of a list of recommendations. And our first product was just, “Hey, was this a good recommendation?” You couldn't even do anything to it. It was terrible. We had to beg people to use it.

RM: That's normal for early stages.

JK: We had to be like, “If you guys recommend 10 things today I'll give you an Amazon gift card.” It was shameless. But essentially what we did was we built up a user base of 14,000 free trial users, essentially, before we went to public with a product. And we just got a bunch of them to eventually increase functionality. It would give them a little bit more out of it. They could make phone calls, etc., but the idea here is that they gave us a one through five star rating on recommendations. That's how we got to a point, once we had an average of about 4.8 stars, that we took away the stars. We felt like our model was pretty good at that point to go on its own.

After that another thing we did was, we have a feature in our email system that it's digging through emails, it's using natural language processing, it's looking for keywords. It'll say things like, it'll make a recommendation to you and it'll say, I think you're selling to this guy and I notice that he's not in your pipeline. Should I just add him? And so we actually will look at it and we'll ask the user, yes or no.

Right now I would say that's at about 75% accuracy rate. Once we get that to a 95% accuracy rate, we're not even going to ask you. We're just going to do it.

RM: A lot of what you're doing is using human feedback loops to control the lower accuracy stuff. Do your prospects or your customers have a problem with that? Do they understand that initially, that this thing is going to take some training?

JK: At the beginning when we were more up front about the feedback loop, it took a lot of coaching. People were like, this is really annoying. We just beared with it. Right now what we do is we encourage usage patterns that give us the good feedback loops. When we on-board folks we actually walk them through the things that make us different versus a traditional CRM.

People choose Spiro because they didn't have success with traditional CRM. They've tried it before. They've tried the Salesforces, the Zohos, the Dynamics, and their team needed something different. We really harp on those differences. Those differences are the recommendation engine. Then email assistant. When we get a client on-boarded we are very upfront about the things that you should be doing. As you interact with it, Spiro will get better.

RM: Is there a certain type of person that is more successful with Spiro than others that's more willing to do the training? Do you ever frame it even against, “Well look, it's going to be faster than training a new sales rep”?

JK: For sure, and you'd be surprised. Our best customers are not in technology, they're not in SaaS. They're more traditional, a little bit older user base. Manufacturing, construction, logistics, transportation, that kind of stuff. The reason they really embrace it is because we train really simply on these things that they know how to do. Where they have met failure before in the past where they'd have to go to Salesforce, they'd have to find a contact, find an account, find an opportunity and log the data manually or figure that out. That is crazy for them. Or, they could tell our email assistant to do something for them.

We've kind of flipped it and we're harping on these kind of features that are really different. By kind of them embracing it, and because they never got to embrace the past where it was go manually log your data, they've really taken on to that.

RM: Before you mentioned natural language processing. It's something we do a lot of here at Talla and it's something that I think is very challenging for the public to understand. I've written a little bit about where I think this is going, because it's a rapidly-evolving technology. It's making a lot of strides, but it's still very problematic in a lot of ways. Can you talk a little bit about some of the use cases, how you guys have approached NLP? Are you using mostly sort of open-source tools? Do you have NLP people on staff?

JK: We use a lot of open-source tools. We take the intent from it and we do our own stuff on top of that. The idea for us, well our dream, is obviously this conversational software. Right? So it's a big focus for us to really make sure NLP's working.

We've really simplified it, I think, for our use cases. We made it really direct and made it so that these are the things that you can do, and until our user base is ready to be a little bit more open with it, I think what we will do is our approach is really targeted, and make sure that it's really direct and super obvious what you can ask it, what you can have it do, what questions you can ask. That's allowed us to train easily on it and really get them to understand how it works.

RM: Tell me a little bit about when you go out to sell this product and you're trying to talk to people. They're high in the marketing funnel. What are the things that they don't understand about an AI product? Where do you get hung up? What are the areas that you're thinking the market is going to grow and mature over the next couple of years, and this is going to make more sense, and it's going to be a lot better for all of us who are selling AI?

JK: It's kind of like how everyone approached the word "automation" what, 5,10 years ago? Oftentimes our customers have no idea what AI is. They get some mandate that, "Oh, AI. That's the thing. We should do that." There's no real concept of what that means.

What we do, actually, throughout when we're leading a potential prospect through what we do and how we help, we take away this idea of AI. It's really, how are the things that we're doing giving you value? How do we make sure you're getting value out of us?

Our sales process is really all about educating how Spiro can take what you're doing already and really give you more out of that. Along the way we'll explain a few of the things. We'll talk about the recommendation engine. We're not telling them about the machine learning, unless they really ask. For the average consumer, they don't really care. They just want to see, I can get these recommendations. That's cool. It's really just about making sure they see the value and the benefit and it's really clear to them.

RM: What have you learned at Spiro, and also just looking at AI more broadly? When you think about your advice to a VP of, fill-in-the-blank at a company who's like, “I'm implementing my first AI tool”, or “I'm figuring out where in my business I should implement AI”. What are the tips that you have for them to help make that implementation successful? I'm asking here a little more just like the more generic tips that would tend to apply to any kind of AI product, not necessarily CRM.

JK: I would say, just understanding what the goal is at the end of the day. Do you want to increase efficiency for your manufacturing company? Or, do you want to increase lead conversion for your marketing? Keep that goal in mind and use AI as a means to an end to really get there.

Stressing AI for the sake of AI, I think, is silly. It's really about making sure that you understand the goals. How we've done it at Spiro is that we talk about the goals and we're taking your use cases and taking your goals and making sure that you can understand how we utilize our part the best way you can to meet those goals. It's really just about, at the end of the day, "what do you need to accomplish?". 

RM: Your leads in your funnel, are they pretty high-qualified, that they're looking for a CRM and all that kind of stuff? Or do you get a lot of AI tire-kickers, who are just trying to figure out what is this and what does it do and they're not really serious? Do you spend a lot of time with them?

JK: We try to filter them out as best we can, but people are getting good at filling our forms and confusing us. I'm sure you guys run into that all the time where you just get the folks who are like, “Oh, I'm an AI consultant. I just want to see what you guys are doing”.

RM: You get a lot of department VPs at big companies who come in and it's like, “What are you looking for?” and they’re like, “Well, we're looking for AI. So the boss has said we need to apply more AI for productivity”. It's like, back to the goals thing. “Well, how do you want to apply AI? What are you trying to do?” Then they’re like, “Well, we don't know. Tell us what your thing does and we'll see if that's an area we can apply it.” It's very different than the other companies that I've been involved with in the past.

JK: It's such a buzzword right now. Obviously people are curious about it, and it's great because we get to educate a lot of folks. But at the same time it's wasting valuable sales time, unfortunately. For us it's a lot of consulting. It's a lot of consultants who want to learn a lot more about it, like “I should add this to my bag of tricks”.

RM: I mean, you don't necessarily want to spend all your time just educating them for their business. Being in a new, emerging space that's very buzzwordy is a double-edged sword. On one hand it's easier to get leads than if you were going into a more stable market where there were lots of established players and it was less sexy and interesting. But on the flip side, many of those leads that you get in are just tire-kickers and not serious. Then you have to deal with the fact that there's a whole bunch of people in our space that are promoting themselves as AI. Like, they're really pretty dumb and stupid and not good products, and it reflects negatively on the ones that are doing interesting stuff.

JK: It does. I know. It's kind of like, anyone who has a basic workflow, they call themselves AI. Right? And it's tough, people have these expectations. Right? We were just talking about Her before, and if you're not educated enough in AI but you were like, “Oh, well, they talk about it in Her, it should do this” and all these preconceived notions of how it should work in the enterprise. The AI technology's nowhere near where that level is. Then people are like, “What are you doing?” I'm like, “Well, we're getting there. It's step-by-step. But we're doing what we can”.

RM: Do you have a really good success story that you can publicly share or any case studies of a use case where you're like, “Hey, we have this Spiro customer and they saw this level of improvement, or something good that you can talk about?”

JK: You know, I don't have any numbers off the top of my head. I mentioned before, our biggest customers are in manufacturing or construction. One of our bigger customers, they're in B2B finance, and what they do is they actually sell loans to furniture stores. If you go to Jordan's Furniture and you need to buy a really expensive couch, you can get financing for that. That's their back-end product.

These guys are all on the road, they're road warriors. They are hitting the ground running. They've seen efficiencies crazy improve. We're collecting data that they would have never gotten before. I believe they were using an older CRM before us and their sales team basically wasn't using it. What we're doing is we're collecting all this stuff for them, these insights, collecting emails, phone call conversations that they just never had before.

Their managers can look at that and we have a back-end tool that looks at risk assessment. Right? They're able to look at that and they can look at the deals they're working on and they can really make educated decisions or guesses about where the forecast should be based on risk assessment. That's huge for them and making sure that they get an accurate pipeline.

RM: You mentioned in there that you were capturing a lot more data that wasn't being captured before. Do you see one of the advantages of starting and building an AI company is that you can go out, and when you're competing against the other CRM vendors, you can do some AI things par for the course; but, then you can build into your workflows ways to collect new data sets that the old guard is not collecting that's going to help you come up with new things you can AI-ify about your product. Right?

JK: Exactly. We've seen a lot of the other, older CRMs release AI add-ons. I mean, Salesforce's Einstein. That's kind of the most pressing out there. A lot of the bigger players have released AI add-ons. One of the things that really doesn't work in their advantage is they have established architectures. These guys have been using their platforms for forever, and they were pushing cloud and customization and ease of use.

The thing is that they've kind of built up all these really highly-customized customers, and there's no way that they could leverage all that data in a meaningful way, where we were able to build it from the ground up. That's been really huge for us because we've been able to think of this from the very start.

RM: I think I've said it on this podcast before, but I think the last sort of decade of enterprise software design was all about making things run in the browser, making them easy to use, make them mobile-friendly, make them a little bit like your social apps. You can upload avatars and you can tag things. I think the next wave, the skill sets that are really going to make enterprise software companies successful, is the ability to build interfaces where you collect this other data. You do it in ways that it doesn't feel like extra work that people are having to do, but you build it into the workflows that you're collecting data.

When humans use software there's so many things we're thinking about that we don't capture as part of the software because there wasn't a reason to before. Now you can capture that data and you can build models with it and you can automate more and more of the work stack.

JK: A hundred percent, I agree. The old way where the cool thing you could do was build a workflow, right? If this bit of data is changed then update something else, that's magic for most people. But, it should just know what you're about to update, do that, and then do the workflow. All this stuff should just be automated.

RM:  I definitely agree. Do you think the same thing holds true for companies and their business processes in general? Let's say you are a VP at a big company and you don't have an AI initiative yet. Is it best to just wait and see where the market goes? Or, do you feel like these models and concepts build on themselves. You've got to step in and put your toe in the water and embrace AI early because it's going to have a flywheel and  you're never going to catch up to the leaders if you're too far behind.

JK: Well, I'm obviously a big fan of early AI adoption. It benefits both of us, right? I'm going to say that for sure it makes sense to get going faster. You're collecting the right bits of data. You're learning more stuff. You can improve upon your models. I one hundred percent think that the earlier you get into it the better it is. You know, with a lot of our customers they are learning things about themselves they just had no insight into at all. It's really just a matter of getting the footings early, and you can iterate on that.

RM: Last question, because you work in this industry, if you have an opinion on this. There's this ongoing debate. Mark Zuckerberg versus Elon Musk. "AI is out to kill us all. We need to start protecting against it. It's the biggest threat to humanity." That's the Elon Musk version. The Mark Zuckerberg version is like, "Nah, you're crazy. We have no reason to think it would be deadly. Even if it might be, that's a long way off. We shouldn't worry about it today." Do you have an opinion on that debate? Or where do you fall on that spectrum?

JK: Definitely on the Zuckerberg side. I feel like Elon Musk is a bit of a sensationalist. You know, he says things that generate headlines. I think he called one of the cave divers in Thailand a pedophile for no reason because he said his submarine was dumb. I mean, so I'm obviously a big proponent of artificial intelligence. I am a big proponent of using it the way you can as long as you understand your use cases and you're using it in a way that's helpful, not just for the sake of using it. I think it's going to be great for everybody.

RM: Great. And if people are interested in Spiro where's the best place to find information about you guys?

JK: You can go to our website, Spiro.ai and poke around. If you're interested you can sign up for a demonstration right on the website. Or reach out to me directly. My email is Justin@Spiro.ai.

RM: Awesome. Justin Kao, co-founder and VP of growth at Spiro. Thanks for joining us today.

JK: Thank you.

Subscribe to AI at Work on iTunes or Google Play and share with your network! If you have feedback or questions, we'd love to hear from you at podcast@talla.com or tweet at us @talllainc.