The AI-Native Services Playbook - with Jake Saper, Emergence Capital

0:07 Welcome to today's episode of the AI and ROI podcast. Today I am joined by Jake Saber, general partner at Emergence Capital. We'll be discussing five topics today based upon the emergence capital's AI native services playbook. First, why product leadership is so critical early on. Second, the mirage that is product market fit. Third, why outcomebased pricing is so important in an AI native services model. Fourth, revenue per employee as a defining metric. And then fifth, the northstar metric. So Jake, could you take a moment to give a brief overview of your journey to becoming a guest here on the AI to ROI podcast?

0:51 >> Thanks for having me, Ray. my journey to coming here. So I'll start with my journey to to emergence which I guess brought me to this podcast. So I was born in Austin, Texas or I was raised in Austin, Texas to my parents are co-founders. They've started a bunch of companies together. so I have inadvertently been sitting shotgun on the early stage startup journey since I was a child. when I was in middle and high school, my parents would bring strategic questions about the business to the the kitchen table and we would debate they should enter this market and how they should think about this new product and pricing. So, I've kind of been doing versions of this job, unknowingly, since I was a child.

1:29 I'm very grateful my parents for that. after college, I did consulting. I'm a very structured thinker. I like frameworks a lot. So, consulting was a very, good place for me to hone those skills. And then I did a startup, which I had to do because of my parents. and where I was raised. my startup was developing utility scale solar power plants in India and Africa. >> So a very intense experience. I was living in Mumbai and building very large power plants and selling the power to governments. which was a wild, crazy, horrible terrible wonderful experience like all startups plus you know living in Mumbai and selling to these bureaucrats as my customers.

2:08 but that journey ended for me in in about 2012 and then I came to America for grad school and I did an MBA and a MS in environmental sciences at Stanford. which is where I got into venture. Started working at Kleiner Perkins in their green growth fund. Decided I really liked investing. I didn't love the like growth equity later stage version of it that I was doing there. I was more interested in doing the kind of stuff I did with my parents which is like earlier stage kind of series A seed stage series B type questions you tackle. and I joined emergence in 2014 and I've had an incredible amount of luck and fortune since I've been here. the very first deal I was able to lead diligence on was our investment in Zoom in 2014 which is an amazing life-changing experience for for me and for so many people.

2:55 and I just feel a lot of gratitude. So it's been been 12 years. I feel grateful to do this job, to work with my partners, and to have this conversation with you. >> Well, you are in the perfect seat to be on the early stages of AI and specifically AI native services, professional services. And when I saw the playbook that you emergence published, I knew I need to have you on the podcast. But in the first three paragraphs, something jumped out of me. And I'm not a gotcha type podcaster, but something jumped out at me that I want you to kind of share with our audience. And that is you talked about how critical deep domain experiences for an AI native services company. Makes total sense to me, but then you also said it doesn't necessarily need to come from the founders. Can you expand upon that?

3:43 >> For sure. and maybe I'll take one step back before we dive into that just define what native service is and then explain why domain expertise is important and I'll explain why it doesn't have to come from the founders. So an AI native service is a vendor that delivers an outcome and they do it generally with some combination of AI with a human wrapper on top. So this looks like an AI native accounting firm, an AI native insurance broker, someone that is delivering a service but doing so primarily with AI. The reason why domain is so domain expertise is so important in this era or in this business model is that in this business model in contrast to SAS you are not selling a product you are selling yourself and you are sell your own ability to deliver an outcome. So when you when you are talking to a potential customer you're not saying hey try my product and see if you like it. You're saying hey trust me to deliver on your insurance brokerage on your fund accounting on your etc etc etc. So you need the credibility on your team to say like yes I am someone that can be trusted and I've got AI you know supporting that below. so as a result domain expertise is very important for the go to market process. that being said in my experience so far and I think we've invested in seven AI native services businesses thus far and hopefully many more going forward. not all of the teams that we've backed that have been successful have come from the domain. I would say over half do have domain expertise but there are a few that where the founders themselves did not start with domain expertise but they did two critical things. The first thing they did is they went as native as humanly possible on the service having not actually performed the service before. the second thing they did was hire senior respected domain experts very soon after starting the company so that they had that credibility. So I can give you a specific example. we're invested in a company called Handover Park which is a fund administrator and you can think of that effectively as an accounting firm for private equity firms, venture capital firms, etc. It's a very specific type of accounting that is necessary for our business. And it's a large very large industry you know performed primarily by legacy services providers which are human-based. the founders Chris and Nick who started this company did not come from fund accounting. However, before they started the company, they spent time with 150 CFOs of private equity and venture capital firms, which is an order of magnitude more than you typically see a SAS founder spend with prospective customers before they start their company. So, just crazy amount of domain expertise. And frankly, I was a little skeptical when I first started talking to Chris because he wasn't a fund admin, Chris the CEO. And so, I started asking him very kind of minutia questions about fund accounting just to see to test his knowledge. And that guy is so much deeper on the on this space than I am and I've been in venture for you know 12 years. the second thing that Hanover Park did is they hired senior fund administrators from legacy providers that had a respected track record so that when they came to their customers to their prospects they said hey we've got this fancy AI I'll show you a demo of it but don't worry we've also got these people that you know and trust that are sitting on top of it that are guaranteeing that it's correct. That's an a really good example and case study.

7:04 Something else that you said early on in your playbook was the importance of hiring a product leader leader sooner than you think is required. Tell me why that's so important in a services company. >> One of the biggest traps in building an AI native services business is over relying on the human service. which is an obvious statement, but I I can't stress it enough. And and this is just so common where companies set out and say, "Okay, I'm going to build an AI native law firm, an AI native, you know, whatever the service is, and they start selling the product and the customer, they get a lot of customer demand because if you promise to sell something faster, better, cheaper, and there's already a market for it, which by definition in these services businesses there is, you're going to get a lot of market pull and then you got to staff up to serve that pull and you may get distracted from actually building the product itself. And the product itself is the AI that is delivering hopefully the majority of the value. And what I've seen is that if you don't hire someone whose full-time job it is to just think about the productization of this service, it will get neglected.

8:12 >> And so I've seen this I've seen this people don't think this is necessary because they're just so focused on delivering. So they they have engineers and they have, you know, the the actual service providers. and they forget that like having someone whose sole KPI is building and owning the product and the platform is really important. >> Yeah. And unlike a traditional professional services company, you are tech enabled. You're AI enabled. So that product leadership probably needs a little bit of product management skill set experience also, right?

8:41 >> Yeah. Because there's someone that needs to like the best AI services companies have a really tight loop between the doers and the builders. So the builders being the engineers that are actually you know building out the the AI product the doers being the people delivering the service the lawyer the accountant the you know insurance broker whoever it is and the PM's job the the the product manager's job is to ensure that that loop is super super tight right it's to ensure that whatever information happens you know in the service delivery gets pushed back into the building process and can be managed you know so such that the the product is is is growing and developing and and delivering more and more value over time.

9:19 Now the other thing that jumped out at me right up front because you see how quickly AI native software companies are growing so fast, you know, fastest to 100 million ever and you came up with this concept in the playbook about mirage product market fit and that the illusion of product market fit is created by revenue growth especially hyper revenue growth when it's powered by human labor rather than AI leverage. So can you tell me a little bit about that trap and what's an early warning sign that maybe an AI native services company is falling into that trap?

9:53 >> Yeah, for sure. this is very related to the previous point which is in a SAS business, you generally knew that if your company was growing quickly and you had high net dollar retention, you had product market fit. like those were it's hard to think of a situation where you didn't have product market fit if if those two things were true in AI native services that is necessary but not sufficient for product market fit because you can if you sell to the earlier point if you sell a service faster better cheaper you promise it'll be faster better cheaper than an existing service provider your customers may be delighted by the fact that like this thing is faster and it's cheaper than the service they were buying before however if you are delivering that service primarily with humans and not with AI. You're just a service provider with like a worse financing structure because you've taken expensive venture capital.

10:47 You have to ensure that AI is delivering the majority of the service to to truly have product market fit with an AI native service. Do you find when there is that especially up front you have maybe more human in the loop process for building confidence maybe reinforcement of the models etc. >> Do you find that that human labor component is pretty important up front though while you're still in the use case? Yeah, in almost all cases the the human is critical and to kickstart the the building of the service and the product, right? So, absolutely the question is how tightly is the leadership metricizing and incentivizing the creation of AI leverage. And this is I think the the thing that gets most missed where again people find themselves with crazy market pull, right? Because if you if you promise you're going to deliver a service faster, better, cheaper, it's going to get bought. That's not the challenge with most AI native services. The challenge is ensuring that you can deliver this primarily with AI and not with with with humans. So what we try to encourage our companies to do is to focus really tightly on both leading and laging lagging indicators of product market fit for AI native services. On the leading side of things, it's really important to identify a northstar product metric that you can use to understand how the AI leverage is improving the business outcome. And that's going to vary based upon whatever service you're providing, right? That's not that can't be universal because every service is different. But you know if you're a law firm it could be something around how much for an average contract for example how much human review time is necessary to to review that contract. If you are doing you know code migration it could be you know what's the average time to migrate a line of code and these are all kind of indicators of how much AI leverage you're getting which are important leading indicators of this concept. On the lagging indicator side of things, there's a couple metrics that I encourage founders to think about. The first is thinking about revenue per FTE. which is effectively like how much you know revenue are you able to capture per FTE specifically per FTE is delivering the service. And by definition, if you're able to get a lot more revenue with the same number of accountants over time and assuming your customers are happy and retaining, then you are getting AI leverage. And then last is gross margin. and gross margin is the ultimate way to metricize this.

13:22 But I would say founders often are wrong in how they're calculating gross margin. >> We're going to double click on that, but let me go to the revenue per employee first. I believe that's also going to be a great proxy for the impact on business productivity that AI is delivering to enterprises because they're not always showing, you know, I invested $10 million and I saved 40 million or I drove 80 million more in revenue. But let's go to revenue employee for a services company especially an AI native. Is there a benchmark that you can put across the entire kind of category horizontally or is it vertical by vertical Jake?

14:01 >> It's vertical by vertical because if you think about it the amount of revenue a lawyer is able to command for their service is going to be different than the amount of revenue that a claims process or insurance can command. So it's a little bit less thinking about it on an absolute basis to compare across verticals and much more thinking about it on a relative basis to two things. The first is how does the I native services AI per FTE ratio compare to the legacy providers and the second is how does the I native services revenue per FTE ratio compare over time to itself and I'd actually argue that one's even more important.

14:40 Let me let me double click because I think about SAS and you know $400,000 AR per FTE kind of a nice little median benchmark for a public SAS company. Now for private AI software companies a million to me is like the floor which is 2.5x. So are you looking at >> for that revenue per FT your employee 2.5 to 3x lift >> is that kind of the benchmark that you're using or not? >> No. So again, I don't think about it as compared to I don't think on an absolute basis. I think on a relative basis.

15:14 >> Okay. >> So the the two ways I think about it relatively are compared to whatever the legacy service was provided. And I guess to to answer your question, if you want to use that which the the two and a halfx certainly you would expect to have at least 2 and a halfx leverage on top of the legacy provider. And then again I think about it a lot and this is very much not not used enough I think in any services business is comparing that number to itself quarter over quarter.

15:39 Like I want to understand for your accounting firm, for your law firm, how much leverage are you getting out of increased leverage are you getting out of each service delivery person quarter over quarter? Because that tells me is the business actually getting more AI leverage, which ultimately is the thing that most matters and the thing that most needs to be proven in this AI native service business model. What's risky about this business model is that on a macro basis, it hasn't been proven yet, right? This has been possible for two years. So it's not like there is a bunch of you know data. This is very much an emerging business model. Hence why we like it at emergence. But the challenge is most companies aren't thinking deeply enough about getting this AI leverage because I think they're hooked on the growth drug. Right? You take venture capital and all you want to do is just grow topline as quickly as possible. The best AI native services companies I know are pausing sales periodically to ensure that delivery and productization works well. And that's something you would never do in the SAS business, right? Because you wouldn't need to because by definition SAS has a much lower delivery burden. But I'll go back to the handover park example I mentioned before. the fund administrator. we invested in the company towards the second half of last year and then in Q4 of last year Chris the CEO stopped selling completely and he he put a moratorium on sales because he wanted to do two things. He wanted to ensure that the existing customers they' sold were deployed well and the second that they could productize you know even more and make sure that the AI was was you know getting better and better. I think that type of leadership is critical to building a long-term successful AI native services business.

17:21 >> Well, you about that trend line and one of the other things that you highlight in your playbook and I've seen is a lot of AI native service companies will start pricing based upon some labor foundation, right? But you're saying that outcomebased pricing is critical in the AI native services business and to me that says you're getting even more AI leverage. Tell me about why you think this. >> Yeah. well, I'll start with the point that I think that technology over time will bend towards outcomes based pricing. Regardless if you're selling an AI native service or if you're selling some sort of software, I think that's the direction of travel. As AI can do more and more of the work, the way a customer is going to want to buy that AI is going to be less and less as a tool which is sold on a per se or a license basis or something like that or even usage and much more on a did you do the job for me type basis. and I think that is true with AI native services as well. There are some AI native services that already effectively charge on an outcomes basis like you either did the audit or you didn't and so you get paid for doing the audit or you did not.

18:32 However, there are also some AI native services businesses that have historic or services business that have historically charged on a labor basis and law being a famous example of this you know charging on a per hour basis. So my contention is that over time I think AI native services businesses will shift how they sell towards outcomes based. Some of them start there because by definition like the service was already charged on an outcomes basis. Some of them are going to have to migrate their customers away from a a labor you know based way of charging towards an outcomes based way of charging.

19:08 Have you seen any of your portfolio companies like maybe too young move from more of a a laborbased pricing model to this outcome base? And did they even do that to existing customers because often that could be a challenge? >> Yeah, great question. so yes, I have seen this. I work with a company called Prosper AI, which is an AI native service in the healthcare space. They do things like prior authorization and benefits verification work for hospitals and clinics to make sure their patients are able to be served.

19:40 They also do some scheduling and front office work as well. but these are services tasks that have historically been performed by large BPOS. and when Prosper got started they said like we should just be charging on you know successful resolution of the question like is the is the patient authorized or not? like that's actually what the customer wants to know, not how much time did I spend on the phone. Historically, this has been charged on a per minute basis, a labor-based model. And what was interesting is the customer initially said their their first big customer initially said, "No, no, I'm used to charging like per minute. Just charge me per minute." And we went back to them and said, "But that's a weird incentive for us because we can actually get the job done faster with AI than was done with a human. And so we'll actually make less money than the the like legacy provider if we continue with this model or we're incentivized to like drag these calls on longer than we actually need to. So it doesn't actually make sense.

20:38 And so what we ended up aligning on was a model where we did some portion of the calls on a per minute basis and then we did some portion of the calls on this resolution basis. So we basically did a hybrid model and it was a way of getting the customer more comfortable with the outcomes based pricing because it was new to them, right? And when you're coming in and selling something that's both AI based and has a different pricing model, that's a lot of change for a customer to digest at once. And so what I've seen these successful AI native services companies do is tiptoe into outcomes based pricing. Ease into it. You know, have 10, 20, 30% of the overall pricing you charge on an outcomes basis. so that the customer gets comfortable with it. So you get comfortable with it and you understand how much risk you're willing to take on the outcome and what that means for economics. You still have the majority of your charge of your pricing and the the traditional way that that you know service was charged for. So the customer is comfortable and it's also maybe perhaps easier for you to forecast. And then over time when you do contract renewals, you and the customer can both look at the outcomes space part of it and say, "Okay, I actually understand this better and this makes some sense.

21:45 Let's let's shift 50% of the contract to outcomes space." I think over time you'll see it go to 100%. You would think if there's some level of uncertainty that outcomesbased pricing also needs to provide some economic benefit i.e. it seems cheaper versus more expensive. Is that accurate? >> It should be cheaper. Like ultimately again the the value prop for an service is it has to be some combination of faster better andor cheaper. I think that the the ultimate goal you know for for most of these services businesses is to do the first two more than the latter right. you would like to say, I can do it faster, better, >> and I'll charge you the same price, and so you're always going to want to go with me. The reality is like over time, as AI is able to do more and more of these services, there will likely be downward pressure on pricing.

22:31 which should be okay because as AI does more and more of the service, there should be upward pressure on margins. >> You know, it wasn't in the original script I put together, but now I'm thinking about that famous SAS word moat. >> Yeah. for AI native services because over time as you get more and more efficient what is the moat is it the volume of transactions and the learning that you have >> yeah it's a great question I think ultimately the the moat for the best AI native services businesses will be twofold one will be the moat the same moat that exists for legacy services businesses and the second will be one that's unique to AI native services on the one that is that is common with their legacy predecessors it's brand like if you think about like why Mckenzie you know retains its value it's primarily a brand based you know value right it's like oh I trust McKenzie McKenzie did this work like I know who they are and I think the same will be true with whoever replaces McKenzie from an AI native services perspective it'll be a trusted brand and I think that will be the the the importance of trust and brand will become even more important in the AI era when you have like you know millions of agents billions of agents running around the internet you're going to want some vendor that you trust whose brand new trust that says like I guarantee I guarantee that whatever the agents that you know I have built and are working for you do I guarantee it will go well that like there it will be safe and you will get whatever outcome you need and so that will make the importance of brands actually higher and then the moat that I think will become new you know that's new in the AI native services era but I think will become increasingly important is data and I'll give you an example there we're invested in an AI native insurance broker called Harper. And what's interesting about their business is because they have AI, they're doing a much much higher volume of of premium placement than a comparably sized insurance brokerage. And when you do really high volume at relatively small human scale, you get a ton of data on which risk should be matched with which carrier, right? What type of carrier wants this type of risk right now? And that knowledge allows you to do the insurance brokerage much faster and get to a better outcome for both parties if you have that breadth of that data. And so that like they're able to outperform and deliver a better, faster, cheaper service than insurance brokerages that are 10 times their size because they actually have much more data than that than that legacy brokerage.

25:05 And they're of course using it in a structured way and they've built AI around it to utilize it. really good example and it doesn't require your most seasoned human resource to be on every call because sometimes that junior person wouldn't know which carrier for which risk profile. >> That's the thing that it's impossible for a human to know as well as an AI, right? The AI is just going to always outperform a human on knowledge like that.

25:26 I would I would lose my reputation with our finance audience because about 50% of our listeners are CFOs, VPs of finance and if I didn't ask this question because it was in your playbook and it was the importance of categorizing your expenses as an AI native service company correctly between COGS and OPEX operating expenses. Yeah. >> So tell me what you've seen there. Where where AI native services companies actually misallocating cost? Unfortunately, it we're still kind of in the wild west era of gross margins for AI native services businesses, which like I don't think that there's intentional deception here. I think it's just we're still in this foggy era of business model creation where it hasn't been defined. The how do you calculate a SAS gross margin? It's pretty straightforward, right? like people understand that you put you know support in there and you put some success and then you put you know your AWS bill and whatever that kind and then you got your gross margin for for SAS for AI native services it can be a little more complicated but at a high level you know the in big animal pictures the way we think about it is you have to allocate whatever human labor was used to deliver that service to cogs.

26:37 I think one of the things I see unfortunately happen a lot is companies say, "Well, you know, the majority of these humans were really R&D. They're really helping us figure out how to make the AI better." But then you double click and you're like, "No, but these humans are actually delivering the service for now." And so it still needs to be allocated to cogs. And so, a lot of the magic here is, you know, a a self-critical and honest evaluation of how much of the doer's labor goes into, you know, the service delivery COGS line versus the OPEX R&D line.

27:10 Something else that surprisingly I've seen a few times is I've seen AI native services businesses not include inference spend in cogs. But you need to include inference spend in COGS because effectively the inference spend is the labor right like if if you are delivering you know some sort of service and it's primarily delivered by AI the like the recurrent the fee to deliver that service the cog the cost of good to deliver that service is the inference plus whatever residual human labor is necessary and so you need to have both those things in cogs >> couple of final points that I've seen and one is the initial model training on a horizontal basis, right? And then there's the customer specific training if it happens to me. I always tell our clients, customer specific training, data training, model training needs to be a >> call because it's part of delivering to a client. Do you agree with that?

28:02 >> I do. I do. I think the training of the base model is R&D. It's an opex. I think if there's customization necessary for a specific customer, it's got to be COGS. >> Yeah, totally agree. Okay. Well, we're already running up on our time. I can't believe it. anything I didn't ask you Jake about the AI native services playbook that you'd like to share with the listening audience? >> there is so much in it. I like highly recommend people go to our website mcap.com and check out the playbook.

28:25 the other thing I would say about it is this playbook has been built by talking to the people that are building these companies. And so if if you are someone that's building an animative services business and you have some other synthesis or insights on what you've learned through that process, I would love to hear about them because we're going to be updating this playbook every six months. The idea is this is an emerging business model. We are all learning and so I want to hear from the community what they agree with, what they disagree with, new ideas that should be added to the next version of it.

28:56 >> Speaking of that, I do have one last question. I I misled you and the audience. You mentioned in your playbook you're going to be launching a benchmark. >> Yes. >> Summer. Can you tell me kind of what's going to be included in that? >> For sure. Yeah. so, I'm glad you brought that up and it's very appropriate because we're partnering with Benchmark it on that effort. you know, we've been we've been running a SAS focused benchmarking exercise for the past few years and it's been very popular to understand you know, relative growth rates and spend and such across different categories of SAS companies. This year we are doing the same thing with a focus on AI native services businesses. And so the goal will be to understand things like you know the the same metrics you care about with SAS growth you know growth rate you know those types of questions but we're also going to >> I'm sorry >> gross margin >> grow and I was going to say gross margin and we're also going to double click on things that are very much more specific in AI native services. I think in SAS land gross margin was always a somewhat uninteresting category because everyone relatively had similar gross margins.

29:58 in AI native services land again wild west we got to double click there double clicking on revenue per FTE those types of questions we're going to try to get some data to help give founders some you know ability to benchmark themselves. >> Jake Saber, general partner at Emergence Capital, thank you so much for being my guest today. great insights and I strongly encourage everyone to look for the emergence capital ENCAP in the AI native services playbook. It's a phenomenal. Thanks Jake.

30:27 >> Thank you Ray. It's fun. >> And to our listening audience, if you're enjoying and finding real value in these conversations, it would mean the world to us to go ahead and subscribe to the AI AI podcast on your favorite podcasting app. And heck, it makes me feel good. So give us a fivestar rating, but more importantly, let me know at Ray Reich on LinkedIn who else you'd like to see on the podcast. Thanks everyone. Thanks again, Jake.

30:50 >> Thank you.