Build the Lead Scoring System Your CRM Can't | Apurva Shukla, Sumble

Actually, Jared, Jared, I'll throw the question back to you. Have you— how have you— have you seen companies do account scoring before?

Oh yeah, absolutely. And I've seen it everywhere from, you know, it's a rules-based system, very simple to understand. To, we have a whole, you know, machine learning team, right, that does pretty sophisticated math to figure out fit scores and readiness at the account and contact level. But then I think where everybody who's worked in this has seen is the The adoption by sales can be quite variable, right?

Yeah.

Sometimes there's just a skepticism like, hey, you know, they see one lead that is marked low, but they, you know, they have a feeling that it's a good lead and they, you know, they kind of discount the whole lead scoring system or vice versa, right? So I think that believability is a challenge. I think the adoption is a challenge. And then so, you know, so there's some people out there that are like, look, we don't, you know, we don't want to give a score. I mean, this is, I think, some of the, you know, some of the contractors.

Yeah.

We don't want to give a score. We don't want to sort of give an opinion, but we will surface for sales some underlying relevant data points as to whether the account is sort of in market and is aware of the problem, has the problem that our offering solves, And if they are aware of us, right? And so we'll break out 2, 3, 4 different, you know, data points around that, sort of let— leave it to the salespeople to make their own judgment from it. So I think that's a different direction. Have you seen that direction?

We definitely see that. In fact, a lot of times when people come to us, they come for this reason. Where they say, hey, like, we feel like our go-to-market org is not aligned. There is a team that did like a quarter-long project or like did like a project once a year where they took some snapshot of data and scored some accounts, loaded them in the CRM, but then they downstream feel that the go-to-market org, which is both the sales and the marketing org, does not really trust it or believe in it and they're doing something else. And oftentimes actually, like when we are approached by RevOps team, it's because there has been some lack of trust. There have been too many valuable accounts that sales team DQ'd maybe after speaking with them. So an example here is we worked with a company that really wanted to know a very accurate count of the number of SDRs at a company. That for them was a very important qualification signal. They sold a solution that helps sales team do cold calls. So they really need a sales team inside one of their prospects to be outbound. And oftentimes they had accounts that had scored very highly, but when the sales team would call on their accounts, they would realize that they don't do any cold calling.

The underlying data needs to be extremely accurate, right?

Yeah, so there was some mismatch between what RevOps was flagging, what the sales team was discovering. Somewhere the reality that the sales team was finding in their calls was not aligned with what the CRM was saying. And ultimately it kind of came down to like, it became like a data problem upstream. But usually like when people approach us, it's When something like this has happened at an account for a long time, that like it becomes an upstream data issue. Yeah. Yeah.

Well, let's, let me, let me do the sort of formal intros here, who I am, why we're running this, who we've got here as our, as our resident expert today, Apurva from Sumbul. Let me get into that and then, and then I'm gonna hand it off to Apurva to to go through some workshop material. Welcome everyone. Thank you for joining today. I'm Jared Waxman and I founded the Go-To-Market Engineer School about a year and a half ago. Before we run our formal programs, we like to do some public workshops to get this knowledge and get the discussion going with as many people as possible. This is part part of a series that we're running this, uh, this week. We have, uh, we have, uh, we've done a couple, we've got, um, a couple webinars, we've got another one, uh, tomorrow. Um, and, um, and let's see here. Um, I'll just share my screen here real quick. Uh, so Go-To-Market Engineer School, uh, we've, uh, we've, we've graduated about 265 people actually over the last year and a half through 4 cohorts. A couple hundred, uh, good companies come through it. Um, this, this cohort which starts at the end of September. We're gonna have a whole bunch of great professors and TAs. You know, some of this stuff— this stuff is moving so quickly. We want to get students, participants in front of real experts and practitioners and have back and forth discussion. So we have live sessions. We also have asynchronous work and a learning management system. And then The first 4 weeks, we're gonna go through different workflows, and you're gonna be hands-on with, with, with a, with a set of tools where you're gonna get free access during the, during a couple months of the program. And Clay, Sumble, Swan, Octave, and we're gonna use Claude throughout, of course. Or if you wanna use ChatGPT or Astra, that's okay too. And then in the, in the back half of it, you're gonna work through a capstone project and then present that. And that capstone project, you could use whatever tools you want. And registration ends next week, and then we get going by the end of the month. So that's my spiel about what it is that we're doing here at the end of the month. And Apurva, great background, was at the growth team at Rippling for a few years, managed, I think, a couple million dollars in a go-to-market data budget. knows good data in the go-to-market space very well, now leading growth at Sumble. I came across Sumble last year through a project I was doing for one of our portfolio companies, and it was the only place that we could find data to help put together our Our account list. There was no other source that had the specifics around what kind of companies were in market in our particular scenario. I was super impressed with it and been kind of tracking the progress of Sumbull ever since. So I'm gonna hand it off to Apurva, who's gonna go through lead and account scoring, and then we're gonna do a bunch of questions at the end. So take it away.

Thanks, Jared. I'll build on my background a little bit. So hi everyone. I worked at Rippling for about 3 years. Rippling, if you may know, if you know, is like in the HR space and we have a really big TAM. That was one of the privileges of working at Rippling. Rippling can sell to pretty much any company. that has, that pays its workers either hourly or through a W-2 program. That means that at Rippling, we really cared about getting to know all the companies in the universe. So like, that was one of the very, like, almost one of the things that gave the company a competitive edge. So anytime we would launch a new kind of an offering or anytime we would expand to a new country, like one of the first projects we would kick off is Can we know all the accounts? Can we know all the possible companies in this country or region? And oftentimes I would be plugged in for projects like these, and I realized that while such a project sounds simple, it's actually a very, very hard undertaking. So for example, like one of— towards the time when I was leaving Rippling, one of the projects was we were expanding our payroll into Germany. And Germany had, like, in the official government data, there were like some 13 million companies in Germany, but like we were only able to source like a handful of those. So leading that project, I kind of realized how critical it is to fully know your accounts. And this is the topic I want to touch with you all today. I think a question which is maybe a good starting point to think about is, And I would encourage if anybody feels comfortable jumping off mute and answering this is like at the moment, how comfortable or how sure do you feel in fully knowing your TAM? So how it happens is a company identifies an ICP. So maybe it's like you sell to medical device manufacturers. Do you know who are all— what's the universe of medical device manufacturers out there? Is there anyone who's done this kind of an exercise and feels, has some commentary on how they did it and how satisfied they feel with it? So since no one is talking, I certainly have done it and have been extremely disappointed, right? Whether you use a data broker or you use LinkedIn or you have Zoom info, and you try to identify the title of the person you're looking for. You get the list, and either the list isn't very accurate or you're missing, you know, your real sweet spot of your ICP. So, you know, I can relate to your German story for sure, but I'd love to understand what you've done. You know, Rippling is a massive success story. So obviously you've cracked the code, and, you know, everyone on the call is, I think, looking looking for your advice and experience to help direct how best to do it. Yeah, certainly. I definitely want to get on that. I think it's a problem of knowing all the companies you can sell to and within those, knowing all the relevant people who are your buyers. I think one thing that unlocks, by the way, and Going back to, by the way, why it was important at Replame is once you know all the companies and all the people, your total TAM, your universe of buyers you care about, it becomes kind of trivial then to ladder on other programs to it. Once you know all your accounts, you can split them. You can take the best ones and give them to the sales team to work, and the rest can be owned by marketing. That's the split we had at Replame. If you know all of them, you can You can do a variety of programs. You can email them, you can add them to a webinar invite, you can, if you know them by city, you can invite them to events. So the way we think about this at Sambal is that this is kind of like the foundational thing that every company should dedicate a good amount of time to. And once you believe that you have somewhat of a good map of your market, everything downstream becomes much easier. Usually till now, a lot of people were not able to do this because of, I think, 4 core reasons. Those reasons were, one, that it was costly, that companies did not know whether this is a $15K exercise or this is a $100K exercise. I think time is money as well, so people did not know whether this is a 3-month initiative. Do I need a team staffed on this? Is this something I could do over the weekend? Is this like a 2-week thing? There was a lot of lack of clarity around how to do it. And coming from and being like an operator that used to work with data vendors and now somebody who is at a data company, I think there is a lot of lack of transparency in this market as well. You don't know which tool you should use, how should you go about it. A lot of these tools require you to speak with the sales team and actually sign up for a plan to get access to their data. There's just generally somewhat lack of transparency in this market. I think the other reason is oftentimes companies themselves are confused. Companies don't know who their ICP is, and that's a very valid place to start from. Most companies in the early stages somewhat have some sense of who their ideal customer is. They could have those customers living in the CRM, But it's really hard to look at a dataset and being like, oh, I can identify who my ideal buyer should be or who my prospective buyer should be. And I think the third is that we'll touch upon a lot is that, and I think Todd, you hinted at this, which is that most of the times data snapshots exist across these various tools and there is no one tool that answers this. And I think that is an unsolved problem. I think at Sambull we have taken the approach of really starting first with LinkedIn data. So if you are a company that sells to buyers that exist on LinkedIn, I think Sambull is a great place to start. But there are many places for which I would also acknowledge that Sambull is not a great tool. So let's say today I was talking to somebody that sells to CPG brands that are trying to break into grocery stores. I don't know how many of those buyers are on LinkedIn, so I would not recommend somebody as a data source. Maybe somebody needs to build out like a specific data source for it. Or if you sell to like real estate agents, that again is not a buyer segment that lives in LinkedIn. And generally it's a much harder segment to track. Maybe we can do a session about those kind of markets some other time, but that's not really like, I think, what we'll find.

I think The—

in this, now before I go into the workshop component of here, I think, like I said, there are like 3 core barriers that prevent people from knowing their time. One is cost. You don't know like how much time it'll take. The other is like capacity. You don't know whether you can do it yourself, whether you need a RevOps team, or Jared, like you mentioned, you need maybe an ML team that does this. And the third team is you generally don't have like a good data source that can answer these questions. I think, can I add something to that? Yeah, yeah.

You know, I think a lot of times too, it's, you know, when people think about, oh, I'm gonna build out my TAM, you know, who's my ICP, you think industry, segment, geography, you know, maybe, okay, B2B, B2C, something like that. These are very, these are okay starting points, but they're very, very broad. Right? You end up with way too many companies and you can't just go from industry and segment to a point where you know that those companies have the problem that your offer solves, that they're looking, that they're in market or should be in market for it. So I think there's a kind of a strategy question and a lot of brainstorming needs to happen. to come up with some hypotheses around what signal out there should I look for. It has to be some publicly available signal. What signal out there is available that tells me that they're in market? I'll give you a couple of examples from some of the stuff I did. Actually, Todd, it was a company I was working with, working on with Todd on the call here. And, you know, they sell— startup is still something in the space of, you know, migrating cloud data. And so we were able with Sumbull to find companies that actually have active cloud migration projects going on. That's not just, oh, they're a technology company, they're B2B, you know, they're between 50 and 10,000 employees. I mean, that stuff's very broad, but And how did we find that they have active cloud migration projects and therefore would be potentially good customers? It was revealed in their job postings. And, and, and Sumbull extracted that out as a feature that, that I could go search for. And that, that made, you know, all the difference in terms of having a list that was going to be actually believed and valued by the sales team versus some some broad generic thing. And I'll give you just one more example, which also actually comes from hiring data, just for my own, for go-to-market engineer school. Who might want to send somebody to go-to-market engineer school or who might want to hire somebody that graduates from go-to-market engineer school? Guess what? A company that is hiring for somebody in that space or somebody with those skills. If I can see, okay, there's these 300 companies in the last 2 months and they're hiring for somebody that knows Clay and various tools in the space, depending on what their title is. If I can extract that out, I've now gone from millions of companies in the world down to the small set of companies that actually I can tell you for sure have the problem that go-to-market engineer school solves.

Right.

And now I have to figure out the right offer to reach them and I may want to I may want to take that list and separate it out further. Okay, there's these 1,000 companies, but— or 300 companies, whatever it is. But do I know some people at this company? Have they ever been, you know, are they subscribed to my newsletter? Are they, you know, do we have any graduates, you know, anything like that? So I'll take it further. But I think it's that brainstorming process in the beginning to figure out what is that public signal that I can find.

Jared, you said so many interesting things here. I think, you know, a lot of times I think like this, our industry started segmenting lists based on industries because that was readily available information on LinkedIn and it's oftentimes self-reported. But, you know, we ran into a lot of issues with it. Because oftentimes, like, you cannot really trust it. Like, for example, is Tesla a manufacturing company or is it a tech company? Is AMD or Intel a manufacturing company, a semiconductor company, or a tech company? Like, LinkedIn qualifications force a company into one of these buckets, and oftentimes it's not representative of what the company actually does. So at Rippling, we had this problem. A lot of times you wanted to sell to startups, but startups would not tag themselves as technology. If it was a biotech startup, it would tag itself as a biotech startup, or it would tag itself as a healthcare company. And in our— we excluded— we could not sell to hospitals, so we would exclude all healthcare companies and as a result miss those. So the industry filter is oftentimes a very crude filter to kind of force-fit like a very large list into something reasonable. And that maybe like 5 years ago or 10 years ago when we didn't have like cheap LLMs, when we did not have browser-based scraping, when we could not extract the full HTML of a website, we could not extract all the job posts that the company had ever published to paint like a more detailed picture of it. That was maybe like a crudely— like that was like a fine way to approximate things, but like Now we have a lot more data, and we have the means to process that data.

Yeah, something has changed, right? This would have been pretty hard 10 years ago, 5 years ago. But now it's like, imagine you have a team of, you know, 1,000 people, and they can spend, you know, a whole month on it, searching, doing all the research, searching the web, but you could actually get that done in a matter of an hour or half hour. Or maybe it's even preprocessed already. And, and, and, and, and so it's a matter of like the creativity again to figure out, okay, if I had unlimited resources to do research, what would we research? Because now we could do that with LLMs.

And the other thing I will touch upon what you said before I jump into the, the sharing my screen is, you know, I think there are 2 components to doing TAM mapping. The first is you really need to know what the universe of companies is. So Jared, in your example, it will be companies that have either already have GTM engineers in seat or have put out a job post, or have maybe not put out a directly relevant GTM engineer title, but maybe are saying something which kind of hints at this kind of a pain point. Maybe they are saying they need an applied AI person in the marketing org, right? Like, that person is still kind of solving a similar pain at the company, and all of that kind of forms your whole universe. So this has maybe, let's say we do this exercise and we come out the other end saying that there are 6,000 companies you care about. 6,000 is a lot of people. 6,000 companies is a lot of accounts. And within that, when a signal happens, that chooses its priority. That's just like the in-market, like you really need to reach it out right now. So you kind of need both. You need to create the whole map. This is the universe you care about. A lot of marketing activities are not— will not just be aimed at like people who are like very ready to buy, who have an active signal. You know, you need to nurture them, you need to educate them. So it's very important, you know, the whole TAM. And this is a good jumping point on how we at Sambal think about this. And I want to leave you all with something which is very, which is easy to do. It's like I want to leave you with something which like you could do after this. So I want to make it as interactive as possible. And but I think the thing I want to start with is we on our website have a guide on how to do account scoring and how to do people scoring. And how we have structured our guide is like we have listed out our thesis on how you should do it, on how our approach to doing it is. And by the way, we cannot do it completely for a company because there are many things which are your first party. There are many things which only you know about the business. But the way I'm gonna show you how it's done is that you can kind of build those additional first-party insights into this. The thing we have done is we have also built this into a skill. So if you, this is the URL. It lists out a skill that you can install for Claude Code or Codex or whatever you use. And what this does is it has like a set of baked-in instructions that we have written for Claude to be able to use like behind the scenes access SAML through the API. And kind of do account scoring for you. So the way to imagine this is, is that Claude is the orchestrator in the middle. It's going to access SAML data. It's going to access your HubSpot data or your Salesforce data, right? Because that's where you have your customers. So we need to kind of take your customers, match them in SAML, see what, what things are we seeing in your customer base and how can we mirror that against the rest of the SAML data. And ultimately produces an output that you can tinker with. So it's not just like, hey, this is an Excel file or a Google Sheet that, hey, here is your answer. It's going to create ultimately— I'm going to show this in a second— it's going to create an output that you can play around with. So the idea here is very simple. This is my slides. All I will do is do simple account scoring. This is my— not resume. It's going to be SAML account scoring. And what this does is basically some cloud is now reading the skill that we baked in, and it's going to ask me a bunch of questions. It's going to ask me, it's going to interview me almost. It's checking whether I have the SAML API key and the MCP setup. That's how it makes, it connects to SAML, and it's asking me who do I want to do account scoring for. So in this case, it's figured out that I am logged in with sambal.com, so it's asking me, do I want to do it for Sambal? So in your case, it'll ask you for your own company, and then it runs you through a series of questions where it interrogates you on what your ICP is. And Jared, this is kind of what the discussion that you brought up in the beginning, which is at some point, like sales, marketing, maybe the leadership team kind of needs to sit together and discuss really who are we selling to. And Claude is like an incredible interrogator when it comes to this.

Do you need to have, for this to work, do you need to have a whole lot of customers tagged with good metadata in your CRM for it to fit?

No, no. So what you— so great question. You actually don't need— the only few things you need on your customers are either the website or the LinkedIn company's page, like that LinkedIn profile. And what it does is basically takes either of those fields that are available in your CRM and it looks it up in SAML. So let's say if you have a company called Airtable in your CRM, it's going to look up airtable.com in the SAML dataset and it's going to be able to connect to the SAML dataset and figure out all the things about Airtable. So you don't need a lot of, you don't need a I'm sorry, I have one more question.

And anybody else on the call, please come off mute and ask questions. I want this to be interactive. Yeah, for— oh, go ahead. You have a question?

Go for it. Yeah, Apoorva, for early stage, I'm a fractional CMO and a lot of it is, you know, pre-PMF. So you're working off hypotheses. So I've done a lot of that. So here you're asking questions on existing customers. But what if you're early? Do you, will it make assumptions and then you go validate? Because you're doing loops of validation, right?

It's a great question, actually. Maybe Jared can also speak to this a bit. But behind the scenes, when you create an account on Sumble, what Sumble does is it builds a profile about your companies. What this means is we behind the scenes go, we scan your website, We figure out what tool you sell, we figure out what your competitors are, we figure out what your complementary tools are. So maybe you sell alongside some tool, and we take that information and we scan the entire Sumble dataset and build not just like a, we just scraped your website kind of a profile, but we build a profile against the whole Sumble dataset. It's not perfect. But basically, like, with the data we have, like, this is— that's what we, like, pull up. The other thing, Claude, that it can do is it's actually super— I think SAML, because SAML is like a rich collection of a very deep kind of an insight on each— on any account. It has, like, all the people who work, they're tagged with the right titles, the job function, the reporting hierarchy, the team structures, the the reporting hierarchy within a team, it's actually really good to test your hypotheses. So there was somebody who worked with Jared who was using the Sumble MCP to kind of like do a lot of product marketing work. They had some hypotheses and they would ask questions like, hey, I sell this tool to companies that like are doing that they need observability in their, let's say, data stack. Like, who is hiring for data engineers? Are they large companies, big companies? And they would kind of validate whatever they believed about the market with Samba data. To answer this question, I haven't, by the way, told Samba any— I haven't told Claude anything at all, but it has proposed an ICP for me over here. If you see, it's figured out what are the key personas I sell to, it's figured out what technology categories I engage with. And by the way, like, this is not just like doing it, just like I said, like through just scraping my website. It's like really doing deep research in the Sumble dataset and figuring out like where do we fit, what not do we fit, what projects are companies typically running when they hire when they hire, when they need like a sample dataset. So it's like figured these out and it's going to keep interviewing me. And this process takes a little while, but ultimately the end output you get looks a bit like this. I'm going to spend a moment here kind of dissecting what this shows you.

And just a quick follow-up question. To calculate TAM, Samsung or early relevant market, could you jump from this and in Sumbull get to the point of if the hypotheses are correct, your TAM is this, make some assumption of ACV or whatever, right? On top of that, which I have done manually, by the way, years ago, I wouldn't do that.

Yeah.

The logic is the same, but the tools are amazing.

You know, what you said is so— it's really possible. The simple answer is yes. A lot of times we work with companies who want to know answers to questions that are really impossible to get. Things like, you were working with a company that offers like an AI support bot, and they wanted to know some estimate of how many support questions a company could be getting. Another example is we are working with a lot of AI companies and they want to know an estimate of token spend or an estimate of AI spend at an account. This is like completely proprietary. All you can do is infer this. But you need a lot of primitives to infer this, right? You need the count of software engineers, You need the count of AI engineers. You need the count of how many times, let's say, Claude, Code, Replit, all of these things are mentioned at an account. You need a count of how fast those companies are growing. You need not just a count of those people, but you need a share. Is this a company where 5% of its workforce is software engineering? In that case, the token spend is going to be kind of low. Or is it like a Shopify where 70% of the employees are software engineers, in which case it's going to be like several millions, and that changes the density. So let me answer this in a second, but I will show how you can answer these questions.

Okay.

So ultimately you go through this whole process, you know, like Claude is interviewing you, asking you, hey, like these are the concentration that it's going to take. It's going to make a first pass at it. You can of course change everything. Um, I'm just going to run through it, but I want to show you what the end output kind of looks like. The end output looks like this. What you get is a dashboard that will be on your computer. It's looked at all the accounts which it thinks are your ICP. It's figured out which ones are in your CRM. So if you see here, it says there are some accounts that are allocated to a rep. There are some which are highly ranked but are white space. These are accounts that your CRM does not have at the moment. And there are some which are already customers. So it's figured out like this distribution. So you can see like the whole, like what's in your CRM, what's not in your CRM, within your CRM, what's allocated and kind of not allocated at the moment. You get all the sliders that you can pick around with. And this goes back to like kind of Jared's point earlier. You know, a lot of times account scoring was done Take all the companies, apply an industry filter, apply a growth rate filter, apply some filter, and then we have a company. That's oftentimes not the case. It could be that you have— maybe Jared is selling to GTM engineers, but GTM engineers exist in companies like OpenAI, and they also exist in 30% orgs. There is pain on both ends. If you apply a very brute-forced way of slimming down the TAM, what you're going to do is you're going to miss out a lot of companies. Which is why we find that this gap of trust between sales and RevOps. RevOps has applied this brute force filter to come down to the ICP, but sales is saying, hey, these accounts are missing out. I'm seeing inbound leads come through which are not ranked. Why is that the case? We somehow missed out on it. If you take the different approach—

How does it do it on the backend here? Like, okay, it asks you a bunch of questions.

Yes.

And then How does it go from there to this? Does it— yeah, does it have to go get new information, or is it all based on information that's already been processed?

So what it's doing is Claude is using the SAML MCP to query it and fetch that information. Ultimately, this information is saved as a CSV on your computer. So it's all running locally, but basically Claude has figured out your company, figured out your interview, figured out your ICP. If you connected your Salesforce, HubSpot, it's read your customers, and it has the ability to go query SAML. So it combines all of this and produces this. It all lives locally on your computer, and you have all the parameters. Like our Claude skill has instructed Claude to basically do something like this, right? It's figured out that SDR count is a big thing for me. So I have a parameter. I can be like, I can work, the sales and marketing team can sit together and be like, hey, actually our best accounts have a very strong, like they have like high sales headcount and actually this is, and they have like high marketing headcount. We want to give persona headcount this much weight. So Jared, in your case, this would be something like GTM engineer or applied AI engineers. Or something like that. It'll be, you want to know maybe how many teams use a tool, right? Like Jared, in your case, it'll be maybe teams using Hightouch, maybe teams using Clay, maybe teams using a lot of these other tools like Cargo and stuff. A lot of people you collaborate with. Anytime marketing teams use something like this, it's a valid account for you. And you can choose how much scoring you want to do to it. You mentioned in the beginning that SAML has this concept of projects. I like to think of projects as internal initiatives. We scan every company's job posts and can figure out what internal initiatives are happening inside a company. You can give it a certain amount of weight. There's funding information. Maybe if an account has raised funding, like maybe this is the highly valued account for you. And but you see, this is going away from a motion where somebody would just do this as a one-time exercise. And would just assign the weights and be like, this is your ICP. Now you have all the knobs to dial in your ICP, and what you can do is you can really see how your customers fare against the score you're setting up. So what's happening over here is I can see within my score range where are my customers falling. A great score is one where your customers score highly. They are the ones you have already met. So here you can kind of collaborate or like calibrate your— you can look at your customers, you can see what makes those customers, you can take those scores and apply it to the rest of the universe and find like, you know, like people say like lookalikes. A lot of times lookalikes are like based on things which you really cannot figure out. Why does this company— why was this company highlighted as a lookalike? Here you have all the knobs and whistles for it. If I go into an account, I can see exactly why this happened. This is all the information that Claude pulled from Sumbull. So for a certain company, it can figure out what the sales headcount is, how many teams use a given tool, how many funding is raised, what's the SDR headcount, marketing headcount.

And when you say it pulls this from Sumbull, Sumbull has it, I think probably A lot of—

let me show you on the—

I think a lot of folks on the webinar here, they might just be thinking Sumbull is just some other, some like regular data source that has like the same data points that everything else has. And I think it might be worthwhile just to take a minute and go through like what, like how we can get some of the stuff where, you know, like SDR headcount, like that's not something you can just get from many places, right? So there's a particular way in which Sumbull processes I think jobs data and LinkedIn data in a combined way to figure stuff out that other sources don't have. And that's why it can do these things, which is why I found it completely fascinating when I, when I came across it.

Yeah.

And, Purva, what begs the question then is, does Semble have the ability to create a custom, right, signal? Because you might, In your space, there might be 1 or 2 that nobody has. Yeah. But do you have that ability in Sembill to do that? Now, can you access it and calculate? That's a different thing, but it becomes a data problem on your side, right?

Yeah. We have some ways to make this happen because this is GTM Engineering School. I'm gonna get a little nerdy here. And I actually kind of explained how the Sembill map is created. How Sumble does this is Sumble basically has all the accounts information. Like I mentioned, it all comes from LinkedIn. We have people, we have people, you can get people data from ourselves. We have things that give us LinkedIn profiles of people. And what we have is a custom web scraper that basically goes and goes through job posts. We go through like some 200 such job posts in the world. So these things like Glassdoor, Indeed, everybody else. We get those jobs, we normalize those job posts. Sometimes they are in different, um, for different languages. Sometimes they could be for different entities. So maybe it is not Stripe, the parent entity, but maybe a smaller entity that puts a job post. We kind of normalize all of this into a big set. And once we have this, what we do is we are able to do a bunch of machine learning. So what this means is once you have a job post, you can extract a bunch of things from it. You can extract all the projects a company is doing, the tech stack, you can extract team names. Once you have team names of all of these things, you can look at who likely is maybe on this team. So let's say a job mentions that there is a platform engineering team. that uses Datadog and is doing some kind of an observability migration project. That's why they're hiring this person, right? And the person's title could be something very generic, staff engineer, right? Staff engineer doesn't tell you anything much. The gold is in the job data, the job description, but it's too vague. Like, it's too— it's like hundreds of millions of job posts that are put out. We have like a finely tuned LLM that extracts these things from a job post, but that's not where it ends. We can combine this. So now you know that a certain team has a platform engineering team. You kind of, if you get all their job posts, you know like what are they focused on, what tools do they use? 'Cause you can correlate that, like I said, with the people data. You now have a list of all, let's say 15 people who are on the platform engineering team. And because now when you have those 15 people, you can kind of create a hierarchy of who likely reports to whom. Claude, you asked me a question, which is, can we do custom stuff? The answer to that is almost yes. We have a new API endpoint we're about to release at the moment, which will allow you to ask custom things. So like, for example, we are working with one company in beta right now. They want to know all, they have a very unique use case. They have found product-market fit when pharma companies make international cross-border payments after they're finished with clinical trials. This is something they really cannot get anywhere else. How do you find companies that are hiring for accountants or controllers with this very specific mandate to do something like this? And now that we have some of the fundamentals, we have all the job posts ever, we can build, we can store this in a new kind of a database and allow you to query it and get custom stuff like this out. It's not yet out. I think it'll be out in a few weeks. But once you have the right foundations, you can do something like this. So this is—

Yeah, this is really, this is a good diagram. I really like this. And I just, and I don't want to bring you too far afield from the leave scoring, but I just wanted to underscore the fact that like, Yeah, there are a lot of places where you can go and be like, okay, give me a list of job posts that have this keyword in it and it will produce a list or it'll produce a, you know, you get it as a web book.

Yeah.

Okay, that's fine. That's not, that's like elementary school kind of stuff. What they're doing at Sumble is, like you mentioned, like with the machine learning, they're actually extracting intelligence from that, not just the job posts, but how organizations are structured and who might be working on different projects or in different teams and putting together the, so, So it's taking that to a whole new level, which is why they can do some of the things that they're doing and that he's showing you here in the lead score and the account scoring project is because they're extracting and combining information in ways that you just don't, you just, you just don't see anywhere else.

So let me, let me share an example on this, right? Jared, you mentioned SDR headcount is like an important, let's say like Samba sells to sales and marketing orgs. So, for us, SDR headcount is a big thing. What we can do is, like, we don't just say that there are 66 people who work at a company that have the SDR title. A lot of times what you get is a very hardcoded number. There is no way to inspect that number. We believe that the work of creating this kind of a unified dataset that cuts across all of these things is that it produces a living map of a company. So you shouldn't just say that there are 66 SDRs. If I want to click on it, I can like go here and let's say click on it and actually see how many show up. By the way, this demo was made a little while ago, so I think numbers have changed. But if I am saying there are 48 SDRs, these are the 48 SDRs. So Jared, let's say you want to sell to like marketing folks at an account, or you want to sell to like GTM engineers. You don't just want to know that an account has 4 GTM engineers. You want to know who those 4 are, and you want to know who they likely report to. GTM engineers don't exist in a vacuum. They likely exist in the RevOps org or the marketing org, whatever. Different companies, different things. We have taken the stance that they exist in the RevOps org. So if I go here, like, its job functions are classified by a certain hierarchy. So we take all of this people data and we kind of force it into a hierarchy structure. So, you can kind of like map the account structure. You have the AEs, you have the SDRs, they all roll up to sales. And all of this, like the way you can slice and dice this data allows you to do very good lead score and account scoring.

Wow.

Coming back to one of the last few things I want to touch before we kind of ask questions, because I'm sure there are. If you do this exercise, what will you get? You will get a list of your TAM. You'll not just get like a static list, you'll be able to go and like move the sliders. And if you want to see like any, any, you want to inspect any data, it's going to leave you with links that you can click to kind of go and see the data inside SAML. Once you have all your accounts, that's usually not enough. The next step of this is going and finding all the people at them and scoring them. That's the next step after this. Let me show you what the best version of this looks like, which is what we have built internally at Sambal. So at Sambal, we built this internal dashboard, and it's a super powerful way to kind of visualize what we are talking about. These are all our accounts that we care about. We have some 5,500 target accounts, 5,800 target accounts in our database. For each one, We can see what the size is, the growth is, how PLG, how much first-party PLG data we have at them, how AI-powered them are concentrated. This is all coming from SAML data, by the way. So you can set something like this up. The Claude skill will guide you. You can click inside a company and see all the people at those companies as well. And this has been scored. So there is a score here. If I click on this score, I know that the SVP of Revenue Strategy and Operations is the most valuable person at Snowflake for Sambal. And the reason is because they're in RevOps and they're an SVP. It's a target high account for us. It's one of the top accounts for us. They're in the US. And because of these reasons, they're super highly scored. And once you create this map of all the accounts you care about and all the people that you care about them, you start like almost going from like living in the dark to like switching on the lights. And you can do a variety of things then. You can do something like this. So this is our universe of people we want to know. And I can do things like I want to find all no customers who have never used, uh, who maybe have used Sambal Who we have never had any meetings with, who are a first-degree connection, or on what claimer, who are a first-degree connection to, let's say, my co-founders and have not attended any prior event by us. So this is the list of all the people that we want to reach out to. I want to find all the— let's say we are doing a dinner in Austin, right? So let me go and see how many such people are in Austin. So this is all the people who you want to reach out to in Austin. And if my— if my— we have an events person who's based in Austin, and if she wanted to do an event, this is the list she'll reach out to. You see, like, we went from not really knowing our ICP to like coming up with some version of our ICP using Claude to map our TAM, using Claude to find all the people, and then storing this dataset. You may ask, why is this different? I could do this in a variety of other tools. One, you don't have this richness. And the second is you want to create an evergreen connection with Sumble. So if somebody moves from one company to the other, you can see, you can get all of this information from Sumble and you can combine this with your first party data. A lot of this used to be—

We have a couple questions in the chat about coverage in Europe and coverage in Asia, with the note that the persona is usually not on LinkedIn for Europe.

So yeah, I would say those people are— I would actually turn you to other data providers for it. Sambal is really good because we get a lot of our information from job post data and LinkedIn data. So if you have— if your buyers exist online, are on LinkedIn, and like maybe post jobs, you will find Sambal to be very rich. But if you're going out to companies that are not on LinkedIn, like, um, you will not find Sambal to be helpful. However, if you are reaching out to companies in, let's say, Europe that have a LinkedIn presence, or like let's say we have like We have tons of customers who use us for prospecting in Europe. Surely works, but it only works because those buyers exist on LinkedIn. Same thing for Canada as well. If your Canadian buyers are online, you'll find it very helpful. The thing I will say is that Sambal, unlike a lot of data vendors, has a very open philosophy. You can go create an account. We have a very generous free tier, and you can connect Sambal. We are available in Claude and OpenAI, so you can just go to those marketplaces and add Sambal as a connector and just ask it questions and see whether it's able to satisfactorily paint you, like, how well is it able to describe the accounts you care about. You can ask it, you can be very bold, you can ask it things like, hey, I want to sell my product to companies in Canada that are less than 1,000 employees, And my buyer is somebody, a company that has a data engineering team with less than 4 people, with more than 4 people but less than 10 people. And maybe the buyer has had these kind of like journeys. So as you can see, it's labeled to query the sample dataset in a variety of ways. Recently, yesterday, I met a sales AE who sells data visualization tools to data teams. And he wanted to know all companies in Boston that had hired a data leader in the last 4 months and had put out job posts mentioning dbt and Grafana, which are like data observability tools. And he found a list of 18 companies that were in his track, and he was able to go and deep research any of those companies. So, So if your buyers are in LinkedIn, you'll find someone to be very helpful. I'm looking at a question here which is, I have a client in Cayman Islands that sells to 25 Caribbean islands. I think, yeah, in that case, you know, in that case you have to do a lot of custom stuff. And actually I would say like Clay is an incredible tool for those kind of things. But yeah, that's not the ICP that oftentimes finds SAML to be helpful. Jared, any last questions, Jared? Not from me.

Well, I do have more questions, but we don't have a lot more time. But if this is interesting, Uh, for you guys, we're going to be going through this in more detail and giving people access to Sumble, uh, and actually turning this into like exercises and, and, you know, workable projects, uh, in the course. So, um, so, so definitely, uh, consider joining if you, uh, if this is interesting to you. Apurva is going to be, going to be teaching that module.

A bit of a question and comment, I guess. Um, I work with GTM Partners, Sandgram. We're all GTM advisors and we carry an amazing, you know, go-to-market operating system as a, I'd say, as a methodology to bring all the departments together. I'm a product marketer by trade, so it makes total sense, the foundation. What I think this can do is combine the framework and actually use it, use Sembull to activate for a client, not take a month, 2 months in terms of data and having lots of meetings. We can come in with hypotheses into meeting with their data. I think that's exciting. Just, just want to say that.

Yeah, you should definitely use that. And the MCP is an incredible way to do it. MCP behind the scenes, the MCP has the ability to query SAML. So as you can imagine, SAML is this like complex dataset that you can like slice and dice in many ways. So if you have a hypothesis or if you have a product marketing framework, you can actually add a lot of rich real-world insights to that framework to really build it out.

And some of us, some of us work with different ICPs of, you know, product market fit.

Yeah.

PMF and then scale up. Some are coming. So I'm in a group with 160 GTM advisors. So I'll follow up with you offline because I'd like to match that framework to assemble, have a methodology Yeah. And I'm not the only one that's kind of struggling with that or has that problem to solve.

I think Claude is an incredible friend. Claude can connect to your framework and, but Claude by itself can only do like very basic Google searches. So it's really like falls short building this framework out. You need to connect it to a dataset. And I think that's where Sembull can play in.

And Jared, for school, engineering school, I read who it's for. People like myself that deal with multiple clients, would you You know, at times we would have external resources come in, but also how valuable can it be to not go into the most complex, because we talked about that yesterday, but have at least a good starting model that if you want to execute for the client. I believe now a senior person can do a lot more execution with the right tools. So how does the school— really, it's a question for myself, right?

Right.

Is it something that's relevant for me? To up, you know, I'm not 25 anymore. The strategy is simple for me. I've done it. But the execution, you can go down the rabbit hole a lot, right?

With Claude and ChatGPT, obviously, the execution is getting a lot easier. But what's not getting much easier is actually just knowing what is all available out there, when to use it, and basically what to tell Claude to go build for you. and then what MCPs you need and what good looks like. I think we go through that. A lot of times nowadays, you should be familiar with the tools. It's okay if you're just a beginner level in the UI, but you need to know what the capabilities are and what makes each platform or technology unique so that you could say, hey, Claude, this is what I'm trying to do, help me through it, go connect to this, that, and the other thing. I know now because I've seen from Apurva what good account and lead scoring looks like. I know what my goal is. If I know what my goal is, then I could have, you know, Claude or ChatGPT to help me through it. So we're gonna, you know, we go through kind of what good looks like. We've got live instruction every week. We've got coaches that you can meet with that will help you through either on a tool basis or really more at for your capstone project, what you're building during the program. You can bounce ideas off of them. These are other practitioners, other folks doing this for many clients day in and day out.

So, yeah, because some clients have their tool, the methodology that you jump in, they have their resources. Early stage, you come in and really you want to install systems and the value of a consult is not time anymore. It's really building and leaving value behind that you run for them or they take over.

Right.

So if that— if you're addressing that in GTM engineering school, that's super interesting to do a fast upscale, at least of a starting model that you can deploy. And then over time, things can evolve, but at least having a good starting point.

Okay.

Yeah, yeah. Reach out with more questions. All right. I think we are at time. Thank you, everybody, for coming. This will be recorded, and we will send out the recording. So Uh, thank you everybody. Take care, Purva. Bye, Lisa.

Thanks, guys.