Reps Work Around Bad Data. Agents Can't. How We Used Data Science

called Sumble, like basically the most interesting and probably best company you hadn't heard of yet. And when I learned more about it, um, I quickly learned that they have some of the best companies using this. So folks like Snowflake, Databricks, we just talked, um, Sam and I from Monaco just talked to last week. He did the CRO Confidential that'll be coming out soon. They're on Sumble. Figma's on it. Figma was just in the pricing teardown that Ulrich did from Willingness to Pay. Uh, Wiz is on it, you know, um, their CMO Roz has been on our— on SaaStr many a time. Um, and so I was just surprised that I hadn't heard of these guys yet and interesting to see what they're doing. I've gotten to know them a little bit more since then. And we've actually tested it ourselves here at SaaStr. So I talked a little bit about this morning on the kickoff, but just to give you guys some quick background and then I'll let Anthony take it over, is we use Sumble in a few different ways. You guys saw it this morning on our prospector. So we use it when somebody comes in inbound, outbound, renewals, just even list enrichment. We use Sumble now as a main skill to help us enrich and find not only the right contacts, but also the company hierarchy. So things like, you know, when somebody inbounds to our website, who do they report to? Is this the right decision maker? How do we pull for us? We wanna know how many people are on their go-to-market team, 'cause that might affect how many people they bring to their booth on site at SaaStr. And so there's all this really great data in their view, and they are doing this every day. This is where I think your data, Anthony, wins on a lot of others. You guys are constantly updating the data. From what I've seen from other tools we've tried and evaluated in the past, yours is the most consistently up to date. And the nice thing too, just to the point of agent-led growth that we chatted a little bit about this morning, is my agent likes it. So across 10K, and also Claude, which I use a lot for enrichment, I actually have a Claude skill now that chains together ZoomInfo, Sumbull, Clay. It will, between those 3 and relying on, on Sumbull as the first pass, it will find basically anybody on planet Earth we're trying to find for outbound renewals, POCs. And so it's nice that our agents have gotten a lot of use out of this. That's actually the main way we use it. I use it headless, if you guys like to call it that. Our agents like to use it. But here to explain more about it in a way that I cannot is Anthony, to explain about how agents can do this in ways maybe beyond how SaaStr is using it.

All right, awesome. Thanks, Amelia. So I'm going to start off maybe just briefly to introduce myself. So I do not come from a go-to-market background. I had, for those of you who have been in the data science machine learning world, I had previously started a company called Kaggle, which is a large AI machine learning community. It sold to Google in 2017, and I'm proud of this, 9 years after we sold it, it still runs under its own brand as a division of DeepMind. So I come from a very strong data background and relatively newer to go-to-market. But as Amelia said, we've gotten up to speed on go-to-market by working with many of the world's leading go-to-market teams. Amelia mentioned a few of these established companies like Databricks, Snowflake, Figma, Wiz, but then a lot of the very AI hot AI companies, Base10, Vercel, Modal, CodeRabbit, and so forth. And so we work very closely with them to sort of understand in an agentic-first go-to-market motion, what are their data needs. And a perspective we have come to is that in the first, as AI is relatively new and people are trying to figure out how to use it for go-to-market, The first wave was very much focused on scaling outreach. AI SDR was a very hot category initially. And what has happened is a lot of digital channels have been really saturated because it is possible to send out messages at a crazy scale now. So what we in some ways are flipping the script a little bit and using AI to do something quite different, which is to identify your targets with great precision. Now I'm going to just explain to you what I mean by precision. Precision really has 2 components to it. One is granularity. So, you know, we can now understand what's happening inside accounts that you're trying to prospect with much more detail than you ever could before. And I'm going to illustrate this with one single job post. This is a job post from Walmart. Now look how much we learn from this job post. We learn that there is a team called the Catalog Engineering Team at Walmart. You can sort of infer from the way the two— there's also another team called Walmart Global Tech, and you can sort of infer from the way the two are mentioned relative to each other that the Catalog Engineering Team probably reports into Walmart Global Tech. We know that they're responsible for maintaining an enterprise-scale catalog platform. We know that they are currently doing a migration to a new platform. They have a scalable catalog orchestration service, Looks like they're currently migrating that. We can infer that their new platform is likely to have NoSQL components to it. Cosmos DB and Cassandra are mentioned, and probably a containerization component, so Docker and Kubernetes. Then finally, we can also make a pretty good guess as to who the people who might be leading this project are. The job post has a whole lot of clues. What's the location? What's the title? What are the types of technologies mentioned? What is the team name? To what extent do those correspond with characteristics that we, you know, of different people inside Walmart? Now remember, this is one job post. Now imagine this, we take every single job post that Walmart's ever put out, and then you also look at their financial filings, their earnings transcripts, random blog posts that engineers at Walmart put out, podcast interviews that Walmart employees put out. And you can sort of imagine how, you know, look how much we learned off one job post. What happens when you sort of aggregate up every single bit of information out in the public domain about Walmart? You can really build a very rich picture of what's happening. So that's one component of precision. The second component is granularity. The second component of precision is A little bit less sexy, but arguably as important, if not more important, which is it's also really important to be accurate. Now, you've probably all heard complaints about your go-to-market vendor having stale data. It's out of date, it's not fresh, it's inaccurate, etc. It's historically been extremely difficult to have data kept fresh and accurate and up to date. And when you look across data vendors, different industries, like some of the those that are best regarded like Bloomberg have historically had 9,000 people in low-cost geographies cleaning data all day long. Well, we now have AI agents. And so what was very hard to do previously is now much easier to do. And I also want to give you a bit of a picture as to why, what makes it hard and why you so often will hear complaints about your data being inaccurate. I'm going to take a simple example. So we have a few companies that care a lot about very accurately measuring how many people are in a particular job function. So Nooks, for instance, sells a power dialer. They want a very accurate count of the SDRs in each company. Figma sell design software. They want to have a very accurate count of the number of designers and specifically UX designers, not chip designers or fashion designers or jewelry designers or CAD designers. And then Weights and Biases care a lot about the count of machine learning engineers in a company, right? And historically, that's been very hard to do. And the reason being is if you have 80% accurate company data, and then you had 80% accurate people-to-company data, and then you have 80% fresh people records, and then your job function classification, i.e., your ability to distinguish a UX designer from a chip designer, is only 80% accurate. And then your job level classification, let's say you care about ICs because ICs are the ones who will consume seats, is 80% accurate. These errors compound, and at the end of this chain, you only have 33% accurate data. Now, with agents, it's possible to push— we are, by the way, are not perfect, but we're trying to get to 99% across all these dimensions. And then at the end of that chain, we have 95% accurate data. So those are the 2 characteristics of precision. And so when you have very precise data, you can start to imagine the things that it unlocks. Marketing can target extremely accurately and with great Also with great context on what a team might be, what people in a team might be doing. Same for sales reps. Now, but where do you get started? So you have these new components, this very precise data. How do you put this into action? Now, the first 2 primitives that I always think that a company, they get hold of very precise data. What do you do with it? The first 2 primitives that I think are worth putting in place might be surprising to you. One is very granular, daily refreshing account scoring. And the other one is, I think for a lot of companies, it's a newer concept called contact scoring. So on account scoring, probably a lot of you are saying, okay, well, I've been doing account scoring for years. We do it before territory planning season. But account scoring, it's now possible to do it not just on basic characteristics like employee counts and and what industry they're in and revenue and maybe some basic technographic data. But it's possible to do it, and I'll give you a demo of this, in updating on a daily basis with much more granular characteristics that also give reps sort of a way to enter the account and sort of good context on the account. And then contact scoring, it's good to know what account is good, but you ultimately have to find the right people. And it's now possible, as I sort of indicated earlier, to have much more context on what people in a company are doing. So I'm going to give you a sense for what this looks like in our platform. So if you log into Sumble, you might see if we've already mapped out your domain or your company, you might start off with a a good first pass at an account score. So you can see here, these are some of the accounts that Sumble, at least within the list I have up, that Sumble might be most interested in. So when I go into OpenAI, which is the 3rd-ranked account on this list, I can see we break out why OpenAI is interesting between 3 characteristics: size, growth and momentum, and then concentration. So size, how large is the opportunity? Growth, how large How fast is the opportunity growing? So what will the size be tomorrow? The other thing about fast-growing companies is very often processes are breaking, so it can be easier to make a sale. And in concentration, how central am I to what they do? Now, when we go in here, you can see we fingerprint headcount. How many GTM engineers do they have? How many salespeople? How many marketing people? How many RevOps people? How many SDRs? And notice these are all linked, and I'll show you this in a minute. Sumble also likes to see whether they have things like Clay or Cursor or Claude Code in their go-to-market team, or Codex, obviously, for OpenAI. Are they AI-forward? If they are, they're more likely to be interested in us. And then obviously we care, do they do GTM enrichment? Now this is undersized. This is telling me there are 6 teams at OpenAI that use some sort of GTM enrichment tool. And by the way, Going back to this idea of precision, you can click through here and the rep can see exactly which teams. You've got the go-to-market team, then underneath that you have nested something else. We can see that there's a bit of close and potentially LinkedIn Sales Navigator. You can see the teams underneath this team and you can keep going through to see the people on these teams and so on and so forth. The account score isn't just a ranking, but it's in some ways like an entry point for a rep on understanding the account. Same goes for if I want to see all the GTM engineers at OpenAI, I click here and I see every single one of them. And then just quickly, growth, we can see their sales headcount has grown 395% over the last year, marketing 269%. And then you'll notice for OpenAI, concentration did not contribute many points because they're not really a heavily sales-led company. Only 5.4% of their headcount is sales reps. So if I got a meeting with Sam Altman or Greg Brockman, their sales tooling is not the thing that's top of mind for them because it's not a heavily sales-related company. That weighs a little bit on their score here. And then within any account like OpenAI, we also have— and we're running out of time, so I'm not going to go into great detail on this, We also have this concept of contact scoring where you can sort of— we rank all the people we most want to know at OpenAI. So again, from a rep perspective, this is super valuable because you immediately have a list of the people you might most want to get in front of. And similarly, from a marketing standpoint, if you're queuing up campaigns, you know who to queue them up with. And our scoring goes across companies as well, right? So this is a list of out of out of the Sumble top 1,000, the people in the world that I most want to meet, right? And yeah, so ultimately, if you can very precisely identify the companies you want to target and the people you want to target, that lifts up a lot of other pieces of your go-to-market system. And just to give you a little bit of a taste for that, obviously reps know exactly which accounts and why and what their entry points might be into those accounts. I gave you a bit of a sense for that. Inbound leads, there's this issue of MQL to SQL conversion, but this is an example of a Slack alert in our channel. We can see that the head of revenue operations started a PLG session. They're the 102nd ranked account. This is the 6th most important person in that company. And so Leah, who's the AE on this, immediately knows that this is a highly ranked account and a highly ranked person in that account. This is not a lead that she's going to drop. Whereas if it was the 10,000-ranked account and it was the 400th-ranked person in that account, Leah might conclude, I'm busy, I have other things on, this is not the thing I need to jump on. And then of course, the marketing use cases. If I have a list of the 50,000 people or 10,000 people in the world I most want to know and I'm hosting a dinner in New York City, I know who to invite. If I have a budget of 10,000 people to do contact-level advertising on social platforms to, I know who should be in that 10,000-person list. You know, one of the most intriguing that we're seeing a bunch is, um, let's say I know who the top 10 people I most want to, um, meet are. I can even monitor for events that they're speaking at. So let's say I, I see the 5th-ranked person is speaking at an event in Phoenix. I might send them an email saying, hey, I noticed we're both going to be at this conference in Phoenix. Would you, uh, would you like to meet me? And if the answer is yes, you quickly book a trip and a ticket to that conference. Lots of clever ways if you can know with great detail who you want to get to. A few examples of how this ends up mattering a lot. I mentioned Nooks earlier. We started off with helping them with their territory assignments and just having a very basic use case, very accurate count of SDRs allowed them to find 700 accounts that they didn't know about before that allowed them to double the size of their enterprise team. And that's just from that one use case, which is the territory planning use case. And that's not even layering on all the, you know, the other pieces, the advertising, the MQL to SQL conversion and so forth. And I might hand over to Amelia, who I know has been using Sumble more on the contact side for, you know, precision enrichment on her contacts. And so, and so yeah, Amelia, I'll pass to you and let you demo that use case.

Yeah, let me show how I'm using it.

And I should stop screen sharing at this point, I think, right?

Yeah, cool. Super cool. All right, sure. All right, I showed this, a version of this a little bit earlier, but for context, so we have what I call Prospector, which brings in all of our inbound, outbound support. Into one place. And so for inbound, very similar to how Anthony has his Slack alerts configured, I don't, I don't like to be bombarded by Slack alerts, so I have it in a queue here with 10K versus a Slack. But this is a really that came in. This is Bamboo HR, their name folks know. But, um, you can see here when I, when they come in inbound, I can hit this generate pitch and this is what it does. So it'll look up our first-party data first. So What do we know already about this company, their contacts, and then it'll fill it in with third-party signals. So in the case of BambooHR, they were a sponsor now more than a few years ago. So they're, they're a sponsor a long time ago. High chance that this person who inbounded, it's probably not the same person who was there back then. You know, what did they, what did they sponsor? This gives me at a glance what they did. So they've worked with us a little bit in the past, but also like what events have they been So for us, this is maybe more important to see. They actually have not engaged with SaaStr in, in a few years, right? So they came, they sponsored, but we either lost contact with them or whatever it might be that we didn't get them back as a, as a customer and sponsor. And we haven't really done a whole bunch with them since. There's nobody in their company who's like reading the newsletter. Sometimes we'll find like, you know, the CEOs are like a super fan and they are subscribed to the newsletter or the podcast or whatever it might be. So. We don't have a ton of first-party data. This is one where it's like, I would say a little bit lighter, even though there are some signals, not zero. There's not a ton here, right? Like all we know is they were a customer of ours more than 5 years ago now. That's not a heck of a lot to go off of for somebody who's just inbounded, who probably was none of these people. So I can see that at a glance, these folks who came back in 2022 was not the same person who inbounded. So. The next thing you do, we've got 3 things hooked up. So Exler, which I talked a little bit about, has, you know, just, just basic company information. A lot of agents use this. You can see what they're posting about. I think sometimes it's interesting just to see like, do they have any announcements? Obviously they're focused on HR, so that's gonna be the core. Is there anything, anything else interesting here I can see? Okay, then I'll look up Sumble. So I've already hit the button, so it'll preload and then it will give me a few things here that I think are helpful above and beyond something just like the, high-level learnings that we get from something like an EXA or even some of our first-party data. So one, how big are they, right? So 1,700 employees is pretty big. They probably have people, says to me, they probably have folks that are distinctly owning field marketing as a subset of marketing versus sometimes at smaller companies, this is still either the CEO, the CMO, if they have a CMO, and then how many people might there be on the marketing team to engage with us? And do they maybe have a separate media team that we can engage with? Engage on the media side. So already I'm getting a good taste and flavor for them. Um, where are they based, right? Sometimes it's not the US. A lot of our event attendees are based in North America. Their website, their company profile, you know, I can always click in if I want to see more. But again, I let my agent use it headless. Like, I don't have to log in to see this data. My agent just brings it in here for me, so I don't have to click in unless I want to. And then, yeah, we have ours configured custom just to show within Sumble. Our agent will go and look who's in their leadership and go-to-market team, because that's what I want to know the most. Who's, who's the CEO? Who's the CMO? Do they have a CRO? So they don't have a CRO, but they've got, you know, a CMO. I can look up their email quite easily here. I'll usually use that when I find if the, you know, if the inbound person is being unresponsive. Or if we have a first— great example is we have a first call with them and then maybe they go a little bit dark. So instead of me going to ask David like, hey, who do we think their boss is that we can get ahold of to help escalate and push the deal forward? What do we know about the CEO? Like, are they a super fan? I have all that information at a glance. So if anything ever happens where maybe the deal stalls out, um, and you can use it also to inform the first conversation, you could say, hey, you know, These folks have not spoken at SaaStr, but sometimes we'll know, um, a CEO or CMO who has spoken at SaaStr or even attended SaaStr in the past. And so that's good data for us to have. Um, we can also say, okay, if this deal falls out, we should probably talk to Adam because he's the CMO. He'll have some sort of pull on getting the deal done to do an event. So already I don't have to like ask David for those answers, who's one of our reps. I just have that data right away. It's also going to give me, and again, we have it configured this way. We want to know who's in marketing, right? Because we want to know. Who are all the potential stakeholders when it comes to making a decision on the event? So we've got the top 5 marketing people here. So they've got— it also tells you a lot about their structure, right? So they've got a CMO, director of demand gen, they have some international marketing folks. Hannah, who's the person who inbounded to us, so we can see that it is, that it is the right POC because it's the director of events. I can look at her LinkedIn if I want to see what she's posting about, but this is giving me really great information and that they do have different marketing departments. And they, they do have a fairly bigger go-to-market team, if you will. There's 433 people. There's a lot more sales than there is marketing. And so that might be something we also bring into the deal to say, okay, we know they have a lot of sellers. This is something where SaaStr is unique in that we don't, we don't typically recommend that our sponsors actually fill their entire booth with just AEs and sellers. It's something where we try and push for go-to-market leaders. Um, as high-level leaders as we can. And so this might be something we say to them too, maybe more so if they become a customer of ours, like, hey, we know you guys have a big sales team. Let's be strategic together on who from sales out of these 357 people you should be bringing to SaaStr, right? Maybe not the new AEs who started, maybe not, maybe one international AE. We always recommend that, like, it's not all North America. So sometimes we have people who bring only North America AEs and then When those folks come from Europe or Canada, they don't have a specific AE they can talk to. So that's another great insight for us here into, you know, how do we approach this customer, just not only at the start but also throughout the entire deal cycle. So for us, this is really great information to have. As you can see, our agent just looks it up for us, which is quite nice, and then we can use this to inform our next steps and action items. So With that, I'll pass it back to you, Anthony. Oh, you're muted.

Fun. I was just interested in looking at the structure of their marketing team. So of the 76, they have 7 people in events. And then you can see immediately that Hannah, who inbounded to you, is the most senior person in events. So that could be another interesting one to pull in. It's like size of events team, potentially. Anyway, just to finish this off, one really nice thing, ultimately, a go-to-market team has to know who are you selling to, both at the company level and the contact level. And if you have a very precise list of those things, then marketing can warm up those accounts, rev-ups can make sure those accounts, that the handoff between marketing and sales is really clean and that sales has context when inbound leads come in on how good they are, as well as allocating territories around the best accounts. And then reps have great clarity on who they want to go out to. So I really feel very strongly if you can get your— if you really land account scoring and contact scoring, it lifts up a lot of the rest of your go-to-market organization. And with that, and keeps everyone aligned, with that, we can open up to Q&A.

Yeah, I know we don't have a ton of time for questions. Let me try and ask one from the livestream just before we run out of time here. On the point of, you talked about, you know, and you just touched on a little bit, like reps working around bad data, the agents can't really do that. Do you have a specific example in mind of when, you know, a human maybe handed a rep bad data or missing data? And then how did a rep actually handle that miscommunication?

Yeah, I mean, you know, historically, a human would— one reason I think it's particularly important to have very clean data in a world of agents is agents don't have the same judgment that a human has, right? So as an example, I'll just give you a really trivial example. If you use most waterfall enrichment vendors, and you enrich somebody whose name is Sam at Anthropic, and maybe they enjoyed a week ago, a very high fraction of them will return sam@anthropic.com as the person's email address. If there are 30 other Sams at Anthropic and the person joined a week ago, he probably doesn't have that email address. And a human will look at that and say, oh, that's probably a bad enrichment, whereas an agent will probably just fire off an email to that Sam, assuming that email address is right. Right. And so something we do in that situation is we will not allow, you know, we will never return sam@anthropic.com as an email address for somebody who joined a week ago when there are 30 other Sams. And that's just a sort of small example of a place where historically a human might have used judgment to say that email address is implausible, where an agent wouldn't. And there's, you know, I give that as an example, but it's illustrative, but it sort of gives you a sense for why I think clean data is all the more important. In a world where, you know, agents are doing a lot of the outreach.

Yep. I know we're running out of time. So for folks that have more questions or want to learn more about Sumble now that they've heard of you guys, now where's the best way for them to reach you?

Yeah, I mean, my email address is a@sumble.com, just the letter A. So you know, send me an email and I'll either respond myself or set you up with somebody on our team. Who can, uh, who can engage with you.

Yeah. Awesome. Thanks for doing this, Anthony. We've enjoyed, not only me, but my agents have enjoyed using Sembull for the last like 6 to 8 weeks. And yeah, it's funny to, um, I think, uh, Ulrich was making this point on how like usage spikes. Like first I was like, okay, how am I like, let me figure out the best way I'm going to use this with our agents, right? Incorporated because we have so many agents already. And then when we were chatting last week, you're like, yeah, you have this nice Nice. Drumbeat of just usage in the platform. I'm like, yeah, mostly headless because I don't, I don't call it up. But you can see we've just built it, wired it into, you know, our inbound and outbound flows directly into our agent. Not to say that you still can't log in if you need to see something one-off. And then I have it just as a Claude connector as a skill. So it's quite nice to just let my agents run, go for it, find me the right person, find what we need to see and and get moving to the next account. So

yeah, by the way, I should, I should have said that we, it should be used headlessly. The purpose of the UI is really just to, as a catalog of what we have. Um, yeah. Yeah. Cool. Thanks, Amelia.

Yeah. Thanks for doing this, Anthony. Thanks everyone. We'll see you at the last session next. Bye everyone.