Helping Sales Professionals Trust AI
AI technology is here to help us. It can make increase research, decision-making, and efficiency in your business. AI is not here to get your job, it's here to simplify it. Join your host, Chad Burmeister and his guest Joseph Miller to talk about the usefulness of AI in the PreSales space. Joseph is the co-founder and Chief Data Scientist of Vivun. Vivun is an enterprise software company that is focused solely on presales. Learn what the job is of the AI in Vivun and it enhances the job of a sales engineer.
Listen to the podcast here:
Helping Sales Professionals Trust AI
I've got a cool guy. A lot of times, we talked to founders, CEOs that are somewhat involved in data science and AI. It's a rare treat that we get to talk to someone who's deep into artificial intelligence. Joe Miller is the Chief Data Officer of Vivun. He's been there for a couple of years through seed round and the rays. Before that, he spent some time in the Ray Dalio hedge fund, Bridgewater. He's worked with IBM Watson. He's got a lot of experience. We're going to dig in and get to know Joe first, and then we're going to dive into some of the AI conversations. Joe, welcome to the show. Thanks for being here.
Thanks for having me.
Before we jump into the day and age of what's going on, I'd like our readers to understand your background. Tell us a little bit about Vivun. What your company is doing now and then we'll go back to how it all started.
It's an enterprise software company that is focused on this niche called presales. Matt and John are two Cofounders. They have decades between them as Sales Engineers at various companies, Pandora, Zuora, and other places. Their insight was to see that this space had unique data to the sales pipeline. It connected sales, product, field and all of these things. It was being neglected. They started out trying to build out a place to live and to prove that the data that was being collected there wasn't being kept with all its value.
I came in about a year, if I'm not mistaken, maybe a little more than that, after they started building this enterprise software out to come in and say, "We have all of this unique data, what can we infer from it? What insights can we drive? How do we synthesize all of this data? How do we build an AI platform on top of it to make recommendations and capture all the value that was in that unique set?" I've been building there for a couple of years. The types of things that we're able to do are not only exciting for Vivun and the field that we're in but in technology in general. The type of recommendations we can make and the type of data we gather is having impacts on how companies evolve their technology. When you start to think about that horizontally and broadly, it becomes an enormously impactful place to work.
I think of places like WebEx where I worked and RingCentral, the seller is the presales person, the delivery person and the collections person. I'm being facetious a little bit. Before that, in enterprise software, that role is obviously bifurcated. You've got an inside rep that supports the rep and SC. A lot of times, there's a disconnect between what the SC knows that person is either on Zoom or on-site and discovery. I have to imagine that turning on the lights in that dark black hole and serving up some of the information keeps the salesperson more in the loop. Also, it helps with forecasting and understanding. There's so much data that's hidden there.
With AI, you can get more with less people.
In general, we always say that every company now will be a technology company so as the narrative that software is eating the world. What's interesting is, as technology companies, specifically as they scale, a lot of the divisions end up being siloed. Everybody becomes more specialized. That's the beauty and fun part of working at a startup. A lot of people wear a lot of hats all at the same time. It's exciting because everybody is also sharing the same goals and missions when you're that young organism.
As you scale, people started to specialize. They have their own OKRs, their own metrics that they start to follow. They have their own goals that they're out to achieve. The field ends up trying to do its own goals and incentive structures. The thing that is weird is that somewhere along the line, these silos start losing the overarching goal of the mission statement of the company. The presale space is about connecting those things back to that original goal and bringing everybody into a common view of how the world is, what's happening in the field and how we should evolve our technology to meet our customer needs, not just now but in the future. Getting everybody back on the same page from a data-driven perspective is enormously profitable. You get to capture all that value that's being lost from the lack of synergy between the product and the field.
Let's go back a little. I like to ask the question. Some of your first memories when you're younger. What was your passion? I like to connect the dots between what you're doing now and what you like then.
I've always wanted to be a scientist. That was first and foremost. When I was growing up, the scientist I wanted to be was the Indiana Jones archaeologist. A lot of people my age, that was the thing and then like Doc Watson, Back to the Future. That was where my head was always when I was a kid, science, archeology, discovery and adventure. I also liked sports and whatnot. As I grew up, I realized more that science was always going to be that thing that made me want to get up in the morning. They got more focused. I realized I didn't want to spend my life digging up dinosaur bones as much as I thought I did when I was little. Instead, I'd like to be somewhere where there's climate control.
My son, he's going into Electrical Engineering at a school in Colorado. He's now thinking of Computer Engineering. I'm so proud of him. It's a field that's here to stay. He's perfect for it. Tell me about where AI is used at your company. We've talked about the need for it. Can you share how specifically AI gets used at Vivun and with your customers?
We have a couple of pillars in the AI unit. One of them is how do you make recommendations about what to do next. What's the best value for the level of effort? What's interesting about the presale space? How we've organized that world, how we represent what's called knowledge representation and how do we represent that knowledge so that we can make inferences about what to do next. We start to make recommendations around deals in the past that have had this concern. Maybe it's price, timing, political alignment or you don't have a good champion. These are things that are natural into the process of the deal that the sales engineers uniquely have that insight into.
We don't have somebody that is a champion on our side here that is touting the technology or the advantages that we can bring from our product. What we're able to do with some of our AI is going through the comments, a lot of the natural language that's left in notes or has looked at in terms of the model that we use to predict how well that deal’s going. We start to look back and say, "Deals that looked like this deal that you're currently in did better if you did XYZ versus if you did ABC." Sometimes I like to characterize as the same Vivun is the best sales engineer in your company. It has the advantage of not walking out the door for a better gig. That's what it is. It's how you capture that historical knowledge that your best sales engineer does have. That person has been through every deal. They understand, "When we get pinched on price, this is how we navigate this. If we don't have a great champion, these are the angles that we used to get one. If the timing’s an issue, these are the levers that we can pull to make the deal survive that long to make it work."
All of that knowledge is in the data. You can make inferences about how it's going, all of this stuff and make these recommendations. One pillar is the recommendation of decision making, managerial decision making, deal decision making, sales engineering decision making. That's one side of it. The other side is a pure efficiency play. Everybody especially in the markets there is now, is trying to figure out how do you get more with fewer people. There's a lot of information out there. There is an enormous amount of information, more information that humans can consume on a daily basis. How do you synthesize all of that information so that you can give the pieces to the decision-makers and they can make quicker decisions? We have this thing called deal gaps. It's a difference between what the customer wants and what your current product offer. It's effectively what it is but there's a lot of nuances in there.
You can imagine that if you have customers A, B and C. Customer A wants thing one as their top priority. Customer B wants thing two, but everybody wants thing four as their second priority. Synthesizing that across would take manually an enormous amount of time. It's a very inefficient process to have to get your hands around all of that. That's another space where if you can parse the natural language and build a model that can understand that this is the thing that pops to the top of like, "Everybody wants this thing as the second thing. That should be your first priority." That's a big efficiency game.
Where are you capturing the raw data? I think of an SC, they're either on the phone majority of the time these days. It used to be in person. How do you take what's in that meeting for 60 minutes and put it into the model? Does it come from different places than that?
It comes from different places. We can get it from our connections with Salesforce or through Jira or other apps. We have a lot of interfaces. We've spent a lot of time making sure that our product plays well with everybody because we are trying to connect all of these silos. We want everybody to be able to talk into the central brain. It comes from a lot of places. You're looking at things like notes and people filling out data in the Vivun platform. We make a lot of inferential data. One of the big aspects of principles that we have on the data science team is that we don't want to burden the user with a bunch of data input. That's one. It's buggy. That's not a great way of building systems because people get fatigued on that. They get lazy. It's a pain. Nobody wants to do that. I don't want the solution to be worse than the problem.
We spent a lot of time representing the system in the way we think a deal can be constructed. That goes back to that knowledge representation, which allows us to make inferences about pieces of data. Instead of having to ask you, I can infer it from other behavior you have. For example, we track things like momentum. Is this deal moving along? We can look at that based on how active are the deals. It could be calendar events or phone calls that you've made or emails that were exchanged. I can look at that and say like, "Are people working on this thing or not?" Use that. We apply some of our logic and inference about that. We use that to go into our models of, "Is this deal healthy? Is there something more you could do to make it better? Should you not be working on this deal?"
If you demand too much of the user, you're probably not doing the job correctly.
I met with the CEO of the company on this show called Traq365. They're listening to the phone conversation. It's not like a chorus or a gong that converts it into a note but put some intelligence around that conversation. That could be one of the feeds that enter your discovery technology.
We can do things like sentiment analysis, entity extraction and pairing. One of our core premises because we are having to work with a lot of limited data, the amount of space that a deal could exist is enormous. You have this issue of saying like, "This deal doesn't look like every deal that you've ever had but it looks like some deals." How do you figure out which deals it looked like so you can say, "If there were lessons to be learned from those previous deals, we can apply them to this current deal?" Things like sentiment are some features that we use. Entity extraction that we can then use to say, "This thing looks like this other deal.” We can put a link between the two and share knowledge across time. That kind of stuff is the type of thing we do. We do a lot of natural language processing, which is the technical term for it.
I've gone to a few of the AI, in general, shows over the last several years. One in New Jersey. The Amazon VOICE Conference was interesting. In San Francisco and New York, there was the AI Conference. This one speaker, I'm going to blank on who he is, he's got the data that shows the revenue per head in the world and certainly in the country, is going up. The headcount is not going up at the same pace. From a fundamental ability for an individual salesperson to drive more revenue, have you seen any trend lines where you say, "That rep who used to do 500,000 can now do 750,000?" When you start to put on the Iron Man suit, what does that enable you to do? What should reps be thinking about as salespeople to be ready for the Iron Man suit? What skills do they need to have that they may not have now?
The first is we actively try to measure these types of things. One, it's good for our own business to know like, "Is our product delivering the value we expect it to?" That goes back to that pillar of efficiency AI that we do, a lot of clustering things that would take people weeks at a time. We have clients that three men will be spending a week a month characterizing all of this stuff. We cut that down to a day. It's a dramatic change, being able to use clustering analysis and things like this to do that operation for them. That makes everybody a lot more productive because I gave you back three man-weeks. That kind of thing is valuable.
The second question you asked is, what should people do to prepare to use those tools to their optimum? I have a little bit of a different opinion about this. I don't think they should have to do anything. The value of AI is not that everybody has to become a computer scientist. I don't think everybody should have to be an engineer. There's a lot of talents and skills in the sales engineering space, account management and product management. Not everybody should have to understand how a combustion engine works to drive a car. In some sense, if you demand too much of the user, you're probably not doing the job correctly. The tool should be intuitive and obvious.
We spend a lot of time to some criticism that, in fact, we believe deeply that the users should be intuitive and understandable. All of our predictions should be explainable. We spent a lot of time in explainable AI. Part of that reason is that we know that our prediction, even that recommendation, is not enough to convince people to make behavioral changes. We built our own natural language generation that synthesizes the results of all of our predictions so, in a way, people can use it as an argument effectively. I have to go convince you to change your behavior. You could be a product manager, an account executive and I say, "I don't think we should work on this deal. We should work on this deal before that deal." Whatever the negotiation is, whatever the decision is. We want to arm everybody with the logic and synthesis of the data historically so you can come and say, "Deals that have this characteristic or this shape and size, we never win. We never beat this space.”
If we want it to be this space, we'd have to do these things. These things cost this amount of time. It cost this amount of money. Instead, if we took that amount of time and money and applied it to this other thing, it'd be worth this amount of money." That's a coherent argument that is difficult to argue. That's it. It's also one that opens up the discussion at the top level. It goes back to that break down the silos. Break down the internal incentive structures. Get back to the company incentive structures. That conversation and argument is the one that people should be having at the company level. It's the highest level goal of the mission statement. That is the kind of thing that we want to be able to provide. We spent a lot of time, money and manpower on making sure that our natural language generation can synthesize the arguments and the predictions in a way that anybody can understand it without having to understand an eigenvector. They don't need to know about that. They get the answer but it’s in a way that is defensible and explainable.
The phrase that comes to mind, I don't know what football team this is, maybe if you're into football, you'll know, "Do your job." Thinking about all of your experiences, this is an exciting conversation. It's where things are going. Twenty-one days of effort collapsed into a day. A lot of our technology automates certain tasks. It saves you 20% of your day. It's dramatic. There are certain technologies that say, "Let's go look at a million data points and figure out which 50 people in that million can get me a meeting with company A, B or C." That would take a BDR a million clicks to go do. Those are the exciting things are, “Where can I point AI to make more rapid decisions?” This isn't like a 5% increase in GDP. I remember hearing Bill Clinton at a Dreamforce conference speak once. He goes, "When I was President, we raised it by 5%. We had the best economic times ever." This is bigger than the internet. You're talking more than 5% GDP growth when you deploy this stuff properly. Would you agree that AI is bigger than the internet?
AI has an anchoring to the internet. It's like irreducible complexity. The value is from the synergy of the data availability, which is what is shaking the world, is that things that were relegated to libraries and the halls of hedge funds are now freely available on the internet. Anybody can get that access. Here's the problem. Computers do at least two things that are better than humans. One is that they can consume an enormous amount of information much faster than we can. That's why going to the library is a much slower process than having the internet.
The second thing is that we are not, as human beings, very good at synthesizing through time. We are not particularly good at remembering the past as accurately as it was. There are a lot of good reasons, biological and evolutionary reasons why we do that but it's the fact that that happens. As a result, if you have an enormous amount of data readily available to us via the internet now then you have AI to crunch that data and synthesize it. You have the potential, if you do it correctly, to have a much more objective, accurate representation and synthesis of how things have moved through all of time. Both because, as a human, you can't simply read it all. Two, even if you could, by the time you finish the last book, you've forgotten all of the value of the first book or you have some manipulated memory of it. It's the same exact thing in a firm. It's that's why we say, "I want that AI to become the best sales engineer in your business because that AI will remember accurately all of the data that has all ever happened in your firm and synthesize it correctly for you.”
To bring it up to a general form of, "Is AI going to be bigger than the internet?" The internet is that library. AI is the reasoning we apply to that library. In terms of GDP, it will drive much higher than even the initial booms of the internet. We are still ways away from that. I know a lot of people are very hyped up on AI. Things like neural networks and stochastic learning systems like this are not the way that human being reason about the world. In fact, we're quite bad at thinking in probabilistic terms. Once we start getting into spaces of causality and causal inference and we start to spend more time in knowledge representation, semantic understanding and knowledge graphs, we will start to see the power of what computers can do better than us. Synthesize all of that data, both in terms of volume but then accurately through time.
My son's home for the summer from college. We're playing this game. I'm not going to remember the name of it but I'll give you the concept. The concept is you get you to pick three chips, colors, red, yellow, green, blue, brown and then you can buy these cards. When you play the first game, it's very basic. My son is an engineering thinker. It's all about the math. If I can reserve a card then I'm losing two chips. It's plus, minus, and card counting. There are expansion packs. It changes everything. You put on two other rows of cards. You can reserve this card. There are all kinds of complexities. At my level in life as a 48-year-old in Colorado, I can feel the ability for my processor, it's not as fast as the nineteen-year-old Engineering student processor. What if everyone had the ability to play that game to where they knew exactly what the right next move on the board was? Now you apply that to companies, countries and running everything, things are going to be changing. We're getting to be a lot smarter.
AI is the anchor to the internet.
The problems get harder too. All of these things, like this game example you have, you add a new layer and the complexity increases because the complexity goes as the number of connections between all the objects. That scales much faster than the number of objects. You end up with a need to synthesize data. Right now, in the general market of any business, if you're a leader in AI or if you're in this machine learning space, understanding that you will always have a profitable product, synthesizing information because there's more information every single day, much more than any human can keep up with. In fact, we're starting to get to spaces where most systems can't keep up with it. Being able to reason about information, have systems that can do that algorithmically. The key point, in my belief, is not to just give you the answer but give you the reasons for the answer so that you can see how it's working, ”Can I understand something that might improve in my own understanding about the world using this tool to synthesize all this information?” That's the key value forever but at least the next twenty years, it could make a very big emphasis.
What's the biggest obstacle to salespeople trusting the AI?
It's primarily that explainability, which is why we invested so much of that upfront. That was a strategic decision that we made a pivot. We recognize that the recommendation is not going to be enough because then you’re asking for an enormous amount of trust from the user. There are two answers to this. One, what is the thing that's going to hold people back from adopting this? It's themselves. The second thing is because people won't meet them where they're at. One is they will think that they know better than all of the data on the machine and the predictions, which they probably do. In a lot of ways, that's been the experience of AI. The question is, will companies recognize that and say, "If humans are going to behave this way, I need to meet them where they're at. Let me invest in explainability. That way, it's not a matter of trusting me anymore. It's a matter of agreeing with my premises. Are they clear and articulate? Do I have a valid argument?
It's not a black box. You're exposing the algorithm under the black box that I trusted.
Not only that. We expose the whole argument. It's not just the algorithm. You could say, "Here are the features that are involved. This is how we waited." It's different than that. It's more of I'm synthesizing it for you and saying," These are literally the arguments." I can go back and point to these other deals so that you can say, "Did we get it right?" A lot of times, you can make predictions. You can even say, “What are the most important features?” People think that what we mean by explainability is using Shapley or something like these algorithms that give feature weight importance. This is different than what I'm saying. I'm saying we're going all the way to the space of making that human argument to another person and then referencing previous cases to say, "These deals look like these deals. This is what happened in these deals accurately. Did we get it right? Is that a reasonable argument to you? If so, that's the way that you and I would discuss." It's not AI as much as it is trying to be true intelligence. Can two people talk to each other? That's what the computer should be able to do to you.
We've been talking with Joe Miller, Chief Data Officer at Vivun and a lot of experience with IBM Watson, Ray Dalio's hedge fund. What an awesome conversation. Great talk with you, Joe. Thanks for sharing with our audience. I appreciate the connection. I wish everybody the best. We'll catch you on the next show. Thank you, Joe.
Thanks for having me.
About Joseph Miller
Physicist and AI/ML expert specializing in causal inference. Ex-Bridgewater Manager of Technology and Intelligence responsible for building the AI platform to systemize management decision making.