Transcript: Eilon Reshef | Gong | How Do You Build a Product for Sales Organizations

Eyal David: [00:01] Hi friends, this is Eyal. You've reached Product Builder, where you can listen to the most interesting product conversations. My journey began in 2012, when I learned what product is at my own startup. Since then I've worked in the industry, and since 2017 my company has been advising a range of companies. My drive is to help all of us get better as product people. And before we start, if you enjoy listening to Product Builder, please rate us five stars on Spotify. It gives me great feedback that the podcast brings you value, and it would make me very happy. So let's get started. Eilon, how's it going? All good. So great to have you here. Thanks for having me. Of course. Thanks for agreeing to come. We have Eilon from Gong, of course. Would you like to tell us in a few words about the company, really take us inside?

What Gong does and how the product evolved

Eilon Reshef: [00:50] Sure, sure, gladly. So it's a software company, software as a service, that has existed for, I think, nine years already, which is really a lot. We provide a platform mainly for sales organizations. Today everyone calls it AI; we started when nobody yet knew what AI was. But in a nutshell, we basically understand everything that happens in the communication between the salesperson and the customers: calls, emails, texts, whatever. We run AI on it, and then we can really help them, whether it's producing call summaries, writing emails for them, telling them what they should and shouldn't do, helping the company understand what works and what doesn't

Eyal David: [01:27] work. Lots and lots of things; basically the whole organization works with it. Really all the way up. Yes. And tell me, from the product you started with, you say in a nutshell, but how much has it actually changed?

Eilon Reshef: [01:38] It has evolved a lot. I mean, we still know how to record calls, yes, but today recording calls is easier. When we started there was no button to record and transcribe. We have our own engine, but today you can transcribe with any engine, yes, it's not new, it's not advanced technology. But there are still many customers who come and say, I use Gong first of all because I can reliably bring in all the emails, record only what's needed, and extract insights in the best possible way. On the other hand, there are also new products: products that let the organization do forecasting, manage the customer pipeline, correspond with customers, lots and lots of things that didn't exist at the beginning.

Eyal David: [02:11] I assume that over time you also started moving up in the organization, right? You started with sales reps and it began to expand upward?

Selling to the CRO: first customers and fast value

Eilon Reshef: [02:19] No, actually, actually we started from the very beginning with the CRO, which is the Chief Revenue Officer, the person who makes the decisions. When we started, we tried to sell to an organizational function called Revenue Operations, which is like the IT of the sales organization. And we discovered that, my brother was in that role at the time. I told him, bro, come buy the product. He told me, listen, I'm so busy, I have no time for you. I said, if my brother isn't willing to buy, it's probably not the right role. And then Amit and I understood that you need to sell to whoever is responsible for the money, and then it's much easier. You come to the person and say, listen, you put in Gong, you'll see better close rates, what's called attainment, more salespeople will hit their targets, you see everything. Faster onboarding. And that's something that speaks to every sales leader. And was that already something you could claim after a while? We could say it at the beginning as a hope; we said this is what we believe the product does. To the first customers we said, listen, it's a first product, check it together with us, let's see. But very quickly they saw it themselves. I mean, our first customers, even the first ten, at some point we came and said, the pilot is over, please pay. And there were, I think, two who said, no way, too expensive. We said, fine, no problem, we'll cancel. And then they said, no, no, wait, wait, don't cancel. And then they saw the value there very fast. It's a product where one of its advantages, and also disadvantages, is that you get value very, very, very fast. And you get value very, very fast, and basically all those companies have stayed with you since? Most stayed. Some simply went bankrupt, unfortunately. You know, tech companies, nine years is a long time. But many have stayed since 2016. Soon maybe we'll give them a gold watch for the tenth birthday. Right, it's also that kind of product, right, that once you're in

Eyal David: [04:01] it becomes another work tool that's hard to rip out of the company. Yes, yes, although again, there are companies that aren't with us today, that's a bummer. There are companies

Eilon Reshef: [04:08] that are with us but don't have a penny to their name, and that's a bummer for them too, and in the end for us too. But yes, we have many customers from the first batch of 2016; I estimate most of them are still with us.

Eyal David: [04:19] Amazing, amazing, that's one of Gong's strengths, I think. Another thing I saw before you came is your values; you have a whole page on the website. Would you like to expand on that?

Company values written bottom-up: no sugar

Eilon Reshef: [04:31] Yes. First of all, we have quite a few, so I don't think it's worth going through them one by one. But what we did was a really nice process. Because in the end, when three people sit in a room, everyone sees each other. When an organization grows, it's like a tribe: in the end not everyone knows everyone, and then it's worth writing down what the values are so people know what they're supposed to do when they join the company. But you can't invent them. Say I announce tomorrow, I don't know, at Gong we're frugal. We're not frugal; everyone will see it, and it'll just be another thing written on some wall. So we grabbed almost all the employees at the time, I don't know, 60, 100, I don't remember how many there were, and we told them: write down what you think makes Gong a place you love to work at. And we did a kind of post-it notes on a wall exercise and made clusters. And non-trivial things came out. For example, one of the things people really love is called "no sugar." We said: we speak, we tell the truth without fluff. Americans find fluff very easy; at Gong there's little fluff. We have something about customer love, very, very passionate, raving fans. We have a few things like that which became catchphrases inside the company, but they came from the fact that it's real. It's not that Amit and Eilon invented them; all the employees came and said, this is really what we feel when we come to work in the morning. Wow, that's amazing. It also shows that you know how to receive feedback and also adopt it. Maybe, but I'd say it's really at the most basic level, because if you try to invent something, it doesn't work. If I really, like, do it top-down, it doesn't work. So you say, let's take the good things and replicate them, not, tomorrow I'll decide I'm 2.20 meters tall and a good three-point shooter. That won't make me 2.20 meters tall and able to shoot threes. Yes, maybe, maybe in the future, maybe.

Eyal David: [06:08] So tell us, take us down memory lane. You started in 2006, 2015, right? And since then, how did you evolve, what does it look like today?

Gong today, and from CTO to CPO

Eilon Reshef: [06:16] So today we're at around 1,000 people, I think. Most are still go-to-market in the US. In Israel we have about 400 people, and we're hiring a lot. We also have a small office in Europe. We sell to all kinds of customers, really from very small customers, startups, including Israeli ones, all the way to the Fortune 50. I mean, a big search engine that starts with G, all kinds like that, a big car company. Gorgeous. And another thing that happened is that you moved from the title of CTO to CPO, right? Yes, yes. When we started, there were two of us, yes, so each did something: Amit did sales and marketing, and I did product and engineering. I even wrote code before we started the company; I pulled out an IDE, which I hadn't done for many years before, and wrote a bit of POC code. But at some point the engineering organization became a monster. I mean, hundreds of people, and that's not necessarily my forte and not necessarily my passion. So about four years ago, I think, maybe already five, someone came in as head of R&D, and I took on product.

Eyal David: [07:22] Gorgeous. So maybe tell us, to dive deeper into the different products, a bit about the landscape, you can tell us, but first a bit about what you do day-to-day as CPO?

A CPO's day-to-day: managing 80 product people and ambiguous features

Eilon Reshef: [07:32] Yes. So first of all, the product organization at Gong is already quite large, I mean about 80 people, let's not count exactly, give or take, it changes; sometimes various functions move between departments. So it's about 80 people. Obviously when you manage 80 people, you're managing managers, not managing people directly, unfortunately. And then a lot of the time really is people management. I mean, today I had a conversation with a product manager who wants to move from group A to group B: how do you do that, the group B manager needs to find him something, really things like that. I'm a bit administrative, but in the end that's management: taking care of people, making sure they have a career path, that they have clarity about what they want to do, hiring people. The second part is, let's call it, a global view of where the product is going and understanding what we'll develop next year. Say, the next three months don't need me. Of course I don't do it alone, but again it's driving the processes, and mostly, mostly, identifying areas that may fall between the cracks. Five people are working, each takes an area; what about the sixth one that nobody saw? That's where I invest quite a lot of thought.

Eyal David: [08:37] Is that also where you see yourself as the best, among all the things you do? I don't know what's best. I really don't, it's hard for me, hard for me to judge, but

Eilon Reshef: [08:48] I try to do the role, first of all, regardless. Maybe I'm the best programmer; that's not my role, yes, that doesn't change. Yes, you probably wouldn't be in love with it. Actually, I really love it as a hobby.

Eyal David: [08:58] Yes, as a hobby, exactly. And today, are you involved at the level of specific features, or really only in strategy, the way you describe it, like you dabble in it?

Eilon Reshef: [09:08] I'm absolutely involved in features, not all of them. I try to be involved in features with what I call a high level of ambiguity. I mean, a customer, a big enterprise, says, I need to manage people better, like users. But what is clear about managing users? It's hierarchy, it's really not clear, and there I think experience helps. Of course our group leads can steer it, but sometimes my experience can help and say, OK, let's look here, let's look there. So in those things that are either very, very, very strategic or very, very ambiguous, I usually inject myself into those processes. But I can't be involved in every single feature; I probably also can't help, because the product managers and the customers

Eyal David: [09:46] know better than me what works. I have, I have nothing to contribute there, really. Yes. Well, you just say "permissions" and I break into a sweat. Those are the least fun things to do.

Eilon Reshef: [09:53] I'm not fun, I love it, but fine. But if it's not fun to do, then all the more reason I need to invest energy in it so people do it right. So yes, it's always that thing B2B companies most hope not to deal with, because it's always a black hole. Right, right, it's a tax, because we all love creating value for customers. Sometimes it doesn't create direct value, but without it there won't be indirect value for big companies. Exactly.

Eyal David: [10:15] Gorgeous. So tell us a bit about which products are now basically the company's flagship.

The products, the competitors, and Microsoft and Salesforce

Eilon Reshef: [10:22] What the landscape looks like, a bit about competitors, I know they've started popping up, and that's maybe something that also changed since the beginning. Yes. So the category we were in at the start is called conversation intelligence, which is understanding calls and emails and what you do from there. There, I think today, Forrester, say, put us alone in the top right corner of the grid, the only leader, which is very nice. Yes, I sent it to my grandmother, but she's no longer alive, so she didn't see it. But since then we've moved to a category called revenue intelligence, which is basically the main platform of sales organizations, with which they run their day-to-day, a platform that sits on top of the CRM. We basically sell three different products, but it's really a platform; many buy everything together. There's conversation intelligence, which is how I take the calls and use them to improve people and share knowledge. A product called Forecast, which is how I predict what the sales organization will sell. And a category called sales engagement, which is how I manage ongoing communication with customers, for example prospecting: I need to send lots of emails, how do I make sure they go out on time. So those are basically the three products. We obviously sell all three, some older, some newer. And really, when you look at the whole category, there are players who came with a sales engagement product and are trying to develop the rest. There are companies that had a forecasting product and are trying to develop the rest. We, I don't know if it's interesting or not, but customers find it interesting, we're the only ones who developed everything in-house. Other companies, out of choice or lack of choice, had to do acquisitions, and then the products don't connect that well together. But regardless, it's real competition; these are already real companies at our scale, nine-figure revenues, and it's a fair fight. It's not some startup with three cats trying something. There are also the giants that have probably started entering the picture, right? Would you like to talk about that? Yes, yes. So first of all, in the end that's always everyone's fear, right? When you look at what stopped a company like Slack, it was Microsoft. What stopped a company like Zoom, again it was Microsoft, yes. It's always the big ones, the ones that have unlimited distribution. So in our world, that's Microsoft and Salesforce, which have CRM systems that don't do what we do, but it's very easy to say they do. Today it's certainly hard for them to compete directly, because they don't have our feature set and they don't, and we don't know, they can't create the value we know how to create for customers. But they could definitely come to customers and say, listen, Gong is good, but we're almost the same, take it free for a year. And then of course we need to fight against someone giving it away for free. So far we're holding up well, partly because someone who goes for it comes back to us after a year, because it doesn't work. But if one day they decide to take it very, very, very seriously, they can also bring 400 people and develop things. So of course we need to run forward fast.

Eyal David: [13:05] Yes, I think what works in your favor, besides the feature set and everything, is maybe also the data collected so far, which is clear, and the contracts you have going forward, and also the very fact that you sit in the slot that brings in money, as you said from the start. In terms of ROI it's the clearest, so it's hard to rip you out, say compared to Slack and Teams, for that matter. The blow they got. So maybe the situation is a bit different. Is there a competitor you keep an eye on more than others, that's worth watching?

Eilon Reshef: [13:36] We give respect to everyone, really. No, we don't have someone where you say, wow, that's the one. I think we see a competitor in probably a third to half of the deals, and we respect them and try to win as much as possible.

Eyal David: [13:47] Come on, I love that. OK, and another question, this one actually came from the community: there are companies that are technology-led,

Product-led and the pod model

Eilon Reshef: [13:55] there are companies that are product-led. Where do you see Gong in that regard? Yes. In the way we built the organization, product and engineering are really intertwined. I mean, anyone who knows Marty Cagan's books, for example, when he talks about that whole pod model, we really work on a pod model. There's product, and there's an engineering lead, and UX, and that's basically the building block we work in. So in a certain sense I'd even call it pod-led: each pod sets an agenda and runs forward very, very autonomously relative to what's common in the industry. Having said that, obviously in the end the goal is to create a product. I mean, as an organization, certainly product-led. The main thing, the main reason people buy Gong and not some competitor, including Salesforce, is that the product is better, not because the salesperson is better. Although Amit, my partner, likes to say he wants to build a company where the product is so good that mediocre salespeople can sell it, and the salespeople are so good that even if the product were mediocre they could sell it, and everything would be fine. But of course we see ourselves as a product company first, no doubt at all. Excellent.

Eyal David: [14:56] Can you describe a few challenges you're dealing with today? You as Eilon, of course.

Challenges at scale and the AI revolution

Eilon Reshef: [15:02] Yes, yes, yes. So there are many challenges; at our scale there are many challenges. We have challenges, for example, where big customers come and sign very large deals with us, and sometimes each is missing a penny to make the pound. How do you orchestrate it? We'll really run and give everything for these customers, but we still have a roadmap and still need to manage it, and we still don't want to promise things we can't do. So how do you manage that whole affair? And I see it as a good thing, not a bad one. But it still needs to be managed. At our scale there's also how you create a coherent picture in the organization of what the product does and what it will do a year from now. I talked about it a bit earlier, but really, how to make sure things don't fall between the cracks when everyone has their own agenda and runs forward with it. And there's one agenda that always interests everyone today, the whole AI revolution, where we were sort of at the front even before there was a front. And we're still leading it, but it's still an industry that reinvents itself every two weeks. So how do you make sure you stay at the front, how do you make sure you run enough experiments to discover something that adds value?

Eyal David: [16:06] Whether there was some trick we didn't find, things like that. Gorgeous. I'll pick up exactly on the last one, then we'll come back to AI. I think it's a challenge, but at the same time it holds within it the fact that AI has become kind of a commodity, let's call it, starting to be, so it probably also opened lots of doors for you, or, for that matter, lots of adoption that used to be harder to achieve.

Eilon Reshef: [16:27] Tons and tons. There are lots of companies that wouldn't have talked to us three years ago. Big banks in the US would never have even thought about recording calls; it seemed strange to them, let alone doing anything with it. And today it's not if, it's how and in what way they get into it. But honestly, it's also the value you can create. I mean, with all due respect, we have a very big and strong AI team and everything, but the value we knew how to create three years ago, we couldn't produce a call summary. We could produce pieces of it, but there wasn't, we can't develop everything an entire industry developed over five years. So now, yes, today the value a sales organization gets from using Gong is much higher than what it got three years ago. That creates lots of opportunities, and it's also very easy to feel. And just now I saw, we systematically do win-loss analysis with customers we won and lost. I just saw a win-loss analysis, yesterday or today, I don't remember, with a customer who was asked why he chose Gong, and he says, listen, it's magic. They asked him, what's magic? It's not us doing it, a third party does it. So he says, I sat with a salesperson, presumably he's the sales manager. We looked at some account, and I saw there's a box in Gong where you can ask it whatever you want. We asked it why a customer wants to buy their product, which I don't know what it is. It wrote 1, 2, 3, 4, 5. He says, I asked the salesperson, is that right? The salesperson says, not only is it right, I remembered the first three and forgot the last two. So your ability today to create value in that way is that really in thirty seconds you look at what the customer needs. And the next question could be why they want to buy and why they don't want to buy.

Eyal David: [17:47] That really is a bit of magic; it was unimaginable years ago. Gorgeous. And that also brings us back to the first challenge you mentioned: suddenly customer types like banks arrive, for that matter, and then you need to adapt to them, and adapt to big customers too, and also add the little extras for them, as you said. Basically everything is tied to everything, so it sounds like good, healthy challenges. I want to ask a question we usually ask here, and we'll come back to AI: in terms of product skills, really, after the whole path you've walked, at Gong and before that at a previous startup, and as noted your path in product started even before, what do you think are the most important skills for a product manager?

Product skills: understanding what the customer really needs

Eilon Reshef: [18:34] So I think product management is a bit of a profession that isn't defined the same way everywhere. There are lots and lots of skills: communication skills, discovery skills, product-building skills, project management skills, communication skills, collaboration skills, which was the hardest to keep, really. I think it depends a lot on what the company's emphasis is and where it applies its product managers. For us, the first skill we look at is the ability to do, there are names like customer empathy, but I don't like it, but it's really sitting with the customer, looking them in the whites of their eyes and saying, OK, the customer needs this, and this is what they'll get from it. Because otherwise you become a kind of paper shuffler; you move the customer to the side of the button on the right, and that's not what we want. So really I think all the product managers here know how to do this really, really well: sit with the customer, understand, listen, understand what they need, not just what they want. That's number one. Yes, there are a few more skills, where in the end, OK, build a fucking tank, I mean in the end you should also build it properly and communicate it outward, even if it's internally to salespeople and so on, but for us that's the first one. And I also think anyone who looked at product managers at Gong today would notice that this is where they're really, really good. And of course it may come at the expense of other skills, which are perfectly fine, but we're not that picky when we interview new people. And you're hiring now, right? Yes, yes, we're hiring today across all roles. It's funny, we're now looking to fill about 60 positions, something like that, and that's rare. We're talking about where we are in September 2024, and it's a bit exceptional in the industry right now. So it's nice, because there really are many people who are looking for a change, and also, obviously, it's been an unpleasant year, so many hesitated before even raising their heads. So it really is a nice process to be in that situation. So there's another challenge, and basically, so

Eyal David: [20:13] of all the skills you listed, there's one that tops them all, which is basically understanding what the customer wants.

Eilon Reshef: [20:19] Yes, yes: sitting with the customer and understanding exactly what they want, why they want it, why they need it, and what they'll get from it. That's number one. Otherwise you just build things that aren't worth anything.

Eyal David: [20:29] Yes, we've all been there, I think, at least I have. Where did it catch you at the beginning of Gong? Was there a situation where you built things that weren't in the right direction, where you needed to bake them more, needed to do the archaeology better?

Eilon Reshef: [20:43] Yes, I think it's no coincidence in the end that we strengthen this muscle, because it's a muscle we really believe in. So I don't think we made big mistakes there. I mean, there were cases where we said, OK, the customer needs this, we built it, and saw that we weren't managing to create value for them in some area. But I don't think so; really very few things we built turned out that nobody needed, if any. But that's because we over-invest in these areas; we sometimes even overdo it a bit: come explain to me why you really need what you've already asked for three times. It's worth a lot of money, as you said, you're chasing it, of course. Yes, but I think that there, really, no. There are other places maybe where we missed; there I don't think we missed.

Eyal David: [21:18] And could it be that your customers are very unforgiving, for that matter? Because it actually sounds to me like salespeople are exactly like that, they're not forgiving, right? If there's a mistake,

Eilon Reshef: [21:27] they'll be measured on it, and so on. I think actually salespeople aren't, in the end they're business people, yes. As long as it's in the right direction, it's fine. It's not, say, if I were building, I don't know, a medical device, that's less forgiving, or an airplane engine, or a pager. A problem in the software business, OK, no, but software also not for airplanes, software for human beings. So at worst, they'll do you a favor and say, fine, I skipped it, it's not all day.

Eyal David: [21:55] Interesting. OK, what other challenges do you notice you have today in the product? And you can also think about other companies, because we're in a period where products are fairly similar, at least from a UI perspective, say. So what challenges do you identify today?

Shipping the org chart, and the Engage product

Eilon Reshef: [22:13] Challenges or pitfalls, you could say. Some of my challenges we've discussed. I think within the product itself there's a cliché I somewhat believe in, which says every company ships its organizational structure. I mean, in the end you say this group does X, it has a tab called X, and the stuff is inside. I'm not against it, by the way, it's fine; just let's arrange the org structure to fit the customers and not the other way around, because it's probably hard to avoid that situation. But I think that still, because we built products, the latest product we launched a few months ago, we still have a bit of that symptom of here's a tab that does something and a tab that does something. We have quite a lot of work to do in the coming year to look holistically at the platform and say, OK, it's nice that the product is called Engage, but actually it should be a bit here and a bit here and a bit here, not a tab called Engage with everything inside. So I think in the user experience there's work to do there, but let's say it's fine, it's not terrible. And what is that product, actually? Fill in the ear. The Engage product is a product called sales engagement. It helps salespeople reach out to their customers systematically. The most common use case is prospecting: how do I reach out to customers I don't know. So you can really say, here are these hundred people, create a task for me every Tuesday to email them, then on Wednesday on LinkedIn, on Thursday call them, whatever. That's of course very systematic, but there are lots of processes like that you also want to handle systematically. For example, I want to call a customer 30 days before renewal to ask if everything's OK, so I also set up tasks automatically, with a built-in telephony system and email system that help you do everything super effectively. Gorgeous. And everything's measurable? Everything's measurable, and we invest a lot, a lot in reporting, because in the end the organization wants to understand what happened, I mean why one person sent 10 emails and another sent 50 emails. What, who prefers which emails, what's better, and so on. And when you say you launched recently, is there already data, a quarter, say? It went to market about a year ago. It feels like yesterday, but it's a year, and there's already, it's positive. There are many hundreds of customers, like eight-figure revenue, so it's already a very successful product. I think as a customer I meet it from the other side, all that goodness, with all the bombardment, basically. Those bombardments, we're also trying hard to educate the market not to bombard. The direction the product is built for is more personalization than bombardment, I mean steering the person, as much as we can, to say: don't send to ten people, send one good email to one person after you've researched and seen they need it. Whether we can force customers to do that, I can't tell you, but we don't have a feature that lets someone send email from six different mailboxes so they won't get blocked, because we said that's not a priority right now, and we'll leave something for the competitors to do. Yes, yes, of course. I think actually in the future you'll head there; it's like how, for a long time now, you tell people, OK, you talk too much on the call. So right, it's basically the same idea. Absolutely. The whole idea is that software systems, data and AI can help you do a better job; ideally they'll replace the boring parts, and what they can't replace, they'll help you do well. Nobody likes doing bad work. Yes, that's very much part of it.

Eyal David: [25:10] Yes, we all want to be effective, no doubt. So really, how do you see the future of AI,

The future of AI in sales: upgrade, not replacement

Eilon Reshef: [25:17] where it's going in our areas? Let's stay at the level of salespeople, for that matter. Yes, I think it's obviously some kind of combination. Yes, there are some professions that will be completely replaced; I think that's a minority, much smaller than what people talk about. Across the street from my house there's a tailor. Really, with a sewing machine, and he's stayed too. He's older, admittedly, but we thought tailors would be replaced, and there are still tailors. But it's true there are professions that technology obviously replaces. I think in the vast majority of cases it's not replacement but an upgrade. I mean, for that matter, if today someone without Gong needs to write an email to a customer themselves, Gong writes you a first draft. It won't replace your need, I think, to write an email, to refine it; invest 5 minutes, invest 5 and not 50. Same for remembering which customer to get back to next Wednesday. In the past you'd, I don't know, open a calendar or try to remember, or you'd keep it in your head a bit. Today you have a system that tells you, listen, you haven't talked to him in a week, he asked you for an action item, and it organizes your life. So I think in sales organizations, and in programming, and in many other professions, AI is basically what will replace, I call it drudgery, the annoying and boring work that a computer can really do very easily. That probably means a salesperson will simply be able to do more. So if today, I don't know, they have a quota of a million dollars, it'll be a quota of two million dollars and half a dollar, whatever the thing is, and they'll simply get rid of the part of the work that's really, really annoying. So there'll probably be more tailors in that sense, because more people will have time.

Eyal David: [26:49] On the other hand, more problems in the world will probably get solved, right? That's basically what you're saying, people freed up to do more things.

Eilon Reshef: [26:57] Right. I think there will be, yes, you'll probably need fewer people to do the work, and people, yes, in the end there are fewer tailors today than there were, say, 200 years ago. And yes, but obviously

Eyal David: [27:09] other, equivalent professions will be created, yes, nobody knows today what they'll be. So on the way there, tell us what your organization actually looks like in terms of AI, what the product organization looks like, let's maybe focus there.

An AI-first product org: the AI group and a single driver

Eilon Reshef: [27:22] Yes. The product organization, apart from AI, looks relatively standard. I mean, each product or each product area has a manager who leads that product group. And where we're a bit unique, because we're that kind of AI-first company, is that we really have an entire group, the AI group, which also has features or capabilities it delivers independently. An example: OK, we now have a capability we deliver that takes a customer and knows how to produce a document summarizing that customer from A to Z according to the company's template. And the company says, I want first of all their business problems and the status, or I don't know, whatever, and it knows how to do that. So that's the AI group delivering on its own, because most of the work there is AI. There's the debugging and so on, but really 80 percent is AI. The second thing the AI group does is basically support: it builds infrastructure and helps other groups whose AI expertise is weak. For example, the product we talked about sends emails, so this group helps them with how I write, how AI can write an email better, and so on. So the feature still belongs to the main group, let's call it the product group, but basically the AI group supports it, and because it's a big group it manages to also build. So, A, how big is it, and B, what does such a process look like? I'm talking specifically about the hybrid process you described. The hybrid process doesn't look good, it's hard, hard. In general, interaction between people is a hard thing, yes, but we've developed that skill. Some group with, I don't remember exactly, I think product managers at the moment, and of course behind them a few dozen developers, I mean there's a big group behind the scenes. And it basically splits; in the end we're very strict that there's one driver. I mean, it's not that the driver is someone from the AI group or that the driver is from another group. So as an example, we released a capability for when companies run a process called pipeline review: the manager sits with the salesperson and they go account by account. They have methodologies, what's called MEDDIC and BANT and all kinds of acronyms, that basically say how I look at the account, what language we speak with each other. And we discovered that the AI can help you; you don't need to go hunting for how to see it. Yes, for example, one of the things you need to answer is who the decision maker is. So the AI won't tell you who the decision maker is; it'll tell you, I spoke last Tuesday with Tzvika, Tzvika said he's the decision maker, that kind of thing. So whoever led that was the same group that manages the pipeline review process, because they understand where it fits into real life: on Monday they do this, on Tuesday they do something else. Whoever developed that AI infrastructure is basically the AI group. So basically we pair one product manager, let's call it the driver, who's from, let's call it, the main product, with a second product manager from the AI group, in this case both product managers were women, and they work together until they connect the ends. That's of course hard, because the original product group doesn't really understand what's in AI, so it has all kinds of fantasies of maybe we'll do something that sings and dances. And usually that product manager coming from the AI world isn't supposed to, or can't, understand the nuances of what the business process is and what they do on Mondays and Tuesdays.

Eyal David: [30:16] So they need to develop a skill of working together, which is fun, but let's put it well. So how are you actually going to solve that, what are you steering toward to develop it?

Eilon Reshef: [30:26] To improve it? So we're building a process that's, let's call it, guidelines, that says, OK, first of all there are drivers. First of all it's clear who leads the process, so he won't say I'm waiting for her and she won't say I'm waiting for him. Second, we explain to people: OK, you now need to work together, how you do it, figure it out, and you're now working together. And third, we put in a few planning mechanisms that support it. For example, at the start of the quarter, before we do quarterly planning, we come and say, let's locate all the places that are dependent like this, and crack them, understand that we do project planning for it at the most basic level: there's capacity here, there's capacity there, it's progressing. So we as managers help that process a bit. And what about, you talked about not speaking the same language, as it were, what about that area of basically improving communication: what can I do with AI, what even is the business process, how do you... Yes, so there's the component of communication between people: you put them in a room long enough and they start speaking the same language. But in general we also do a lot of work. Once we actually built a syllabus for a kind of mini AI course, but today, say, at the GenAI level, all the product managers, and the designers too by the way, all did an exercise that tells you: do something GenAI. Take a customer call, do 1, 2, 3, 4, 5, produce a call summary, a title for the call, I don't know what, translate it into Georgian. And then they had to, first of all, I now need to get to know the console I'm using, it's not ChatGPT, it's another one, doesn't matter, and I need to understand where the data comes from. Suddenly I see it writes it with bullets and I don't want bullets. I tell it to make it short and it makes it long. So you learn the language. And the goal in the end is that a product manager won't be what's called a prompt engineer, they're not the one who needs to write the exact prompt for it, but they at least need to be able to look and say, I have an idea, can I ask the AI whether the call was good or not? Take a call, put it in, tell me what the AI said. It said fine? Does it seem in the right direction? It'll always say something. But at least you're not afraid to do the initial evaluation, just as I think every product manager knows in their head to say, if I build a screen with three buttons, it'll take a week, two, three, and not a year. Tell us more about that challenge you set: how long it was, what the participation rate was, what it looked like, and whether you felt an impact afterwards? So first of all, it's too early to know the impact; we did it recently. Participation rate, you're not asking the right question, because with me, whoever wanted a salary did it. So I think everyone did. I didn't check, but you'd have to look at the payroll. And I think everyone pretty much enjoyed it. We also have someone who's a prompt engineer, so in the end she did the, let's call it, she gave them grades, went over it, gave comments, said what was good and what wasn't. I hope, time will tell, whether they really lose the insecurity, or gain the confidence to do things on their own.

Eyal David: [33:04] Excellent. I'd like us to talk for a moment about other processes happening at your company in terms of AI. I think you're leading in that respect, and it's really worth it if you'd like to share.

Evaluating GenAI and Elo ratings

Eilon Reshef: [33:12] There are many things. I think it was really convenient for us because we came from the AI world, where GenAI wasn't some kind of, wow, let's reinvent everything, but rather let's adapt existing processes and try to understand it. So I'll give an example: I think what's called evaluation is a super critical thing. I see many startups that are like, I ran a prompt and it was fine, and then I shipped a product and it'll be good. But a customer comes and says the output isn't good; what do I do with it now? So I don't know what to do. So, for example, there's a structured evaluation process, which means that every time we release something, we measure it. Now, in classic machine learning it's easy to measure, because in the end there are metrics like precision and recall, I mean how many, doesn't matter, error rate, yes, doesn't matter, I succeeded or I didn't. In GenAI it's harder, because in the end it writes an answer; this one's correct and that one's correct, the question is which is more correct. So we built, for that matter: first, in places where there's precision and recall, or where I need to check accuracy, we sit and build a test, people sit and check and score it all the way, and each time we measure how much we improved. Automatically, that way. In GenAI places, we really compare old versus new: what's better, what's worse. We use a metric called Elo, which people use. Elo comes from the world of chess: every chess player has a rating. The world chess champion is a Norwegian guy named Carlsen, and he's the highest, and below him, not champion, he's basically just the highest-rated. And then there are gaps: he's 2800-something, and the one after him is 2800. And basically what happens is that every time there's a chess game, the winner gains points and the loser loses points, and the number of points depends on the balance of power. I mean, if he loses to me he'd lose his whole life in one go, but that won't happen. And in the AI world we use it to compare models too. I mean, basically you let people say: look at one thing, look at another thing, which seems better to you? Both answers are correct, but this one is written better, this one less well. And then the better model gains points and the lesser one loses points. And then we can simply take models and say, OK, this is a 1710 score, this is 1720. That gives us, it quantifies the distance between them. And also, if I offer the market a product, I say, OK, I improved by 13 Elo points. Now, obviously that means nothing to the customer, but 13 Elo is less than 100 Elo and less than 1,000 Elo, so if later we release 400 Elo, I'll know I really, really improved. So there's nothing clever about it; it's a math formula from about 10th grade, but it's really nice that in the end it gives a numerical score to something that isn't mathematical. Yes, and you say it doesn't matter to the customer, but still, if there's some score and it went up, right? Why would you basically... Today there's a leaderboard of models, not ours, where they give them generic things, like, I don't know, answer me an SAT question like this. So there really is a leaderboard like that in the world of all the models. To us it doesn't matter,

Eyal David: [35:51] because our models don't need to answer the SAT, they need to answer about sales, but our equivalent is exactly that. Interesting. So what does it look like inside the process? Do you define it as some KPI you need to reach, not a KPI but some place you need to get to, a bar you need to reach, otherwise the feature isn't good enough and you don't release it until you get there?

Eilon Reshef: [36:11] That's very hard. First of all, yes, but it's very hard specifically with this method. Because really, when we launched Gong, at first we used an external transcription engine. It was a catastrophe, 30%. Today it's Whisper, yes, today you can do it with Whisper. Yes, so we had a 30% error rate. Now, I could have said I want 25; it wasn't possible, in the world, the world was at 30. Yes, then it dropped to 20 and to 10, and today we're at, I don't know, an 8% error rate, Whisper at 9, doesn't matter; it's already a problem that's much easier today. So I can't say, give me 90%; the technology doesn't work, what will I do? But what we can do is, A, compare them. And really, yes, if there's a feature that, say, for that matter, is 50% accurate, which is what we discover, I won't release that feature in most of those cases, yes, if it's a feature that's not in the way. But if, in that example I gave earlier, I want to identify the decision maker and half the time it tells me nonsense, then the person will just say, leave me alone, why are you bothering me.

Eyal David: [37:02] OK, yes, that's really an easy example. There are probably more complicated examples where it's harder to understand where the difference or the value is. So what does it look like? There's launching the thing and there's exposing it to the customer, right? Are you more lenient about exposing it to specific customers, what does it look like?

Showing customers early: feature flags and betas

Eilon Reshef: [37:19] The methodology at Gong has always been that customers get access to things with what's called feature flags, I mean to immature things very, very early. I have a story about a product manager who came to me one day and told me a customer had complained about some feature, doesn't matter which feature, I don't remember. And I said to him, how did you show it to him, it's not ready. He says, I know it's not ready. So I said, what did you show? He says, the window. I told him, there's no window. He tells me, there was a window. So I said, and what did the customer say? He tried to save and save didn't work. We show customers very, very, very early, a true story. And in the AI worlds it's even stronger, because showing them Figma, which regular product managers show, or Miro or whatever, doesn't help. Because it's nice that the AI can produce an answer, but the UI isn't what's interesting; the answer is. So very, very early in the process we start bringing in, let's call it in today's jargon, prompts or results or texts or whatever, something I can basically show them in the world: a summary of this customer's last week, come look, is this good, is it not good, what do you say, long before the UI. Sometimes we literally send them spreadsheets with it so they can see whether the data is right or not. Not long ago we had a customer we showed exactly that kind of thing: a biweekly summary of what, but once every two weeks we want to know what's happening, yes. So we produced a biweekly summary for them. I listened to the call with the sales manager, and she's looking, and my product manager is sitting and showing it to her, and the sales manager says, wait, I'm missing something here. So she said, what is it, like, the product manager said, what did you miss, what's going on? So the sales manager asks, wait, is this the AE's version or the AI's version, the salesperson's or the AI's? OK, we passed that test. When you can't tell, that's excellent. Of course that's an exaggerated, anecdotal example, but we show it to them and say: is this good, is it not good, shorter, longer, do I use bullets, don't I use bullets, no, wait. In the GenAI worlds, the gen generates. Right. So you can also vary between people. My partner Amit is a CEO, he has no time, so if I give him a word, five words, he'll say: that's too good. A mid-level sales manager has more time to read. So there are also nuances here of what the use case is, who it's meant for, how much customization there is, how much ability the user or admin has to decide what it outputs. But all of this gets to customers very early, because otherwise they won't love it in the end.

Eyal David: [39:20] And that's something that might also set you apart beyond... If we compare both of them to Microsoft and Salesforce, classic, they don't have that. They can't release that early and have that conversation with customers, right? They won't send out a spreadsheet and start talking, a spreadsheet in Microsoft's case,

Eilon Reshef: [39:34] and actually start talking to the customer at such an early stage. So first of all, they can; I don't know why they don't do it. Microsoft is a bit more like that, and so I think they have somewhat better products. I also doubt Google does it. I mean, I think it's something in a company's culture. Salesforce certainly doesn't do it. I even once talked to someone there and asked him; he asked how many, whether we do betas. I told him, what do you mean betas? Each such pod runs 20 betas; right now there are probably 300-400 betas somewhere at Gong, because each PM has 13-20 betas. So he asked, what do you need betas for? I said, what do you mean, for the process we just talked about. He said, but you can see what happens after launch. I said, OK, that's an audience difference of very big players. You can, but I want the version that hits, not that version 7 hits. But they succeed, let's say, more than us in terms of size, they're bigger, so I can't say mine is good and theirs is bad, the fact that they've scored on this specific thing.

Eyal David: [40:22] Yes, I believe this is the right way in 2024-2025, but there are really many ways to succeed. And if you could now recommend to an organization how to bring in the AI methodologies you advocate more deeply, or in general how to bring AI into an organization, how would you recommend starting?

How to start bringing AI into an organization

Eilon Reshef: [40:40] I recommend starting small. I mean, coming and saying, let's find the simplest GenAI use case there is, some GPT wrapper that does something, even if it's not really important to customers, because it's hard to know at first. Come and say, let's put it in and see how customers respond. Usually the appetite comes with eating. I mean, at some moment, so why didn't we get a call summary? And then people said it's too long; we have it short too, and then we have long, both long and short. And then you say, wait, if it can do short, maybe it can do... and suddenly you discover the richness. But until you start, customers can't imagine it either. When they can imagine it, it's either because a competitor did it and it's too late, or, I don't know, it's already really become a hole in the product. So trying to iterate is really easy to do today. If I compare what we had to do three years ago, today just throw something into some GPT, see what you get, maybe you'll get something good. And it's not that expensive either. Expensive also depends on scale, but there are expensive models and cheap models, the issue of... Yes, in terms of development, certainly not. The cost issue, let's call it, of running models is part of the consideration. I mean, if you run on, I don't know, ChatGPT 4 today, it's expensive; run on ChatGPT, I don't know, the mini, and it'll be cheap, or whatever, Claude,

Eyal David: [41:45] there are many options and those also need to be tested. There's also another thing here: the sooner you start running some GenAI,

Eilon Reshef: [41:52] you'll start building a layer that's unique to you and start building your expertise, which is also something you need to start running to do. You must, must, must, because really, with all the hype where everyone says AI AI AI AI AI, fine, hype aside, but today products that don't know how to put it in the right places won't survive, yes. So it's not if, it's where. There's a product here where you can take an email and say, make it more formal, less formal. I don't know, thirty-something percent of users use it. A user like that won't move to a place where they're now looking at a blank page and saying, what am I, why am I doing this.

Eyal David: [42:22] Really. So OK, that's a great piece of advice, and maybe another one: we looked toward the product, but what you said that I really took away is actually, inside the company, introducing some challenge or some requirement tied maybe to something like people's salary. No, no, I wasn't serious. I'm joking, but it's an excellent habit for actually bringing people in. You must.

Eilon Reshef: [42:42] I find again, it's like a kind of management one-on-one, but in the end a person who isn't versed in something, doesn't feel comfortable, will always postpone it: I'm busy, we're always busy, we're all busy. So you need to inject people with some self-confidence or knowledge, artificially, so they'll come and say, wow, I'm diving in now, I'm doing something with it. Because I'm not afraid of it, and it won't take me three months, and I know how to get into it. So of course you need to inject that knowledge, and the designers at our company got the same exercise too. Because I think someone in UX also says, wait, maybe instead of a button we do something else, but if I don't know, then I say, whoa, AI AI, where's the data scientist here, where's the chief scientist, where's the AI product manager. You don't need all those things. Come on, I think even my parents know how to do it, it's not something super complicated. And UX is also going to change around this, so you must get to know it, and there's no escaping it. Absolutely. And there are also things it can do and things it can't do. So now if you go to one extreme that says, I don't do anything, that's terrible. If you come and say, OK, I read in Scientific Yedioth that it'll do everything, then you put a button, ask a question and it'll give you all the precise answers, and that can't be done either, so.

Eyal David: [43:50] You have to develop intuition to make sensible decisions. Excellent. And I'll try to squeeze one more piece of advice out of you. Today is a very hard period for product managers to break in, especially juniors, but even up to five years of experience, the market is hard in general today.

The challenge of onboarding juniors

Eilon Reshef: [44:04] What do you most recommend to get into the market, to start in a product role, what would you recommend? I have to disappoint you here, because we also haven't cracked the junior issue yet. I'm very disappointed; it's one of the things, especially one of my big failures, that we don't know how to bring in a junior today and make them succeed. The pod model we use, where a PM works with a development group, really doesn't work well with juniors, because the juniors look at it: what am I supposed to do here, a quarterly plan, what do I do with that? I think a junior PM, first of all, needs to find a mentor, some more senior mentor, and also find a constellation in the organization where they can slowly taste parts of the organization. Then they come as a more mature product person, and then they can hit the ground running, as they say, start running with some mileage. We cracked it at my place: of all the 20-something product managers, almost, almost all came to Gong as seniors. That's interesting; I actually tied it in my head to a flight simulator: you need to log flight hours before you... Yes, but someone has to pay for that simulator. And it may well be that there are companies where it's easier, I don't know, I'm just making it up now, I don't know, maybe banks, maybe Microsoft, maybe really companies with more of that kind of guidance, as opposed to companies like ours that live in a very intense market and constantly need to get things done. We haven't cracked the method. I hope that the next time we meet, I don't know, in two years or a year, I hope I'll have more insights, but so far we

Eyal David: [45:24] haven't cracked it. So I'll take it from here; I'll actually put it on the positive thing you said: it's better to take on some mentor, better to get in, maybe from another role, and start learning both the industry and the product, and that way break through. That's what we're doing now. We currently have two junior PMs, let's call them PMs, poor me, there are teachers and they went easy on them, doesn't matter, PM wannabes. Two moved within the organization; they're wannabes, but they're in a supporting role right now, and each found themselves

Eilon Reshef: [45:51] a senior product manager who sits with them and mentors them. The hope is that the senior product manager will give them tasks that are relatively small for the senior, more significant for the junior, and after they do five or six of those, they'll be able to tackle problems that are a tiny bit bigger and more significant. I don't know yet if it works, I really hope so, but it's a challenge for us. And is it in your interest? Do you have an interest in such a market at all? As an employer and a company, no; as an idealist, as someone who wants to advance the industry and the country, then certainly yes.

Eyal David: [46:24] Gorgeous, great to hear that. OK, cool, we're approaching the end of the episode, Eilon, and I want to ask: what do you wish yourself in 10 years?

Eilon Reshef: [46:32] For myself, I don't really dwell on it, you know. I wish all of us years a bit better than this past year, certainly at the level of the country and also at the level of the industry. I think, I hope at least at the industry level, Israel will keep flourishing, and for all of us, in the end we're all ships at sea. So when there's a tailwind at sea and not many waves, everyone sails.

Eyal David: [46:53] Wow, that's wonderful, I really connect with that. And at the level of Gong?

Eilon Reshef: [46:58] At the level of Gong we have only one aspiration: to keep growing, provide more value to customers, and really implement the vision we've had since day one, which is a smart, automated system that comes and helps salespeople do, not only sales, customer-facing, post-sales, presales engineers, anyone who works with customers, so that their lives become like: just do the things you're good at, which is talking to the customer, doing the fun things, and in the end creating more and more value for them. Which really connects to the AI vision you talked about earlier. Yes, yes, that's how we started in 2015, and finally the technology is getting there, which is nice. One of the reasons we started Gong, me at least, is that we saw AI starting to happen, we saw the... back then GPUs and LLMs and very, very basic things were starting. We said, OK, we'll see, we hope we'll be on time. I think in hindsight it was probably the time, but I bought NVIDIA shares back then. Wow. Unfortunately, I sold them at some point, but we definitely saw the light at the end of the tunnel. Wow, amazing, really. OK, it was very interesting and enlightening,

Eyal David: [48:00] and thank you very much, Eilon. Thank you, happy to chat. OK, bye bye. Hi friends, thanks for listening. If you found this podcast valuable, you can subscribe and follow us, of course, for more episodes on Spotify, Apple Podcasts or any other app. And of course, if you couldn't find us on some app, I'd be happy if you wrote to us. We'd love five stars on every platform, and for you to follow us so more listeners can discover us and find the podcast. You can also find previous episodes on any app or on our YouTube channel; there are links in the description. So until next time, come on, be efficient, and bye bye.