Jacob is the Vice President of the AI Platform Division at Gong – the AI platform that transforms your revenue growth – and he joins us today to share his joy of giving people the freedom of a data scientist even if they have a limited technological background.
Jacob shares how Gong uses AI, how it empowers its customers to build their own models, and how this ease of access for users holds the promise of a brighter future. We also learn more about the inner workings of Gong and how it trains its own models, why it’s not too interested in tracking soft skills right now, what we need to be doing more of to build more trust in chatbots, and our guest’s summation of why technology is advancing like a runaway train.
Key Points From This Episode:
Quotes:
“We don’t expect our customers to suddenly become data scientists and learn about modeling and everything, so we give them a very intuitive, relatively simple environment in which they can define their own models.” — @eckely [0:07:03]
“[Data] is not a huge obstacle to adopting smart trackers.” — @eckely [0:12:13]
“Our current vibe is there’s a limit to this technology. We are still unevolved apes.” — @eckely [0:16:27]
Links Mentioned in Today’s Episode:
Jacob Eckel 0:00
It is the demarcation point because it looks like it is capable of thinking.
Rob Stevenson 0:10
Welcome to how AI happens. A podcast where experts explain their work at the cutting edge of artificial intelligence. You'll hear from AI researchers, data scientists, and machine learning engineers, as they get technical about the most exciting developments in their field and the challenges they're facing along the way. I'm your host, Rob Stevenson. And we're about to learn how AI happens. Here with me today on how AI happens is the VP division manager and first overall hire over at Gong, Jacob Eckel, Jacob, welcome to the podcast. How are you this morning,
Jacob Eckel 0:48
I am great. Thank you for having me on your podcast,
Rob Stevenson 0:51
I am really thrilled to have you on Gong as a company I know decently well. And for anyone listening, I'm sure they know it as well, too. But if you don't go ask anyone in your sales department. And I promise you, they know all about going. It's an amazing company. And you're doing some really cool things that technology, which is what I want to get into with you. But first, I would just love to know a little bit more about you and your background and how you came to be gone.
Jacob Eckel 1:15
So I spent most of my career with startups. I like this thrill of taking startups from from zero to one. I did this a few times. My early career was in the financial financial technology. Again, today, we would call it AI. But these days, we call it just all simple statistical inference and, and predictive analytics, stuff like that machine learning, they would call it AI. And so with Gong, we happily call it AI. And the same technology has just evolved immensely since the early days. So I did started with Gong in 2015. I joined the founders, Alana Demeter, John, the actual before the company was incorporated formally. And since then I'm here at Gong. And we're still trying to take it from zero to one. So though, as you mentioned, we probably already arrived, maybe maybe at one. Let's see if we can arrive at two.
Rob Stevenson 2:13
Yeah, it feels like you're maybe one one and a half these days. None for me to say, Yeah, I have to say, I love speaking to people in your position who remember when AI was math. And now it's funny that you have to, like you said, the technology evolves, here we are and gone. For anyone who doesn't know, maybe I'll not pitch the products. Since I have employee number one on the line here. Would you share a little bit about what the company does? Because I think that will set some important context for when we go into the cool sorts of tech you're deploying right now.
Jacob Eckel 2:41
Yeah, so So Gong is AI powered revenue intelligence platform. That's how we look at it. So it's a platform, a suite of products that leads and supports salespeople on their journey from lead to closure of the deal. And we cannot wrap them with this technology with AI technology to make their work much easier and effective. So that's our mission to make customer facing teams much more effective using AI.
Rob Stevenson 3:14
Yeah I've used gone a couple of times, I actually there was a moment where I spoke to an early marketer about a job at Gong and my life would be much different if I had taken that. It was unfortunate because I had just taken another role when I sat down with like one of your first marketing hire. So we might have been coworkers, Jacob, but as it is, it's you know, six years later, and we're doing the podcast. So maybe we were destined to meet but it can try again. Are you hiring for podcasters? I don't know.
Jacob Eckel 3:39
Maybe you can check. It's possible.
Rob Stevenson 3:43
would actually love to upload one my podcast episodes to Gong and to see what it says. And they're like, oh, Robert says here, You talk way too much. You're not asking enough questions.
Jacob Eckel 3:51
certainly possible. But they I don't think that will happen today. Probably I will be talking too much.
Rob Stevenson 3:57
That is the point no one's tuning in to hear me talk about AI. That's for sure. So my experience with Gong is you have these sales calls, you're able to feed the sales calls through Gong, and it gives you all these insights and tells you Oh, you need to spend more time here in the conversation. We've compared it to all of these cases where deals closed from this call. So it's fantastically valuable insight on the sales process. And so I would just love to know, where does the AI come in?
Jacob Eckel 4:21
Yeah. So the magic word Gong happens because we have a bunch of information bunch of data about what is going in, in the sales processes. Mainly, we can think about the data in in three categories. One is the conversations sales conversations, which would transcribe and then we have all the texts of conference calls and phone calls, where the emails of the customers and we have the CRM data. And so the mix of these three main we have a little bit more additional events, but mix of these three components or data components is what creates the six the like view around the customer will see everything going on. And we can combine the data using AI to provide a help and insight to the customer. And so AI actually used in several ways in common in many ways in Gong, first of all the actual transcription of conversations. It's like an old story these days, but you'll be amazed of what is done to make these transcripts correct. And to capture exactly the conversation that is going on. It's a non trivial AI technology, even though that was the thing, it's it is like, all technology, it also evolves all the time. So it's like basic, basic AI technology that that we have. On top of this, we have several types of models, some of them are built by gang, some of them are built by the customers using their data in Gong. Some of them are the standard models that you can buy in the market like GPT, or entropic. So we use the models available on the LLM, mainly available on them in the market. So from these three types of models, we can construct any use case that we need for our customers,
Rob Stevenson 6:15
I'm pleased to hear that you are sort of empowering customers to build their own models that feels like an important shift, really in software. In general, right? It's not merely here's our product, you subscribe to our product, it's like our product is going to allow you to build your own product, essentially. And so while empowering those customers, what kind of things are they building with their own models?
Jacob Eckel 6:35
At some point, we decided that we cannot build every model for our customers, because we don't really know which models to build. It's a huge scale, huge number of customers, and they have their own ideas. And so at some point, we decided that it would be a great idea to give them the ability to empower them to build their own models using their own data, the data they keep with us. We call this capability, smart trackers. And basically, it's a active learning system that allows like, not data scientists, customers, they're not there to say, well, we're not expected, our customers suddenly become a data scientist. And we'll learn about modeling and everything. So we give them a very intuitive, relatively simple environment in which they can define their own models. They can decide what they want to track, which events and the conversations they want to identify. And I'll build these models for them. So things that the customers do, for example, identify competition for each customer, we don't know who are their competitors, they know perfectly well so they can train a model, which identifies when their customers mentioned competition. And like this one example, another example, like things that customers complain about, okay, let's say that some of our customers want want to know, when their customers complain about their support. It's a very important event that they want to identify. Or many times they build a trackers related to a legal compliance. They are salespeople are supposed to say certain things in certain way. It's a legal legal requirements. Surprisingly, they're quite a long list of legal requirements that apply to salespeople. And so that depends on the industry and the situation. And so the companies want to know whether their their salespeople actually comply with these these requirements. But these are typical models that they build. Also things like a various playbooks says playbook says, But playbook is a set of kinda recommendations, standard recommendations that many, many companies apply, and they want to know whether their sales force follows these these playbooks and so they wanted identify the elements of these playbooks. Can you give an example like for example, metrics, what does the organization need to succeed or economic buyer whether they identify the economic buyer, right, when they decided during the calls, who the economic buyer is and the company wants to know whether these steps is basic steps of sales process are followed? And? And if not, if they forgot, for example, to identify the economic buyer, they can come to another meeting and start from from this question. Exactly.
Rob Stevenson 9:21
So Gong is sitting on tons of data, right? Because all of these sales calls from all the customers are recorded. But when it customers build their model, are they leveraging all of that data? Or are they leveraging just their own data?
Jacob Eckel 9:34
It depends on the model. So if the model is built by the customer, it's their data and their model, they cannot use other companies, there is no sense to use the company's data, because they want to build something for themselves. However, we also provide prepackaged models built by Gronk. In this case, we build this on all data that we are legally allowed to use not all data we are allowed to use, but We have very large actual amount of data that we can legally use for building models. And so once we build the models, we would give it to all of the customers.
Rob Stevenson 10:08
And then what about if a customer like a smaller company, they've not been around that long, maybe they don't have that much of their own data? Is it still meaningful for them to build their own model?
Jacob Eckel 10:17
Yes or no. So like, in the beginning, you can start building your models because you don't have enough data. But typically, this waiting period that you need to wait and accumulate data that can build a model for yourself is relatively short. Like I can give you an example, let's say a 10, salespeople company probably will need to wait a month or maybe two, until they have enough data to build basically any model they want. So it's, it's a kind of a waiting period, but it's not too long.
Rob Stevenson 10:48
So it's a couple of months. But what is like that critical mass of data like that could probably take a varying amount of time,
Jacob Eckel 10:54
then is just the basic limitation of basically the number of words that has been said, until we get enough data that you can bring the model and we control for that. So we would count the conversation that happened. And once we see the critical data is the chip, we unlock this capability. Again, they still can during this time, they still can use the pre package models available from day one. On the other hand, if the company is large, they can reach the critical data like in one day, I told you it's like a small company with 10. Salespeople, they needed to wait a little bit, but a large company they like in two days, they have a sufficient number of data.
Rob Stevenson 11:31
How many words is it?
Jacob Eckel 11:33
I said? It's I think it's around 500? Conversations.
Rob Stevenson 11:39
500. Okay, yeah, so huge company a couple of days. And I guess it depends on like every company is different, right? The sales cycle are longer some half hour call, hour long call long demo, whatever. So, okay, but yeah, that is being tracked, and they're building towards, okay, here's your progress bar ding, now you have enough data to make a model that's like, as a part of a normal sales cycle, you will reach that it's not like you need decades of legacy data to be able to play.
Jacob Eckel 12:04
In many cases, by the way we import the data they already have before they joined Gong, some companies collect data even before and so again, in this case, usually on the first day, they already have the needed amount of data. So it depends on the customer. But really, it's not a huge obstacle to adopting smart trackers.
Rob Stevenson 12:24
Gotcha. So what do you think of the models that Gong is building? How are you training them,
Jacob Eckel 12:30
we have dedicated teams to train. Now we're switching to the models that are prepackaged by Gong. So models prepackaged, the gong, obviously, are built by Gong. And we use our Dell teams, we have a great team of data scientists in my division and in other divisions also. And so they work on this, this is their job to build these models. They produce them, they call the factory, it's a kind of a process of pipeline that they need to apply how this data is collected the filter, how the models are created, there is a pipeline. So they work around this pipeline to create this model. So we add the new models all the time to the process.
Rob Stevenson 13:06
Gotcha. So the models that God is making anyway, they're not merely like a dashboard, here's all of your insights, like you would expect from software, look, the user is able to query it, in addition, correct?
Jacob Eckel 13:18
the models what they produce. So we didn't talk about the generative models, these are classifiers, that the models that we mentioned up to now that classifiers. And what they do, they identify events in the stream of conversation. And then you can do something with these events. So it's like you have all this streams of conversations and emails go every single is going on. And they'll just like pin, pin, pin pin pin put here, they talk about this on here, they talk about this. And here, they talk about this, and they can collect in various parts of the application. All these events are collected and then present in various parts of the application to help with the task they're doing. For example, in the pipeline review, they can see these events and say, Okay, this is our pipeline review is like seeing all the deals that are currently going in the in the company, and they can see, oh, let's look at this deal. In this deal, the decision process didn't happen, the conversation or decision process didn't happen. We don't see this market, we don't see this pin created by the model. We need to need to talk about this. So we know we know what we need to talk about as the customer on the next meeting. So it's it's it's like provides you the assistance you need in the process. So that give you a little bit points of interest in the in the conversations.
Rob Stevenson 14:37
Okay, right, right. So in those cases, how are you pulling out those important things? Like how did they decide okay, this is really crucial to a deal.
Jacob Eckel 14:44
If the customer decisions really depends on how customers structure their business processes. We're not in a position to tell customers how they do their business and different customers do seriously different things in the database process. They different industries, different people, different managers, you know, there are various sales methodologies in the market. And different managers adhere to this or that methodology. Some of them do not believe in sales methodology, some that some of them do. And so they decide so we have this configuration screen where they can kind of decide they were interested in these events were not interested in in these events that will pop up what is really interesting for them for this specific customer.
Rob Stevenson 15:27
Got it? Okay, I asked, because, as you called out, there are all these specific things that are really important to hone in on no matter to whom you're selling, where you're selling things like Did they mention competitors is the economic buyer involved, these are just like this is the things that salespeople are trained on, right? Like, this is what you need to look out for. So of course, that's kind of made its way into the product. But also, if you spend time with salespeople, they spend all this time developing rapport, small talk, building trust, trying to add value for a long, long time before they make asks things that I would call the softer skills in a deal. That's like less listening, you maybe can't log in Salesforce, and then you can't say like, oh, did they have budget on time? Or do they have budget this quarter? That's like a yes, no answer. That's information you can solicit. But it's not like the joke you tell in the tee box at the golf course, you know, like some of these sort of stereotypical salesperson things? Are you tracking those like softer skills, these like rapport building? Is that part of it?
Jacob Eckel 16:25
The question is no. And we have been asked many times about this aspect. And I think our current vibe is there's a limit to this technology, we are still a very evolved apes and stuff. Not everything that we can do. LSA is really trackable by automated tools. And all that interesting that the history of natural language processing. Sentiment analysis was a very significant part of the reasons why a natural language processing is developed. And people use it successfully for various things like resupport, or to categorize feedback in sales since applications. And so so people ask us from time to time with a sentiment analysis is like a thing. And in our expert opinion, we did lots of work in this area, it still looks like there is some element, as you say, the softer elements of human communication, that are not yet trackable by machines. So no, we don't. We don't track jokes yet. Some people ask us, do we track the facial expression of the customers? That would be very interesting. We don't do this yet. But who knows the technology evolves all the time, it is possible that we will start tracking, like facial expression and body language and stuff like this. That starts to sound a little bit scary, but but we know it's not there yet.
Rob Stevenson 17:56
Right? Okay. The answer is not yet. But I asked because I'm wondering, like, doesn't matter? Does that stuff that like all a lot of salespeople I've met that they tend to prioritize? And like this whole relationship building aspect of sales. And they I've heard that a million times sales is relationship based. Is that true? Or would it be better off to just cut to the chase and just do all of these things that say, Hey, if you do these eight things, Gong has proven that this is means a deal is more likely to close? I don't know.
Jacob Eckel 18:25
Yeah, we cannot disprove that the softer parts of the communication are as useful as the like the following a playbook we cannot prove that. And my feeling is that there is lots of things that that going through the conversation that are nonverbal, soft stuff, jokes, stuff like this being bringing, like a great conversation to a friendly to a friendly level stuff like this, I think it makes a lot of difference. We don't have yet the science to prove this point. I'm quite convinced that there is lots, you know, as if you ever saw anything or bought anything. And I from time to time I buy stuff from vendors, I feel there is a huge difference whether the salesperson is friendly, and how how they position themselves and how they talk about things and whether they are interested in you as a buyer. It's a valid question. And I think that there is something there. We are not yet ready to do science on this, but it should come at some point, I suppose.
Rob Stevenson 19:25
Certainly, when you say that we're not quite there yet. With this sort of analysis. It makes me think about how every company is they're trying to build an AI chatbot they're trying to mimic human conversation. And the expectation is that a human will interact with this AI as if that AI is another human. But yet the part of being human the telling the joke at the tea box right is not there yet. So is that a crucial piece to solve before we can expect people to trust a chatbot to trust an AI avatar?
Jacob Eckel 19:59
It's a long way, I think we're very careful not to say, you know, like five years ago, it would say, oh, it's science fiction. Now. Now we'll come. Now we are much more careful saying no to such ideas, because we see the rate of development of models, and they are becoming smarter and smarter by the day. And so we are careful not to say it will never come. It's not there. The models, the chatbots, are not capable at this point, even the smartest smartest of them, that they don't have a drive. They don't have ideas of themselves. They cannot ask questions. The basic thing that no chat system in the world today is capable of producing the questions of itself or like clarifying itself. Then, if I ask something, and it's not very clear to the model, it could ask please explain what you mean. It doesn't do this at someone's or stripe does this but they don't do this reasonable like we people do like trying to maintain a conversation, which which goes both ways. These models are very unidirectional. If you ask them questions, they will answer. It's not a real conversation going on there.
Rob Stevenson 21:09
Great point. And like you say they will answer they won't say, I don't know, they might say I can't tell you that.
Jacob Eckel 21:15
They won't say Can you clarify that they won't say what else you would like to know about this, unless it's like just a mechanical thing that they ended the conversation, what else I can help you. But the human conversation goes way beyond that they like when you talk to me, you're trying to understand what bothers me. And then sometimes you ask questions that I didn't understand or didn't have the idea to ask you help me because you're like putting yourself in my shoes. These models are not capable of doing anything like this. And so in very, very simple sales scenarios. Yes, the agents will, will work in very simple support scenarios, agents already worked well. But for real, significant sales conversation, they're not yet there.
Rob Stevenson 21:59
When you say that you're very careful not to say to never say never right? Say no, you want to have this. You say it's because just the rapid growth and progress we've seen recently is so is so immense, right? It'll make your neck snap, right. And I hear that on the show like, oh, the pace of change, and the progress in this space is so fast. And I was tickled I had a guest refer to traditional generative, meaning like degenerative we were doing in like October last year. And so you know, we take that for granted. Of course, it's aI moves fast as tech. But here's I don't know, I mean, it's a squishy question. But do you have an opinion on why things seem to be getting faster and faster? Is it like increase in processing? Is it more investment in awareness, or more people working in the space? Why is the space moving faster than ever,
Jacob Eckel 22:45
we suddenly understand that the opportunity in this area is more or less endless. It was not the realization like five years ago, we knew there is the opportunity, we knew that we can make many interesting things we have gone started at the like, start as an AI company, basically. And we knew that lots of things can be automated, lots of things can be done. But suddenly, in the last few years, with the development of things like GDP team, we suddenly realized that the bar of what is possible, is way, way, way higher than we saw before. And this changed really the mindset of people working in this area suddenly realized that the ROI there is like infinity ROI. We've never been in this position where we could say, Okay, we invested this technology, we develop stuff such and such and we will get some ROI reasonable one or very good ROI on investment in this area. And currently, we are in the position that many people think that ry is more or less infinite. And this is why people invest more money. People find ways to bring data, people bring more hardware, which which is basically money energy. People bring more brains, like people into the business, because when the ROI sounds like infinity, like infinite ROI, you know, people will will stop doing everything else and focus on their thing. And currently in like in our environment in tech, this is the only thing that we think that they are wise is infinite. So it's not surprising that everyone talks about AI everyone loves him.
Rob Stevenson 24:33
What do you think is responsible for that realization? Is it generative MLMs? Is that the breakthrough?
Jacob Eckel 24:39
Yes, I think that the first GPT or around GPT strip and five maybe or around this point in time, we suddenly realized that the potential is much higher than we saw before. We had this glass ceiling in AI. We knew more or less what things can be done. But this glass ceiling like Oh, He went upwards. So high, we don't see this new ceiling, we don't know where it is what will stop, something will stop, nothing grows exponentially forever. So I belong to the, to the class of people that think it's more like S curve. It's not like exponential grows to infinity. At some point, you'll see the rounding of the S curve, but we don't see it yet. So, count, it looks like no glass ceiling at all.
Rob Stevenson 25:26
GBT three and a half and later, is that kind of demarcation point, just because this is the conception of what we always thought AI was this idea that this ability to commune with a machine as if it were a person,
Jacob Eckel 25:37
it is the demarcation point because we suddenly realized that it looks like it is capable of thinking. Because before we said, Okay, I will recall that machine learning machine, learn something, and that's something. But suddenly, around GPT, three or 3.5, we realize that it fakes thinking, but it certainly looks like it is thinking and so it brings the insight that was not possible before absolutely not possible not not dreamt before. So people didn't expect this level of thinking.
Rob Stevenson 26:16
That's really well put that it looks like it is capable of thinking, which is also what my first grade teacher wrote on my report card. We'll see what he does with it. But in any way that was really beautifully said, Jacob. So thank you for sharing this with me and for sharing all about Gonzo models and your opinions on generative and where we are now we are creeping up on optimal podcast length here. So I think this is probably a good place for us to wind down. So at this point, I'll just say thanks again, Jacob, for being here. I really love chatting with you today.
Jacob Eckel 26:43
It was great chatting with you. Thank you for having me on the podcast.
Rob Stevenson 26:48
How AI happens is brought to you by Sama. Sama provides accurate data for ambitious AI specializing in image video and sensor data annotation and validation for machine learning algorithms in industries such as transportation, retail, ecommerce, media medtech, robotics and agriculture. For more information, head to Sama.com