Anders Krohn on this episode
Anders Krohn is the founder and CEO of Kernel, a company that builds entity data infrastructure for go-to-market organizations. Before starting Kernel, Krohn spent years deploying AI into enterprise revenue teams and kept hitting the same wall: the data underneath was too broken for any of it to work. Rather than continuing to build AI agents, he pivoted to solving what he discovered was the root problem.
The core finding is stark: 86% of CRM records have some kind of entity data error at the structural level. These aren't missing fields or stale phone numbers. They're fundamental mistakes about which entity is which, accounts mapped to wrong parent companies, subsidiaries treated as headquarters, local offices confused with global entities because they share the same domain. Every enrichment tool, territory plan, and AI agent built on top of that corrupted foundation inherits those errors downstream.
The reframe here is critical: AI didn't create this problem, it just made it visible. Before, large teams of humans worked around data inaccuracies so consistently that organizations never realized how broken the foundation was. Territory planning and AI deployment are the two moments when companies suddenly see the problem they've been living with. Once you're trying to automate anything at scale or allocate accounts to reps based on CRM data, you discover that roughly 20% of account assignments are flat wrong.
The conversation covers how 86% of records with errors compounds across every downstream system, why companies spending millions on data providers remain unhappy, the domain problem that causes confusion between Starbucks HQ and a Starbucks location with the same domain, the two parts of solving data problems (bad data vs. correcting data), how to diagnose if your organization has this issue, why territory design is the highest-leverage fix, and what rev ops leaders can actually do right now to start fixing the foundation before layering anything else on top.
What we cover in this episode
-
1:15
Every revenue leader hits this moment The universal experience of discovering broken data when attempting territory planning or AI deployment.
-
4:06
Garbage in, garbage out lesson How Kernel pivoted from building an AI SDR to data infrastructure after realizing data accuracy was the blocker.
-
5:49
86% of CRM records have entity errors Findings from anonymized proof-of-concept data showing structural entity-layer errors across dozens of companies.
-
8:18
Spending millions on data, still unhappy Why large investments in data providers and tools fail to solve the underlying entity and hierarchy problems.
-
10:22
How deep the entity layer problem goes Entity data accuracy is a prerequisite for all downstream data accuracy, from contact data to forecasting to planning.
-
14:57
20% of accounts assigned to reps are wrong Direct revenue impact when territory allocation is built on corrupted entity data at the CRM level.
-
20:21
The domain problem and entity confusion Why Starbucks HQ and a local Starbucks both appearing under oracle.com breaks enrichment and AI automation.
-
23:20
Data debt compounds as you scale Why organizations accumulate data inaccuracies over time and why fixing governance early matters less than pragmatism.
The lines worth sharing
Nothing will work with AI if the data is not accurate enough. It's very much a garbage in, garbage out situation.
Anders Krohn · 4:06
86% of records in a CRM have some kind of entity data error. That's the root cause of a lot of data inaccuracy.
Anders Krohn · 5:49
People spend millions on data and people are still unhappy. That shouldn't be the case.
Anders Krohn · 8:29
20% of all accounts assigned to sales reps are wrong. If you reassign those towards ICP accounts, that's millions in sales capacity.
Anders Krohn · 14:57
Common questions from this episode
What percentage of CRM records have entity data errors?
86% of CRM records contain some kind of entity data error, according to Kernel's anonymized analysis of proof-of-concept data across dozens of companies. These are structural errors like accounts mapped to wrong parent companies, not just missing fields.
What is the revenue impact of bad entity data in CRM?
Approximately 20% of accounts assigned to sales reps are incorrectly assigned. This directly translates to millions of dollars in lost sales capacity per organization when accounts are reassigned to reps with matching ICP. Territory planning improvements alone often run into tens of millions.
How do AI deployments expose data quality problems?
AI agents don't work around data inaccuracies the way humans do. When you try to automate at scale using corrupted entity data, the errors become immediately visible. Language models can't distinguish between Starbucks HQ and a local Starbucks if both share the same domain.
What is the domain problem in CRM entity data?
Many companies share similar domains. Oracle Spain and Oracle Global both use oracle.com, but are fundamentally different entities requiring different treatment. Data providers keying off domains alone return the same data for distinct entities, breaking enrichment and AI.
How should revenue leaders start fixing entity data problems?
Start by identifying where data problems surface: territory planning, AI initiatives, or field complaints. Take five accounts and trace what went wrong end-to-end. This usually reveals domain errors, duplicate records, or name mismatches, creating shared understanding across leadership.
What is Kernel entity data infrastructure?
Kernel provides verified entity data with a unique identifier (the Kernel Entity Reference Number) for every corporate entity. This creates a shared definition of which entity is being discussed, enabling AI agents and humans to reliably operate on accurate information.
SEO meta description
Anders Krohn, CEO of Kernel, reveals that 86% of CRM records have entity data errors, blocking AI and territory planning. Why fixing the data foundation matters most.
Target keywords
Anders Krohn
Kernel
CRM data quality
entity data errors
territory planning
AI deployment
data governance
rev ops
go-to-market data
data accuracy
entity hierarchy
sales forecasting
Read the complete conversation
Auto-generated from the episode audio and lightly cleaned. Expand to read or search inside.
Read the full transcript · 58 KB · Anders Krohn
ANDRES0:00I think we just learned the hard way that like, nothing will work with AI if the data is not accurate enough. So it's very much a garbage in, garbage
ANDRES0:07And that's how we go to to where we are today.
ANDRES0:10think a lot of revenue operators think that their situation is uniquely bad.
ANDRES0:14I spoke
ANDRES0:14probably more than a thousand people at this
ANDRES0:16is not unique like everyone has data challenges.
ANDRES0:19we did a test a few months ago where we
ANDRES0:22five like proof concepts and we anonymized and aggregated the data, and 86% of records in a CRM have some kind of entity data error.
ANDRES0:32But I actually think AI has done as a bit of a favor
ANDRES0:35in the sense
ANDRES0:36it's now very, very visible
ANDRES0:37that the data is not good enough to reliably do things in an automated way. It's just that we used to live in this world where we accepted the fact
ANDRES0:45we would have hundreds of people who would work around the data inaccuracies.
ANDRES0:49you know, you look inside of sales or like you might not be able to trust it,
EDDIE1:00actionable playbooks from the trenches working as go to market strategy and rev ops consultants for our clients here at Union Square. Consulting and candid conversations with revenue leaders in the market that have been there. Now let's get into it.
RACHAEL1:15Every revenue leader has had this moment. You try to plan territories, run a pipeline, review or deploy an AI tool and the data underneath it falls apart. Accounts mapped to the wrong parent companies, reps, working deals they shouldn't own and accurate enrichment, etc.. Let's talk about it today on go to Market Science. I'm with Andrew's Crone, founder and CEO of Colonel Andrew, spent years deploying AI into enterprise go to market orgs and kept hitting the same wall.
RACHAEL1:41This data was too broken for any of it to work. We get into what he keeps finding inside companies like Naveen Gong and ZIP. Why the entity and hierarchy layer is the thing that breaks everything else downstream and how to fix it.
ANDRES1:57Thanks for having me. Looking forward to the conversation.
RACHAEL1:59Yeah. All right, so give us the quick version. What is kernel and what problems does the company solve?
ANDRES2:05Yes. So maybe I can start like why why do how do we get here and like why are we here. So the way we started started off by building like many others, it turned out an AI, SDR with the purpose of like an autonomous prospecting. And really what we learned is that the key blocker to getting agents or humans for that matter, to make good decisions in revenue organizations is really like accurate data.
ANDRES2:29Like most CRM today, they have a ton of data, but usually that data is not accurate enough that you can reliably use AI on top of it. As we went down this rabbit hole of like, how do we get data accuracy to a level where we can reliably use it for making decisions using AI? And so what Colonel is today is what we call agent first entity data.
ANDRES2:50So entity data is the data about entities like, you know, you have a company like Oracle. Oracle has subsidiaries. There's hierarchical linkages. And really kernel is the most accurate entity data in the world specifically aimed at making that data.
ANDRES3:06Good enough that you can reliably use it for AI to make decisions. And so our big thesis is that we will get to a point where revenue operators will be using tools like cloud code tools like Clay to automate a very large part of revenue operations. And the kernel will become the foundational data layer upon which they do that.
ANDRES3:26And so it's a very exciting time for us. We've been growing growing very fast. We work with some phenomenal companies and very excited to have conversation
RACHAEL3:33Yeah. Awesome. And you know, that's what we we talk about that all the time. At USC's. You need to have the right data and accurate data consistently in order to implement AI into your systems, because otherwise you're just accelerating and amplifying the completely wrong things, and you have no idea what direction you're actually headed in unless you actually have accurate data to build off of and use AI with.
RACHAEL3:56So you guys didn't start out building the data infrastructure company in the beginning, I believe, right. The first product was an autonomous AI, SDR. What happened with that?
ANDRES4:06I mean, I think we just learned the hard way that like, nothing will work with AI if the data is not accurate enough. So it's very much a garbage in, garbage out situation. And that's how we go to to where we are today.
RACHAEL4:17Okay. So you found the problem. You saw the problem yourself. When you're trying to make this thing on your own and you're like, you know what? This is actually a problem you should focus on instead.
ANDRES4:25Yes. We spend like, you know, a ton of time inside of Salesforce trying to get this thing to work. And it just didn't work, like the data was just not good enough reliably be able to use language models on top of the serum. And so we said, okay, like let's let's not focus on that. Let's focus on actually how do we solve this data infrastructure challenge.
RACHAEL4:42Do you guys think you'll start doing autonomous AI SDR stuff now or you just you to to into the data infrastructure stuff now?
ANDRES4:49I think we see ourselves as like a data infrastructure or company like data accuracy is really our bread and butter. That's what we promise people. And really the vision for, for kernel is like, how do we build like a verified view of the economy that agents can interact with? And I think there are great tools in the like orchestration automation space, like tools like clay, and even the model providers themselves with tools like code.
ANDRES5:16And so it's much more likely they will start to do data infrastructure for other types of personas than it is that we would start going back into orchestration and automation.
RACHAEL5:27Okay. That's right. So kernel works with companies like Naveen. I don't know if I'm pronouncing that correctly. Yeah, yeah. Naveen Gong zip alpha. So these large organizations with big go to market teams, when you're when your software plugs into their CRM, what are the patterns that you keep finding what's consistently broken in these big organizations?
ANDRES5:49Yeah. So I think a lot of revenue operators think that their situation is uniquely bad. But what I can you know, they can. I spoke with more probably more than a thousand people at this point are is not unique like everyone has data challenges. We did a quick we did a test a few months ago where we took five like proof concepts and we anonymized and aggregated the data, and 86% of records in a CRM have some kind of entity data error.
ANDRES6:17So like would be like the wrong name, the wrong website, wrong legal name, wrong headcount, wrong wrong revenue. And so CRM are consistently broken in somewhat unique ways. But there are patterns and say the most consistent pattern. That's a bit of a sort of like earned inside. Is that a lot of data inaccuracies in CRM or in data warehouse come back to confusion around which entity is this, right?
ANDRES6:47So like let's say you have a record could be like Oracle Spain where it has the website of Oracle. Com and like you know it's like it's some notes of tell tell you that it's Oracle Spain right. Like the a lot of what tends to happen is that you plug in different kinds of like enrichment providers, but they are talking about a different entity like they had kind of Oracle global is very different from the head kind of Oracle Spain.
ANDRES7:10And so I would say that's the root cause of a lot of data inaccuracy. And like until you solve that you can really get your data to a state where you can reliably make decisions on top of it without some kind of human verification.
RACHAEL7:26What is surprised you about how far off, you know, even some of the most well-run organizations have been?
ANDRES7:33I would say it's actually not so much that the organizations themselves are bad. I think quite often what surprised me is that how much money people are spending without the problems being solved, right. So I think a lot of revenue operations professionals will feel this. Like, you know, the sales reps are like, why don't you fix this?
ANDRES7:58Like, this is your job. But the reality is that for a lot of organizations, like fixing the underlying data has historically been outside of the control of the web ops team. And a very big part of it was like, how do we give people the controls, the tooling to be able to fix the data in their in their system work.
ANDRES8:18So that's been the most surprising thing. So people spend millions, in some cases millions, in some cases tens of millions of dollars on on data. And people are still unhappy, right? That shouldn't be the case.
RACHAEL8:29Is there a common moment that people see, you know, when their team realizes that the problem is way bigger than they thought and like, oh my God, we're spending these millions of dollars on this data. We didn't realize it was this bad,
ANDRES8:40I would say historically, territory allocation or annual planning has been like the two big triggers for organizations to realize that the quality of their data is not good enough to reliably make decisions. One thing that is an interesting pattern that we're starting to see reasonably is that any attempt at using AI at scale, on top of the CRM makes people very quickly realize that the data in the serum is not good enough.
ANDRES9:10And so, like one thing is like using an AI tool as a sales rep, like using some kind of like copilot or using ChatGPT directly. Like usually that's like good enough, but like once, just once you start to get to the point where you're operating organization wide on top of, say, like a 20 year old Salesforce instance, that's when people realize, okay, actually, like this big grand old like, you know, AI roadmap we have is like, there's this massive wall in front of it, which is data accuracy.
ANDRES9:37And until we fix data accuracy, we're not going to be able to do anything with AI at scale. So that's probably like a year ago, it was like 0% of our opportunities that had that as a compelling event in knowledge, probably like 10 to 20%. I think a year from now be 100%.
RACHAEL9:55Oh, 100%. Well, I think right.
ANDRES9:57Now everyone is like everyone wants to build with AI and it doesn't. Yeah. You know, so at some point people are going to like, you know, everyone will have to fix this.
RACHAEL10:05Yeah, absolutely. That's a really good point. Like at some point everyone will have to fix this because everyone's going to be using AI, right? I mean, as far as we know, something big could happen in the next ten years. Who knows where. They're like? Everyone's like, you know what? AI gotta shut it down. The robots are going to take over the world.
RACHAEL10:22But where we're sitting right now, that's probably not going to happen. So how deep does the problem go that you tend to find? So, you know, the entity layer, the corporate hierarchies? Can you explain, like, how deep it really goes and why? Why are these things breaking everything else downstream?
ANDRES10:44Yeah. So I think the easiest the answer is it goes
ANDRES10:48all the way to the bottom, like as deep as as deep as you can imagine. Like it's like entities. Data accuracy is a prerequisite for any kind of other types of data accuracy, whether it's contact data or forecasting or annual planning or whatever it might be. The easiest way to illustrate it is domain names.
ANDRES11:07So take like Oracle Global the domain of Oracle Global Oracle. Com take Oracle Spain domain of Oracle Spain is also oracle. Com but fundamentally like those are very different different entities. And so being able to distinguish between those is is really really important. And historically data providers have operated off of the domain oracle. Com. And they will then return the same thing for two things that are fundamentally different.
ANDRES11:40And an AI agent will not necessarily know the difference. And so that's when like errors start to compound. And so yeah so it starts from from the beginning basically. And I think unless you solve that root cause you're not going to be able to truly solve the problem of data accuracy in your CRM. And people are waking up.
ANDRES12:01So that is like it's a bit of a nuanced thing to understand. But like once you see it, you cannot unsee it.
RACHAEL12:07Yeah. And I don't know if this is I mean, I'm guessing it's related, but you can tell me if it's not. We've just recently started focusing more on LinkedIn campaigns and trying to like, build targeting lists for for LinkedIn, for thought leadership posts and stuff like that. And while building these lists, we found like sometimes the name of the company that the AI will collect from the website or like the domain name isn't the exact same name that they have on LinkedIn, like on their company page.
RACHAEL12:35So when you're targeting for LinkedIn, you have to have the exact name that it shows up on the LinkedIn company page, or else it's like harder for it to find a match. Is that similar?
ANDRES12:44I think yes, that's one example. Like every everything that's related to any kind of data that relates to the external world starts with a need for agreement about like, what are we talking about? Right. And so like it's a kernel. As an example, kernel has the domain kernel AI. We have the legal name momentum AI limited. That's what our company used to be called.
ANDRES13:09There's also another company. It's a brain scanning company called kernel. And there's also another company in the rev up space that used to be called sorry, that is called momentum that got acquired by Salesforce. And so
ANDRES13:21unless you are like squeaky clean or like this is this is this kernel,
ANDRES13:26you can end up with all sorts of different things downstream from that wilderness, like through LinkedIn campaigns, ads, contact enrichment, intent data
ANDRES13:33because they're very different companies.
ANDRES13:35And so so you're you're absolutely right. Most companies will probably be having ads that are going to accounts that are not the same accounts as the ones that the sales reps are prospecting into, because the marketing team does not use the same entity definition as a sales organization.
RACHAEL13:50And so what does this mean for companies then? Like I assuming, you know, salespeople are looking at their lists and they see the data that they get enriched with and they're like, hey, I got to go to this website to look up this company. Before I call on them, they click on the link and it takes them to like the wrong website.
RACHAEL14:05So first they're wasting time having to realize that it's the wrong website and then realizing that, like whatever else data that they have on this account might be wrong, and then they're spending their own time needing to do their own research initiative already just been handed to them. What else like goes wrong and breaks down operationally when this data is you said 80% of these records have some kind of.
ANDRES14:29Has some.
RACHAEL14:30Yeah, yeah.
ANDRES14:31The way I would think of the,
ANDRES14:34the way I would think about the cost of entity data inaccuracy.
ANDRES14:37It starts with direct spend. Like you're spending money on data providers, consultancies, deduplication tools and is not necessarily working. Then you have a lot of indirect spend that usually sits with the sales organization. Probably on average, it's somewhere between like two hours and a day a week per sales rep.
ANDRES14:57That goes to some kind of data collection or data disputes. So that's like a drag on revenue when it starts to directly impact revenue is accounts that are inaccurately assigned to sales reps. So is like probably 20%, 20% of all accounts that are assigned to sales reps are wrong. Like they're not ICP. Like isn't B accounts assigned to enterprise reps out of business, whatever it might be?
ANDRES15:23And if you can reassign those sales resources towards accounts, there are ICP that usually like quickly comes into the like millions of dollars in sales capacity. Yeah. And then there's like a lot of like sort of hidden costs, like, you know, worse planning, worse decisions, worse segmentations, worse ads that can that can be a bit hard to measure.
ANDRES15:45But I actually think AI has done as a bit of a favor in the sense that it's now very, very visible that the data is not good enough to reliably do things in an automated way. It's just that we used to live in this world where we accepted the fact that we would have hundreds of people who would work around the data inaccuracies.
ANDRES16:05But you, you know, you look inside of sales or like you might not be able to trust it, but a language model doesn't know, right? So like, you have to somehow give the language model the tooling to say like this is the verified truth and you can now make decisions based off of that truth.
RACHAEL16:21how do we and I know like, the answer might be like just yeah, Colonel, but like, how do we start working on this, this technical debt and like, how do we figure out, like let's say CRO revenue leader listening to this is wondering like, oh, how bad is it in my organization? Like you're saying, 86% of all records have an error.
RACHAEL16:41How do I figure out what my percentage is? And like, how do I see the symptoms of this issue? Because they it seems like a lot of like pinhole leaks that aren't super apparent right away. So how is somebody supposed to, like, really figure out where this is happening in their company?
ANDRES16:58Yeah, I think like a very practical way to do it is to say, okay, like where, where do you hear a noise about data? Right. So most chief revenue officer who can be like, they hear that accounts are missing in APAC or like the S&P team is unhappy because we inaccurately segmented accounts based on headcount. But the S&P rep saying the headcount data is wrong.
ANDRES17:22So like usually there is no somewhere. And I think the easiest way to like figure out like what is the root cause of that is to literally just take some of those examples and just like trace them from like end to end, like what went wrong here? What is the truth? And it doesn't have to be that many.
ANDRES17:45Like, even if I think most people like if the Rav ops leader and the chief revenue officer went through five accounts together like this is an account that the rep thing should be in the serum, but it was not in the CRM. This is an account that the S&P rep thing should be mid-market, but it's not like whatever that might be.
ANDRES18:02And what will usually emerge is the domains are wrong, the names are wrong, they are duplicate records. And I think that creates like a shared understanding of like the cost of inaccurate foundational entity data. And then it's just a question like, how do you then go out to how do you then go out to solve solve that? And yeah, that could be a kernel.
ANDRES18:31You can try other things like I don't want to turn it into like a grand sales pitch. Like the reality is like, you
ANDRES18:38you can try all sorts of different ways, but like if you if you're not in line of what the problem is,
ANDRES18:42then what tends to happen is people just layer and layer and layer things on top.
ANDRES18:46So for example, a lot of go to market tools operate off of the domain.
ANDRES18:52buy more go to market tools, you will have more tools that operate off of the domain. But the vast majority of CRM one the domains are usually not accurate, and two of the domains are not a reliable predictor of unique entities. For example, you sell into large enterprise.
ANDRES19:06A lot of entities have similar domains. Nothing like getting that sort of slightly more nuanced understanding makes it easier for people to then come around the table and say, okay, like we have to fix this problem first.
RACHAEL19:18So how does kernel fix the problem? Like when you guys were when you when you found that this problem existed for your other product you're trying to make and you decided to fix it for yourselves, what were the steps that you first took to start creating something that could do this at scale?
ANDRES19:35Yeah, I mean, we've been through all sorts of different edge cases. Like this is what makes this problem hard, is that there is a gasoline different edge cases, and you really have to like capture all of those. But the the most important question that we have to answer on behalf of our customers is which entity is this? There is a record theorem and usually has a name, usually has a domain, and might have an opportunity.
ANDRES20:00You might have some knows it might have some other fields, needs to be a shared understanding of what is it. And and the way we do that at kernel is with what we call the current ID, the kernel entity reference number, which is a unique identifier for every corporate entity in the world that we can use. This ensure we're talking about the same thing.
ANDRES20:21Once we know we're talking about the same thing, then we can start to do all of the downstream things like, okay, what's the corporate hierarchy? What's the headcount of this entity? Maybe you can start to use web research agents for more custom data points. But if you're talking about the wrong thing, like if you think I'm talking about Starbucks Corporation, and I think I'm talking about Starbucks around the corner, who, by the way, has the same domain.
ANDRES20:46Then everything else is going to be a complete waste of time.
RACHAEL20:48Interesting. So from from like a company's perspective
RACHAEL20:53who like not not like Colonel's perspective that you guys, you guys already know like you have all these programs and ways to figure this stuff out. But how before kernel existed, how would a company be able to do something like this?
RACHAEL21:07Just people? Yeah.
ANDRES21:08I think that is the reality. Like it's very, very common that there's like two
ANDRES21:13there are two parts to solving a data problem.
ANDRES21:16There's by data
ANDRES21:17and there's correct data.
ANDRES21:19before AI, the way to correct data was with humans.
ANDRES21:25And you know, I would like strongly encourage like whether you use kernel or not, like anyone should try and like clean and correct data with AI.
ANDRES21:32Like it's absolutely like, you know, a worthwhile endeavor. I'm obviously biased towards kernel, but like, I'm not here to argue that you should try and do nothing. Like you should do something for sure. Yeah. And the way that people used to do it was either they would ignore it, and then a large number of humans would work around it.
ANDRES21:57And by large numbers are mean. The entire go to market organization say like 500 people. And they would have specialists, freelancers, outsource labor, pose a small number of people whose full time job it is to try and correct the data.
RACHAEL22:13Yeah, and that's just not scalable. Not once you have thousands and thousands of accounts to deal with.
RACHAEL22:19At what level is it the responsibility of processes and you know, the salespeople and marketing or whoever to follow a specific process and be inputting things into the CRM as accurately as they can.
RACHAEL22:36And then for like the rev ops team, to ensure that they have the correct fields and stuff in the CRM so that it creates like guardrails for these teams, inputting this data consistently and accurately and as much to the best of their ability.
ANDRES22:51that is for sure part of it. And I think once companies reach a certain scale, like data governance is a real thing and is important. But I also think it's like, you know, it's important to be pragmatic and say, like, if you do that as a young company, that is for sure premature optimization. Like if I, if I that has like $1 million in AR and they're like, oh yeah, we have like we use the Salesforce sandbox and like we don't allow people to create reps allowed to, to create accounts.
ANDRES23:20And we have this like support process for things that fall between territories. I would sell like that sounds like a complete waste of time for companies age. And so what makes it hard is it as companies grow, they accumulate a large amount of data debt from having not had that data governance. And I think what great rev ups and data teams do is that they come in with a pragmatic view and say, okay, there are two parts of this equation.
ANDRES23:46One is like, how do we cure the pain? And then the other one is like, how do we prevent the pain from coming back? And a big part of prevention for sure is to have great governance, great tooling, great systems, great processes, but don't do it for the fun of it. You should do it. If it solves a business business problem.
RACHAEL24:06That's fair. Yeah, there is something to be said for not like putting too much on your reps, because if it's too complicated and there's too much for them to like, add into the CRM, they're not going to do it and they're just going to be very upset with you. So almost every go to market team we talked to this built some kind of AI tooling.
RACHAEL24:25So they've got a bot here and automation there. None of them are really connected to each other. Do you see this across your customer base to like what's actually happening with these disjointed AI workflows?
ANDRES24:39Yeah, it was interesting. I was at a Riff Fest conference in New York last, last week, and there were sort
ANDRES24:47one thing I was interesting is that it feels like was there are two caps when it comes to like AI projects. There's the governance camp, and then there's the let the reps loose and like learn, learn. And I think both of them have merit.
ANDRES25:06My best guess on and this is a bit of a politician answer, but like my best guess, like where things will land is somewhere in the middle. And what I mean by in the middle is there are certain processes that should be centralized. The obvious ones are the ones where people have different incentives for data's be in a certain state.
ANDRES25:32Right. So like reps might have incentives for the headcount number to be X, and the company might have an incentive for the headcount number and serums to be. Why? Because that's how the territories are like. That's a clear thing that should be centralized. But then you have certain things where you might benefit from reps having some level of flexibility in terms of like understanding, you know, like the one of these example could be like count research, for example, or some level of flexibility for A's and like how they use AI tooling for their prospecting.
ANDRES26:03And so I think what what we're seeing that the best rev UPS teams do is provide some level of flexibility, whether it's for sales reps is an open question, but at least for the wider, you know, organization, but with clear guardrails for how to do that well, because most likely will now.
ANDRES26:30We have the way I think about like the phases we've gone through, like we've met through like the sort of mega hype phase, then we probably where we are now is like the make it work in production phase, like things are working is a question like what's working? And like it has to be somewhat orchestrated chaos.
ANDRES26:50I think the next phase is going to be around, like how do we as part of that, like now, slightly more orchestrated way of thinking about AI.
ANDRES26:59How do we then decide
ANDRES27:01what's a distraction and what's not a distraction? Right. Like because it's it's helpful as an organization for sales reps to be familiar with the use of AI.
ANDRES27:11But the reality is that that's not their job. Their job is to sell. Right. So there comes a points, okay. Like, you do this thing on your weekend and you do this thing on your job.
ANDRES27:22And like, this is where we draw the line as an organization because there is a lot of like productivity theater. It's it's easy to feel productive with AI.
ANDRES27:32That doesn't mean you are productive. And I think that's all ultimately like the sort of line is going to be drawn somewhere. And I think people are sort of still figuring out exactly where that is. But I think it's important to be deliberate about it.
RACHAEL27:46Quick pause. Everything we talk about on this show. Diagnosing go to market ops. Prioritizing projects for revenue. Impact processes. Metrics. Insights. Building a predictable go to market engine. We've built frameworks for all of it. They're free and undated on our website. Union square consulting frameworks. The link will also be in the show notes, so make sure you check that out.
RACHAEL28:09All right. Back to the episode.
RACHAEL28:12can't remember who Eddie was talking to on the podcast a while ago now, but they're talking about phase, I think. Craig Craig, Craig Rosenberg It might have been he was talking about phase one and phase two AI and what they were seeing in their data on companies that were actually getting results with AI. So companies doing phase one AI was like productivity based.
RACHAEL28:37So being able to send out more emails, being able to cut down, you know, hours doing a certain task. Phase two AI was stuff around a certain business outcome. So like AI, call scoring, getting AI to help with coaching to make your sales people better. So it's like, I like listening into your sales calls and being able to pick out like where you're falling off, where the where the lead is, you know, disengaging stuff like that, what you can do better.
RACHAEL29:10And they found that companies doing phase one AI did see some results with it. But the the vast majority of results came from companies with a combination of both specifically with phase two, but a combination of both saw the best results. So not just focusing on productivity, because it can be so easy to just be like, oh, we use AI, use AI all the time, we're AI first, and all you're using AI for is to like be kind of your assistant to like, cut down tasks, but having a specific business goal and business outcome in mind, something that ties directly to revenue, and finding ways for AI to help with that.
ANDRES29:51Yes. I mean, yes, I think it's like it's a ultimately like
ANDRES29:57outcomes have to move or nothing. Nothing not moves. And I think like there's sort of like the it's easy when you go through these like big changes to the way that the world works to like forget about the like physics of business. But the reality is that, like, you know, businesses sort of exist to like make more revenue and reduce cost.
ANDRES30:18And like the same will hold true eventually for like how we use how easy I.
RACHAEL30:23So do you see like with the with the companies that you're seeing using kernel and stuff. Which ones. Or I guess you probably can't speak like actual companies, but even just anonymously, what use cases are you seeing them getting the most results
ANDRES30:38I would say like the
ANDRES30:40by far most leveraged intervention that everybody's team can do with better data is in territory design. Like there's just no no question like that. Territory design sits off stream from pretty much all revenue generation. And getting it getting it right is a multi-million dollar problem to solve. And it can cause both a lot of frustration, but it can also cause significant revenue loss.
ANDRES31:05And so I think that's that's remains the main use case. Then there's like a long list of things that are somewhat adjacent to that that could be optimizing ad spend a bad debt prevention from like making sure you know which company you're selling into. And will they have any kind of like credit credit risk. It could be like headcount, segmentation, headcount, sorry, headcount planning, forecasting like everything like somewhat relates to to data.
ANDRES31:38But I think the biggest reuse case is for sure territory design. I think the biggest re use case a year from now will move more towards productive use of AI tooling, but we're not quite there yet. Right. So like we're saying, customers adopt tools like cloud code, like some type of the data warehouse and able to do things that can move like millions and millions of dollars, but it's still a subset.
ANDRES32:10I would say that like, does that effectively at scale, but it's changing. And if you're not on it, I think, you know, not to be like fear mongering, but like there is a there is a big difference in terms of the go to market efficiency, often go to market organization where the rev ops team is able to effectively use AI tooling across the entire system of record, and one where the rebels team is only able to use AI tooling for individual like rep use case like the difference is massive.
ANDRES32:46That doesn't mean that like there can still be like a difference in like what you build. What do you buy? That depends on like capabilities. It depends on where you think you want to build competitive advantage. But I don't think as a rebel organization you really have a choice but to think about, okay, what needs to hold true in order for us to make this technology work effectively at scale across the entire system of record.
RACHAEL33:12So what do you think is the solution to that?
ANDRES33:16Like how do you get started?
RACHAEL33:18Yeah.
ANDRES33:21I do think that like there are sort of two parts there. Like one is like where do you put your resource in terms of like which problems are solved? And then the other one is like, how do you get your current state to be good enough in order for you to be able to solve it in terms of which problems to focus on?
ANDRES33:38I think that's like the principles of prioritization have not necessarily changed. It's like, you know, like which which one do you think will have the biggest, biggest business impact? What do you hear from the field like? Is it like count research takes a lot of time. Is it like the territories are not good enough or like we're not ramping reps fast enough?
ANDRES33:55I think that's just a question of like, these are all the possible initiatives. Let's rank them in terms of which ones you then get the benefit of using AI the most is really a question like where are humans today processing data in a way that can be processed by AI, and that that's sort of like ties it back to this requisite point of there's no question in my mind that, like the most important thing to be able to use AI effectively is to get your data in a state where when a language model uses that data, what you get out on the other end is reliable and it's trustworthy.
RACHAEL34:33Is there a specific result or store your use case that you can anonymously talk about? And it's probably might be like sensitive information, so let me know if you can. But like results that you've seen or like,
RACHAEL34:49like a specific issue that they were having and the way that they fixed it and then the results they saw on the other side.
ANDRES34:55Yeah. So one of the case studies that's live on our, our website is from a company called Alpha Sense, which like many of our other customers, uses kernel as the primary provider of entity data for for all global go to market operations. And their estimates were that 20% reduction rep time spent on like research disputes, data corrections and so on, and a 15% increase in overall revenue capacity for for the organization.
ANDRES35:29When it comes to like the efficiency of the territories that are assigned and like that, that quickly runs into the tens of millions of dollars. And that's also why I think is so important as a rev ops organization, that you think about these
ANDRES35:41system level initiatives rather than the like rep level initiatives. The system level initiatives are the ones that will move your entire PNL, whereas the rep level initiatives might give you like a productivity boost for individual reps.
ANDRES35:54But it's just not.
ANDRES35:55I would say the juice is not worth the squeeze to the same extent.
RACHAEL35:59So when that data layer is solid, what changes for the people on top of it? The rapture of ops leaders leadership. What can they do now that they couldn't do before?
ANDRES36:09I would say the the key word from an emotional perspective is trust, right? Like once you trust your data, you can start to make decisions. Using that data, like data on its own, doesn't really have any intrinsic value. Data is only valuable to the extent that it changes the decisions that you would make. And to do that, you need to trust it because it's specifically for for rev up.
ANDRES36:33Seems like that could be more confidence in like how many reps can we ramp and where should be random. That's obviously a very big question, right? For big companies. That drives a very big part of the penal. Another key component is like, do we have confidence that the accounts that are assigned to sales reps are for sufficiently good quality, that we give them the ingredients to hit quota?
ANDRES37:02So those two things combined for most of businesses drives the entire business, right. So that's really, really, really critical. And that's also why like often when I speak with companies, they will have gone through all sorts of pain to try and solve that manually before.
RACHAEL37:20That's so important. Yeah, right.
ANDRES37:22Sorry.
RACHAEL37:23I was just saying that's so important. Yeah. Especially for companies that have stakeholders and like, they want accurate forecasts and the Crow needs to give an accurate forecast of what you think. You know, the revenue that you think is you're going to produce in the next two quarters, three quarters out. Right. If you don't have an engine that you can like predictably and consistently, you know, see the outcome of, then it's going to be so much harder to be able to tell your stakeholders, like, this is what we're going to be making.
ANDRES37:51Yes. So six months. Those are those are definitely the two biggest things that are like downstream from trust. And I think there's also like a sort of slightly more emotional thing for a lot of robots, people, which is like
ANDRES38:02the reality is that like robots, professionals are usually like highly educated, fairly technical people who have the capabilities to have an impact organization wide.
ANDRES38:12But a lot of people are stuck, like answering like individual data requests and slack, or doing like list uploads from conferences or whatever it might be like. I think being able to like, automate a large part of that away frees up rev up's resources for things that contribute more proactively towards growth, but is a bit harder to quantify.
ANDRES38:36But that doesn't mean that it can't be valuable.
RACHAEL38:38Yeah, we talk about that all the time. We have a framework actually, the rev ops go to market roadmap, where it's essentially trying to help revolves professionals like figure out how to triage big rib ops projects with the fire drills that are constantly coming up. Because you're right. Like it's it's the bane of probably any rev ops person's existence is trying to do projects that they know are going to be better for the company in the long run, but they just have no time because everyone has something that they need them to do, like.
RACHAEL39:06And it's always urgent. It's always like a fire that they have to put out right now.
ANDRES39:10Yeah, I mean, I know, I know, a lot of people have just like deleted slack, you know, it's like way to deal with it. But it's maybe not the most, you know, culturally earned way to do it.
RACHAEL39:21Yeah. I couldn't imagine man. So you're you're running workshops for your customers using cloud code for rebels. Use cases like territory planning and hierarchy mapping. What are teams actually building in those sessions.
ANDRES39:38Right. So I think again, it comes down to like
ANDRES39:41the key thing that's been unlocked by AI is the ability to do things that would previously require either a lot of developer resources or a lot of human resources. And so it's sort of like reduce the cost of customization. And so what people tend to do with those tools, whether it's now workshops or outside of those workshops, is to do the things that they wish they had the resources to do, but would just have been like completely out of reach historically.
ANDRES40:09An example of that could be, how do we make sure that the corporate hierarchies that feed into our territory location for enterprise segment where we accurately reflect, like what are actually the relevant buying centers within these hierarchies that we want our reps prospect into? Historically, that's been something where it's like, do you do it with freelancers? Do you just wait for the reps to collect in the sales?
ANDRES40:33But now we can execute like we have the data foundation. Now let's use AI to like here's here's Disney or here's Google. Which ones of these entities do we actually want to assign to? Reps like that could be an example of something that's very well suited for AI at scale. And that's the lens that I would generally, whether you use tools like cloud code or other things, that's the lens I would generally recommend that people take when they like look at like, is this problem suitable for AI or not?
ANDRES41:05You sort of have to ask yourself one, can a human do it with unlimited time? If a human can't do it with unlimited time, doesn't mean that is impossible for AI. But like at least like it makes it harder. And then like assuming a human can do it with an unlimited time, it's like, okay, like is there, is it like data processing where we can somehow give an agent directions, then it can that it can reliably trust to process things at scale.
ANDRES41:37And again, like this is very these are people who deal with large complex serums and, and I think is not super valuable to think of. Even if you might test at a smaller scale, it's not super valuable to think like, what can we do for this one rep? It's much more valuable to think about, like, what can we do for the system that changes the behavior of all of our reps?
ANDRES41:59We can then test it with one rep. But I think wraps people can waste a lot of time optimizing for individual reps rather than optimizing for the system itself.
RACHAEL42:08And I think maybe I just use AI differently than in the ways that you're describing, because I hear like can of human do it without unlimited time.
RACHAEL42:17My answer for for a lot of things I use AI for, the answer is like, well, yes, but can the AI do it at all? Because like, I don't know, maybe it's different for what I use AI for versus what you use it for.
RACHAEL42:29But I find that like AI gets things wrong so often and it's such a pain and like I'll have it doing a task and even just like data sorting and stuff, I'll have it do a task and I have to always double check what it's doing to make sure it actually did it right the first time at all.
RACHAEL42:46Right. Do you do you find that as like an issue or how do you deal with that when trying to do these workshops and trying to build out these AI use cases?
ANDRES42:55Yeah, that's a good question. So.
ANDRES42:57There is probably like a slight difference in where I have like a bit of a bias here is like our entire business is predicated on like how do we get AI to behave? Like how do we do what we wanted, what we wanted to do. And we obviously invest in enormous amount of resources in trying to do that.
ANDRES43:12And so I do think that your experience is probably a more aligned with the general experience of rev ops professionals, which is, well, it can not do this thing. And clearly you're right, like often fails, like why would it be able to do this much bigger thing. And I think that is a good question to to to ask.
ANDRES43:38And the answer I think is like of this bigger thing like that's sort of like cut it down into like something that's like a manageable chunk that can be processed reliably. So for example, it's like it's much easier to get AI to reliably fix, say like one custom data point from like web research that goes into one field in Salesforce that can be validated with like a test set and with good examples.
ANDRES44:13Then it is to get AI to do something that has like slightly more vague, vague instructions. I think we'll probably get there, but a very big part of like kernel, the product, which I think, you know, people can take inspiration from, is they're also like, how do we break down this, this, these problems are like manageable chunks that we can evaluate the agent's performance against.
ANDRES44:35Because you are right. Like when you just let loose, it's a bit hit or miss for sure. And there's a lot of yeah, it's it's always easy for I've read like an interesting tweet the other day. There's like, you know, CEOs of like just completely like lost touch with like what wasn't actually like what can reliably be done with AI because I don't necessarily see the last mile.
ANDRES44:58And I think that is probably true. Right. The devil is in the detail. And I think people are very good at like getting the problem defined clearly enough that you can get to the level of detail and align everyone around. Okay. Like how do we actually solve this?
RACHAEL45:12I think there's a lot of times too. It's just like like AI just can't do it. Like you need a person, like you need a human touch to to figure it out. And I do like a lot of creative work as a marketer. So maybe it's less maybe that's less relevant for people doing work with mostly numbers and data and stuff.
RACHAEL45:32I'm sure it is. But yeah, I feel like sometimes just like AI can't do it. Like maybe a person could do it with unlimited time for sure, but but AI can't do it with any number of tribes,
ANDRES45:43Yeah. I mean, yeah, I think it's some, some of it is also just trying and
ANDRES45:48seeing if it, if it works.
RACHAEL45:49So one thing that came up in your prep is that. So this isn't just a rev ops problem. It touches finance billing forecasting everything. How does fixing entity data become a cross functional initiative. And who needs to be driving
ANDRES46:05I think rev UPS is very uniquely positioned to drive it in most organizations. Once companies reach a certain size, they might have like data teams or master data management teams, within which it makes sense for some of these, these data specific problems to to set the the way that it becomes cross-functional.
ANDRES46:23Is really that like the.
ANDRES46:24The foundational data in the serum answers to questions like, who are our customers? Who are our potential customers and who are our partners? Which for most B2B businesses like that is what the business is. So like answering that question sort of like feeds into pretty much everything in the, in the organization. To give some specific examples of that could be
ANDRES46:45how do you align marketing and sales.
ANDRES46:47Right. Like that is like the age old question books. Books have been written.
ANDRES46:53Good. First step is to make sure that they are focused on the same target accounts.
ANDRES46:58What's a prerequisite for having the same target account is being aligned? And like, what is the data that defines whether something is a target account? A prerequisite for that is accurate data. So that's like the most obvious thing. But then when it comes to for some of the interaction between the go to market organization and the finance function, that could be everything from like, how do we are there patterns in our customers where we can see that they are bad buyers?
ANDRES47:24That could be literally an eyesore, or it could be things like bad debt. Right? Then you start to have an interaction between the rev ops data and some of the like, you know, credit credit risk data and other interaction with like strategy or finances for our annual planning. How many sales reps can we hire and which region should we hire them for?
ANDRES47:44If you don't have a good data foundation is very, very hot to answer that question without just sort of like praying for for good, good weather, which is usually not a very productive way to. So so it's sort of interferes with everything. And I think one of the great things for rev ops, like technically minded revenue leaders in general, and for real specifically, is that these technologies will elevate the importance of that function to like more strategically important, because they are so uniquely positioned to fix the data.
ANDRES48:21That is a prerequisite for running the entire business.
RACHAEL48:24Absolutely. And I'd say even customer success to like you have as you go after the sale. Because especially with like these tech companies that are recurring revenue businesses like
RACHAEL48:35if you're selling, let's say you have an ICP, you think is your ideal customer persona because they buy really easily, but they never stick around. They, you know, they have horrible adoption, they don't actually use the product, and then they churn after six months.
RACHAEL48:51So that's probably not even your ICP. But if you don't have that data layer that's communicating to marketing and to sales, that like this customer is actually a horrible customer, and we're not even getting enough like lifetime value for them to justify what we're spending to get them. You're going to be hemorrhaging revenue and not understand why, because everything's going to seem like it's running perfectly and like everything is where it should be.
ANDRES49:15Completely agree.
RACHAEL49:16Yeah. All right, so we're running up on time here. Anders, I wanted to ask,
RACHAEL49:22was there anything that any question that, like I missed asking or didn't ask that would encompass something that you've been wanting to talk about or that you feel like you don't talk about enough?
ANDRES49:32I think like one. And we have sort of touched a bit upon it in a few different places is like, like tomorrow. How do I start? Like, I think a lot of, a lot of people feel like they're behind. I can tell you that they're in the vast, vast majority of cases. They're not like everyone. Everyone is behind, you know, like it's changing so fast and like, it's very hard to keep up.
ANDRES49:55my recommendation to people is like, try and just like do some targeted experimentation. Like, you don't have to necessarily have the perfect tooling, but usually just starting with the model providers themselves, a good, good place to start and like try and break the problems down into small pieces of it that you can test at a small scale.
ANDRES50:15And I think like building, building confidence from, from there. And so for those who, like sit out there and like, oh, I don't know what to start, I feel like so behind like get some revenue leaders in a room, they're like, these are all the problems that we have. Like which one do we think is most important? Like maybe even ask ask AI how to work to use AI or just like like how?
ANDRES50:39Try and break it down, start one place, like take 50 accounts or whatever it might be and get going from there. Otherwise it can be people can quickly get stuck in sort of paralysis analysis. The reality is that,
ANDRES50:49everyone is trying, you know, no one has really figured this out yet.
RACHAEL50:53Yeah, I think a lot of people, especially on LinkedIn, if you're like me and you're chronically online, on LinkedIn, it can be easy to think that there are so many people that have it all figured out and there's so many people that are, you know, showing their workflows and their AI hacks and everything and and feeling like, oh, man, I don't.
RACHAEL51:12I just learned what Cloud Code was yet last week. Like, I'm so behind. But yeah, everyone really is in the same boat. And the people I think, well, I'm sure there are some people who like, really, really know what they're talking about. But I think a lot of the people who act like they're gurus right now with AI,
RACHAEL51:29probably aren't nearly as far ahead as you think they are because it's new.
RACHAEL51:33All of this is so new to us, and it's changing every single day. How could you be a master of something that just really started being a thing and in the last couple of years? So that's my take on it.
ANDRES51:46Completely agree. Completely.
RACHAEL51:48All right. Anders, so where can people find you if they want to follow along with what you're doing or check kernel
ANDRES51:54Yeah. So if in some of like following along like my LinkedIn profile is is a good place to start, just go on LinkedIn, search for Anders Krohn. I'm sure my name will appear wherever you listen to this, this podcast, or if you want to understand more about kernel then our website kernel dot.
RACHAEL52:10Awesome. And those links will also be in the show notes as well. Thank you so much Anders. Have a really great rest of your day.
ANDRES52:17Yes, great to see you. Thanks for joining.
EDDIE52:19Thanks for listening to the episode. If this resonated. Please give us a five star rating and a follow. It helps us reach more people, and you get our latest and greatest content without having to search for it. And if you're looking for hands on help and go to market strategy and or rev ops, please reach out to us.
EDDIE52:33We help our clients with everything from annual planning to improving processes and go to market, implementing systems to support those processes and go to market AI. We're always happy to offer a free consultation to help you identify the best opportunities to improve your go to market engine, with or without our help. You can find us at Union Square consulting.com and the info will be in our show notes.
