How Zapier's CFO Automated His Entire Finance Team

0:00 I've noticed that you've made a change to how you hire and AI is a required competency. >> Yeah, we talk about it in four buckets. There's unacceptable, capable, adaptive, and transformative. >> Ryan, what's the most mission-critical finance workflow currently running on a Zap? >> Our month-end close. We've got literally any data source that you can imagine, whether it's a T&E tool or your credit cards or banking, payroll, all of those things are connected into that month-end process.

0:29 >> You've said that most strategic debates end before someone does the math. What does a good enough model actually look like before a decision gets made? >> It's easy to build a very complicated model and sort of try to solve all the problems in the model, but the reality is is there's only a few things that are going to change the outcome and cause you to take a different approach. >> What do you think is the CFO's role ultimately in M&A?

0:54 >> You got to be upfront as part of it, helping to screen the opportunities. How I would frame it is like the cost of being wrong. Put a little fear in the folks. Yeah, a little bit like, "Hey, this is cool and exciting and great, but let's talk about what could go wrong." >> Is this thing on? >> YESTERDAY'S PRICE IS NOT TODAY'S PRICE. >> [music] >> WELCOME BACK to Run the Numbers, the show where I interview the world's top CFOs and finance professionals. Today, we got a repeat guest, Ryan Raccaon of Zapier is back on the show. He's going to walk us through the AI competencies.

1:36 Competency is a hard word to spell and pronounce that he runs potential candidates through in order to get a job at this fast-growing startup. He also takes us through the CFO's role in M&A. They've done a lot of big partnership agreements which he has negotiated and they've also acquired a number of companies over the last 5 years. He's been at the center of this and uh this guy's a shrewd negotiator. And finally, we touch on the metrics that matter when scaling a company of their size.

2:02 Let's just say there are a lot of numbers you can put on a dashboard, but not all of them are that meaningful in driving impact. We talk about this and much, much more coming up next. Zapier makes you happier. Ryan, welcome back to the podcast. >> Thanks for having me. Glad to be here again. >> I think I got it right this time. >> You did. >> I've been working on it for a year, Ryan, just in preparation for this.

2:22 >> Well, you've done a good job. >> You got to give me the rundown because I had sent you an email. I'm like, "Hey dude, what are we going to talk about?" I don't know how you're managing to do all of this in in one day, but what uh what functions are you responsible for today at the company? >> As of right now, it's a pretty wide-ranging scope. So, I've got your traditional finance, accounting, and tax.

2:42 I also have biz ops. Pretty standard stuff, but then I also manage the ecosystems and channel team, which is made up of a few different parts. The data team, which is sort of a full end-to-end data team, so that includes things like data engineering, data insights, data AI/ML, the corp dev team of one, to be fair, strategy and planning, which is sort of like a chief of staff type role. Uh and then I also manage the pricing and packaging initiative cross-functionally. So, that's across design, product, engineering, finance, data, and so on.

3:10 So, putting together the list of what things we want to talk about, it was a bit of a laundry list, but uh it's it's fun. >> How does the finance guy get placed on ecosystems and channel? >> The team was sort of having trouble figuring out what influence they were having on the business. 9,000 plus integrations and applications that are connected. >> In many ways, it's been a cornerstone of our of our growth and our strategy for for many, many years. But as it sort of gone on, it's become a little bit more standardized, commonplace, and you start to actually break down what that team does, and frankly, it's a difficult thing to measure. This idea of what does a relationship with partner A actually generate? So, we've built all sorts of models around partner attribution and trying to sort of segment the tasks associated with partner A versus B and everything. Candidly, even when you do that and you start to look at, well, did that co-marketing campaign drive incremental value beyond what we expected or did that new in-bed that we placed with the with the company drive any sort of top of funnel traction or what? It gets messy and it's difficult to sort of really understand whether you should be placing more bets or less. And so, that team had sort of been a little bit rudderless for a while. Candidly, they said, you know, what would be helpful here is some financial rigor, a little bit of just measure what matters, set us straight on the things that are most important and reprioritize. So, we I spent the first few months just really working with that team to kind of give me the list of everything that you're doing and let's go through and just get rid of the stuff that we don't think adds value and laser in on the things that are or at least measure them much better.

4:38 >> I'd imagine it helps you with the other areas of the business that you work on because you're at the nexus of how revenue gets created. >> Every partner plays a different part. I mean, obviously some more market, some more down. Some have been around forever, some are sort of the new AI native rocket ships. And in many ways you start to see the trends of like, who is sort of the new best-in-breed go-to-market sort of CRM tool or call recorder or etc. It sort of helps you as a business focus on your use cases, your verticals, like where we should be spending more time promoting, you know, externally, building content internally.

5:17 So, yeah, it does. It certainly helps you get a better perspective on what folks are doing within Zapier and how to sort of like coordinate that with people who are supporting you, aka the partners. >> And then the data side of the equation, does that mean you're managing some engineers? >> If I'm being frank, managing the folks who manage the engineers more than anything. We do something called a bar raiser for every hire that we make in the company. And that is an executive of the function has to be the final interview for every hire. Within data, that means I'm the the final interviewer. And as part of that, we go through and we look through all the interviews that happened previously, and we look for flags and gaps that folks raised as part of the interview process.

5:53 We sort of put together a package of like, "Okay, here's the things I want to double-click on." And oftentimes within the data engineering zone, there'll be some pretty technical feedback like, "Oh, I don't know if this person has had as much experience with this coding language or this sort of regression analysis." I find myself having to get educated pretty quickly on how to effectively evaluate these gaps when I'm maybe not sort of the the savviest there. It's been fun sort of understanding how these folks are building today.

6:19 >> Well, you tee me up nicely because there's poker on on your website, and I've noticed that you've made a change to how you hire, and AI is a required competency. You call that out. So, I'm hoping we can we can go through the different levels from a finance perspective. I guess it's wider than that within the company of what lens you look at people who who are inbound. >> This is something we talk about a lot publicly. Controversial? I don't I don't think so, but it is sort of a different way of approaching this. So, yeah, we talk about it in four buckets. There's unacceptable, capable, adaptive, and transformative. They're pretty self-explanatory, but maybe to kind of like give you some examples. So, unacceptable would mean that you might be aware of AI tools. You you may have dabbled in them. Maybe you've tried ChatGPT personally. You haven't really incorporated it into your work. You don't understand the potential of these tools, and you haven't really taken the time to get to know them well. And that for us is truly unacceptable. We we want folks who are pushing this bar forward.

7:14 >> I don't find that controversial personally. >> Us, either. Then you get to capable. And this bar is interesting because ultimately, what we're looking for is you understand the tools. You've used the tools. You are finding ways to incorporate it into your work, but more often than not, it is in a reactive nature. So, somebody's asked you to put together a brief or to evaluate this thing, and you've sort of gone, "Oh, I think AI could help me with that." You're starting to use them and realize where there's leverage available. It's more in this sort of reactive way. Then you go up to the next level, which we call adaptive. And here the difference is really more of that proactivity, right? It's this idea of you recognize that you can do more and build systems and accomplish things that you would have done manually without any sort of interaction or with a limited human in the loop interaction. You're building these things sort of on a regular basis and you started to sort of push that forward. And finally get to the category we call transformative. And that's where you're ultimately taking that same idea, but really doing it almost end to end across a very sort of impactful or complex process and you're starting to teach others and sort of incorporate it in other parts of the org. And so within finance, you know, taking up a piece of the sort of month end close process and automating it it would be more in the capable type bucket, recognizing that you can extend it to have it automatically working in the background via triggers and self-healing based on, you know, a reinforcement learning loop just to go into adaptive. Transformative is, oh by the way, this script, the thing that I've written, works very well for tax reconciliations as well. And for this other team that's doing some other part of the accrual process over here. Like I could build a system that does all of these together. So that's sort of how we we think about it.

8:47 >> How do you test for this? >> A few different ways. So it depends on the role, candidly. I mean, data AI engineer, like they're going to get tested as a deep AI ML modeling sort of exercise. In many ways you're going to learn whether they truly understand the power of the tools hands-on right there as part of it. With somebody more like in finance, let's say, you know, an accountant. One is that we're asking about it throughout the interview process. And as I mentioned earlier in the bar raiser, we test for the gaps that were flagged as part of the interview process, but I also come back to the AI competency again in that final interview and one more time sort of double click here. And we're asking questions to really understand what exposure they've had, what they've built, and what problems they've solved. And we're looking for real examples. Now sometimes, one thing that does happen occasionally, is folks will say, well, at my company we're not not allowed to use these tools. What we're doing there is we're We're looking for an understanding of what's possible.

9:37 So, we're sort of like, "Okay, well, say you join Zapier in this accounting analyst position, you know, what do you think, based on what you've done previously, you would like to automate? Describe that process to me. How would you do it? What tools would you use? Why would you do it? Why is that the most important thing to do versus something else?" And we're really trying to understand their thinking behind it, right? Like, this idea of can you explain that reasoning? Even in some of these skills tests, we'll ask them to put together a brief and sort of describe like maybe a larger problem, and go to market "Hey, describe this funnel and sort of analyze this sort of reconciliation, etc., etc." And we'll suggest use AI, right?

10:11 >> Yeah. >> Like, it's a tool, you should use it. But as part of that, we want to hear how they got through that. What what prompt did they use? How did they sort of create that that outcome? And so, it's really just trying to understand their thinking behind it. >> Part of that without the AI reminds me of the old like Google test case questions of how many hubcaps are there in New York City? And it's less about getting the answer right and just going through your thought process of how you got there.

10:36 >> I cannot remember the name of what those questions are. That is a very specific category of questions. >> Is it? >> It is. It is. You should go Google it. It's very interesting. It's like there's like a whole bunch of science behind it. And I love those questions, for what it's worth. >> Have you heard of a wacky one before? >> I've always gotten the example of like how many dentist offices in New York City or something like that. I do think they're really powerful questions. But the thing that I've been finding myself double clicking on most often with these sort of AI competency type questions, so you kind of go to that first level of like, "Oh, what would you do? Describe the process." But then what I really push for is when they give me an example, I kind of keep pushing and go, "Okay, but that's part of like a process, right? Like, keep keep expanding. What else is there?" The thing that I find is that there's some folks who realize that we're on the cusp of this breakthrough, and they understand that the art of possible here, and they're really thinking about these things from like, "I can not only automate this accounts receivable process, I can remove myself from the whole thing." And the reality is is that you're not going to get it right the first time, but this is like iteration where you're going and trying something and building and improving and building and improving and getting to that sort of like really incredible end-to-end outcome. The people who recognize that that's even possible are the ones that are really going to thrive even versus the one who's kind of just like, "Oh yeah, it's a good process. I'm going to automate it." So, pushing for that is important.

11:50 >> Do you have a recent favorite question that you've been asking people? >> I mean, on the AI front, exactly what I just described. It's sort of this idea of like, "Describe to me what you would build if you if you joined." On a non-AI front, the thing that I like to ask pretty often that's pretty fun is, "Tell me about a big mistake that you made recently." Ooh. And I try to frame it as at work and big and recently. Sometimes you stump people and it's a little telling when they're like, "Oh, I don't know." But most of the time you get some pretty good responses that kind of give you an indication of how they think about problems, how they own things, how they follow up, how they close loops.

12:23 It's It's a good one. >> What are you testing for there because I've gotten questions like that before that are framed as, "Tell me about a weakness that you have." And you feel like you give them a weakness that makes you look good. Like, "I just I just care too much. I always prioritize speed and it just It's really hard. Ultimately, self-awareness. It's like one of our values at Zappier is empathy, no ego, and grow through feedback, and default to transparency." There's three of them for you actually. And you get a little bit of all of those in that answer. It's like, "Can you be transparent? Can you show how you grew from something? And can you do it in a way that is not like an egocentric, 'Oh, I made this mistake, but it wasn't even a mistake and I actually was kind of right.'" Nah, that's not really a mistake. So, we're really just sort of testing, "Are you self-aware enough to recognize those types of things?"

13:05 >> Hey, thanks for listening. We'll be right back after a word from our sponsors. >> Maybe the first thought you had when you started a tech company wasn't about scenario modeling or runway management or the fine print of an S-1. I love a good S-1. That's not where the magic happens. It's more fun to build [music] something new, design your quarter zip, or network with VCs. But what if you got your financials right early, long before your investors started asking for it?

13:31 There's no limit to how far you can go and how many headaches you could stave off. Now, that's where EY comes in. EY has been a part of Silicon Valley since, well, it was just a valley. And it's helped some of the most successful names in tech go from startup to exit to mega cap. You build the next big thing, and with the teams across strategy, tax, audit, and transactions, EY [music] will help you build your next big thing right. Learn more at ey.com/techstartups.

13:57 That's ey.com/techstartups. SaaS and AI spend is heating up, and most finance and procurement teams are already searching for shade. Uber burned through its entire 2026 AI budget in 4 months. ServiceNow did the exact same thing. Your vendors are counting on you to sweat through your next renewal and pay more than you should. Cool down your SaaS and AI spend this summer with SpendHound. Through June 12, qualifying companies that sign up get $500 just for getting started. Teams using SpendHound reduce software and AI spend by up to 30% using real pricing benchmarks [music] from over 1,200 companies across 10,000 AI and SaaS vendors. So, when your next renewal hits, you're not taking the heat. You know exactly what fair pricing looks like and what you should negotiate for. SpendHound also includes on-demand procurement experts who build negotiation strategies, review contracts, and ghostwrite emails on your [music] behalf. That means you can stop getting burned by the tools you need and cut the ones that you don't before that auto-renewal torches your butt. Rated number one on G2 in SaaS spend management, SpendHound is free [music] forever for teams up to 1,000 employees, and only $10,000 per year for enterprise teams with $150,000 in [music] savings guaranteed.

15:08 Don't let your vendors enjoy your summer budget more than you do. Sign up by June 12 [music] and get $500. How's that for a hot deal? Go to spendhound.com/cj. That's spendhound.com/cj. The CFO role has evolved faster than the tools built to support it. Most finance teams are still running infrastructure designed for a job that no longer exists. The reporting, the reconciling, the close that bleeds into the next month or quarter. [music] That's not finance, that's overhead with a really crappy title. And the cost isn't just your time, it's everything your best people aren't doing while they're buried in it. Intelligent finance shouldn't multiply your output, it should eliminate the work that was never worth doing in the first place.

15:52 That's why I run Motley Media on Brex, an intelligent finance platform with AI-powered agents that capture expenses automatically, enforce policy before the spend happens, and close your books in minutes instead of weeks. 35,000 companies like OpenAI, Coinbase, and DoorDash already run on Brex. It's time to get Brex AF. Learn more at brex.com/metrics. >> If you take a step back, Ryan, and look holistically at your team, have you found that AI reduces the need for overall head count, or is it just changing the profile of what people do and who you hire for?

16:30 >> Changes the profile primarily. I can't deny that it has changed the need for growth in places. It hasn't caused us to pull back in a really strong way as you might have seen, you know, plenty of headlines across news around folks who who've done just that. For us though, it has mostly been a change of profile. That is both the folks we're looking for and how the folks internally they're being measured and how they're operating, you know. In many ways, we are looking for the type of folks that are multipliers. They're not coming in to replace a function one-to-one, but to rather to expand what that function is capable of. This isn't just a nice-to-have as AI floats anymore, it is a hard and fast filter at Zapier. One of the cases I like to share an example around, I think it's a pretty classic case to be fair, is our support team, which we have now revamped and renamed customer op because they they are doing a lot of sort of interesting stuff and maybe this will explain why. So, over the years support team has always been very forward thinking as far as how to automate, how to sort of create efficiency and leverage.

17:28 And over the last year or so, they've really done a pretty phenomenal job. Mostly relying on AI tools, blending that with capabilities within Zapier. And the result is they have been able to reduce headcount across that team. But that team hasn't just cut the headcount, they've ultimately sort of re-tooled and re-purposed those folks to go build fixes and feature requests on the integrations across those 9,000 apps, which used to be an engineering team's responsibility. What was a support team has now become this sort of customer-facing forward engineering team fixing bugs and and again shipping features for customers, driving revenue as a result of us using AI tooling and us getting that sort of like AI profile in the door to just do more with less. And so, that's sort of how we try to we try to approach it.

18:19 Ultimately, we just want to extend the the capabilities of folks with with these tools. >> Out of all the interviews I've done, that's the most positive outcome I think you could hope for in terms of upskilling existing workers. >> We always hold a high bar when it comes to hiring folks at Zapier and we know, given the right problems faced and the right set of tools, folks are capable. Like the support team being able to do that within an engineering reason an unlock and it's like, yeah, why wouldn't we do this?

18:43 >> Something that me and you were kicking around before the call is that Zapier already automated a ton of workflows before AI in an RPA fashion, remote process automation. So, where does AI actually add incremental value versus just being a more expensive layer on top of what you've maybe already built? >> Really right down the center for Zapier, determinism. Zapier, the core sort of offering, it is really good at handling structured rule-based work. It's great for things like accounting. Yeah, it's very technical, repeatable. AI obviously, the where it comes in and where it's most valuable is when it's handling these ambiguous inputs, making judgment calls, having to do the synthesis, right? And so all tell you what we find is that AI really adds value at the edges. And that is where human sort of triage or judgment used to sit. And so within accounting as an example, we've automated away our month end close years ago. Like years and years and years ago is like everything we were doing manually we got rid of.

19:40 And I've talked to a lot of folks about this and it was built on deterministic zaps, you know? Pass them through. When the invoice hits this thing, go take that action, route it here based on these rules. So on and so forth. So like that has always existed at Zapier. What's changed is if you take that same filter that I just shared is that often times you sort of had to have a lot of different paths for all the edge cases.

19:59 All that error handling and sort of interesting stuff that isn't always the same. And so that's where at Zapier what we've done is effectively replace a lot of the things that we based that were strictly deterministic. And we put sort of an AI step in the middle of it. And so you allow the sort of very transactional pieces that are very rule based. Like when the invoice gets generated, take this step. Don't use judgment. I'm going to take this step.

20:22 Only do it when the invoice gets generated. Only, you know, pass it based on it being above or below X dollars. Like that is a very binary condition. But once it gets to this step, whether it is like synthesize or catch an error or flag for review in Slack or get a human in the loop on it. Those are the things where it's going to use judgment or a coding looks like it doesn't match the GL. That's the type of stuff where we've inserted the AI step to really add that value and remove the need for this sort of like human judgment. That is our sort of rule of thumb, right? Like determinism where you can and then AI sort of more where you must.

20:57 One interesting anecdote that I'll share here at Zapier, we offer a standalone agents product. You know, you can go buy agents separately from the core Zapier workflow product. And it is a traditional agentic solution. Type in their prompts, tool call, LLM messaging, so on and so forth, model of choice, all that good stuff. But we over the last few months have really started to go through and parse the use cases. We're trying to understand what are people building with these things and what does it look like compared to what they're building with Zapier, traditional Zapier. And what we found is about 80 or 90% of what folks built was deterministic.

21:31 And only small sliver required an agentic input. >> not thinking big enough originally? Just not knowing that they could figure it out? >> I mean, the short answer is no. I mean, even if you look at an agent, most of the things that you build, it's code. It's And it's generally speaking triggered on events and follow a predictable pattern. And there's a set of of instructions where the judgment comes in. But like the before and after of of that set of instructions, I mean, in this case, eight times out of 10 is deterministic in nature. It's like, take these actions upon these things. And then when you get here, that's where I want you to sort of use a little bit more judgment.

22:06 >> It does sound like you would agree that there are still many, many cases where a simple deterministic Zap still beats an AI-driven workflow. >> 100%. I mean, any workflow where the input is really clean, the logic is fixed, absolutely. Like you mentioned earlier, AI is expensive, tokens are expensive. There's latency. There's a little bit of opaqueness of like, what did it do? How do I audit it? How do I explain this to my auditors? Like, you know, I Oh, I threw in an Excel file, it spit out this thing. It's a bit of a black box in cases. So, certainly when it meets those conditions, determinism still wins in my in my book.

22:41 >> Hey, thanks for listening. We'll be right back after a word from our sponsors. I've seen a lot of FP&A tools, and Alaf is one of the few that gave me that aha moment within minutes. I remember watching their founder, shout out to Albert, connect my NetSuite data and build me a full P&L live in minutes. Alaf is now trusted by hundreds of leading companies. I've had the CFOs from Turo, Eight Sleep, Zapier, and more on the pod, and every one of them is a huge advocate. I also just published my second annual CFO tech stack report, and Aleph has been on the podium both years, including a number one finish in the 50 to 100 million dollar segment this year.

23:16 Instead of being just another planning tool, they built a real enterprise grade data foundation for finance, implemented at startup speed with AI native workflows woven into its DNA. All your systems, ERP, CRM, HRIS, ATS, product usage and more, powering one clean governed data layer that finance can actually trust. And with AI moving as fast as it is, they're pushing even further. MCP, custom AI chatbots, AI powered variance analysis, and the list keeps growing. Try it with your own data at getaleph.com/run.

23:47 That is g e t a l e p h.com/run. Tell them CJ sent you. Here's a growth tax that nobody talks about. Every new pricing model you ship creates a nightmare for your finance team. Usage based pricing? Now you're tracking usage against commitments. Product bundles? Now you're untangling what to recognize and when for every line item. Mid-cycle upgrades? Good luck manually reallocating revenue. The pricing strategies that drive growth are the same ones that break your finance process.

24:18 RightRev turns that irony into a competitive advantage. Your product team can ship new pricing without asking finance for permission, and your sales team can close deals without worrying about downstream chaos. But I've seen too many companies where sales are celebrating a huge quarter, while finance is still trying to figure out how to recognize half of it. It's actually me. So here's a good place to start. RightRev built a free tool at calculator.rightrev.com. It scores your rev rec process, shows what's exposing you to risk, and tells you exactly where to focus before it bites you in the rear. Probably explains why your last close took so long. Check it out at calculator.rightrev.com.

24:55 All right, without the Boston accent, calculator.rightrev.com. I got news for you. The ERP category is finally getting disrupted. And if you haven't heard of Rillet yet, please pay attention. It's the AI-native ERP built specifically to replace NetSuite, and it's already won over hundreds of finance teams. Their mission is to make the zero-day close a reality, and they're actually [music] doing it. We're talking teams closing the books at 1:35 p.m. on the first day of the month.

25:21 Companies like Wind Surf, Mercor, and hundreds of others run their entire finance stack on Rillet. Revenue recognition, close management, multi-entity, native Stripe and Salesforce integrations, woo! Everything a scaling company needs. They've got 5.0 star 5.0, wow, 5.0 stars on G2. They're backed by a16z and Sequoia, heard of them, and CPA-led implementations that get you live in 45 days. That is simply unheard of in the ERP space. [music] If your books aren't running as fast as your business, check out Rillet. Book a demo at rillet.com/cj.

25:55 That is rillet.com/cj. That's me. Zappier talks a lot about drinking your own champagne. Ryan, what's the most mission-critical finance workflow currently running on a Zap or an AI-powered Zap? >> It's a difficult question to answer with one example, to be frank, because they're more like systems. They're more like Zaps that are connected to Zaps and agents with agentic steps, and it's like a big sort of ecosystem where there's the most activity is within our month-end close. Our month-end close, we've got literally any data source that you can imagine, whether it's a T&E tool or your credit cards or banking, payroll, all of those things are connected into that month-end process, ingesting automatically using webhooks, using scheduled ingestion, whatever the case, going through some sort of reconciliation process within that Zapier sort of workflow, generally speaking using that agentic step. And we've built a ton of really cool stuff here, as I was alluding to earlier, where we're flagging for mismatches, for risk, for error handling, for deltas against plan, like certain things that we want to have a human's eyes on. And then all these things are sort of getting drafted automatically. We use We use NetSuite, so it's going to NetSuite, ready to go, drafted. All All you got to do is click publish and we get a notification in Slack. So, in many ways that process when it starts on day one and we close the books on day five, like within that, there's stages of just big chunks of accruals and indirect tax and treasury entries and all these things that are just running sort of like on a schedule and just having us sort of check the box that we reviewed before we click go.

27:28 >> I really like how you describe it as an entire process of things being done because at the end of the day isn't that like our whole goal? It's not to just do a task, it's to accomplish some job to be done. That That is a a bundle of many things. >> Interesting segue on that. I gave a talk not long ago about the ROI on AI spend and I'll tell you a bit about what I shared because honestly we'll spend the next hour talking about it. Folks kept pushing on, you know, what what are the metrics? What are you What are you paying attention to? And I was using accounting as an example. So, yes, we built these types of things for month-end close, but we've also built them all of the places that accounting touches, any sort of process. So, another one as an example is within sales. So, when we're writing a invoice or a quote for a customer, one thing that happens is the sales team will often get customers who are mid-subscription and they want to renew and upsell to a higher tier. And we have to calculate the credit they get from sort of like canceling their old plan and then the cost of the of the upsell and then sort of give them that credit calculation.

28:22 That used to be a manual process. They would go and flag the salesperson, ping accounting, say, "Hey, can you run this credit calculation?" Used the AI again sort of for some of the uniqueness of the type of of contract that customer is on. Could have been annual with monthly pay or quarterly payments, annual payments, multi-year, etc. etc. So, you have to sort of incorporate that in the calculation. Automated that whole thing. Now, when the salesperson asks in Slack, it is a bot effectively that we built that answers them immediately. Almost all of the metrics we talk about today are time saved and it's funny because there's this like such a strong adverse reaction to that metric. Every CFO or operator on the planet seems to gag a little when you say time saved and I think it's kind of funny because I mean ultimately in many ways that is very important and it is a sort of a leading indicator but it is just that. It is an input ultimately what is the output. And in that exact scenario that exact example I just shared, what happened was yes, of course that accounting team saved a bunch of time and the sales team saved time they didn't have to wait for accounting to get back to them. Might have been a 24-hour turnaround in some cases. What happened though was that that sales person got to answer the customer like instantly. There's the upcharge.

29:30 And their probability of closing that deal went up like 30%. The system means we get to do more, there's a better output, it's faster and so on and so forth. But ultimately what really matters is do you have to grow the team anymore? Do you can this team go focus on higher value things? Does that enable them to go drive revenue versus be a cost center? So on and so forth. That's really the downstream sort of implication that you got to go trace the thread all the way through.

29:53 >> Sometimes you just got to give the guy the ball and let him cook. That was an amazing example. When you get the cost for AI and it lands on your desk, do you look at that more as a cost of goods sold or do you look at it as an investment in your product or an investment in your workforce? >> It's a kind of a loaded question for Zapier because as a company selling AI products, there are clearly ones that actually are cost of goods. They're they're built into the product for the end customer. For the sort of more internal usage, it is a bit of a mixed bag I would say. In some cases, it's more of an investment in the perspective of we recognize it is a productivity engine, right? It is going to create output that exceeds what we're doing today. And as a result, it has an ROI that we should measure. And we try to do that very explicitly. We look at all of the spend that we have across all of these tools. And to be clear, Zapier has all of the tools.

30:43 I mean, we got a big budget that goes towards AI spend, product facing, you know, customer facing, and internal usage. We we use them all. And as part of that, we try to measure it all. We try to look at everything from, you know, engineering code being written, QA, deployed, shipped, the quality of that code. Similarly, within accounting, looking at, you know, the number of transactions, the number of errors that we're getting on quotes that we produce or or invoices that we generate, then go-to-market speed to lead, how quickly we're getting into the sales team's hands. So, we try to create little chunks of metrics that are associated with the AI usage and tools to really come back and say, "Well, when this sales person has access to this tool and it costs us this much in tokens, what's the outcome that they're driving, you know, in that speed to lead kind of case?" And we try to pair that up to say this is an investment that's justified. But, it is uh it is a tricky thing to get right, and it is this blend of some of it is more of this sort of infrastructure, but ultimately we want to treat it as much as we can as an investment and sort of feel confident that we're getting a a positive return on that investment.

31:47 >> I was going to lead you into M&A and partnerships next, but sometimes when I get dog with a bone, let's let's talk metrics and measurement. So, when a metric moves in the wrong direction, what's your first instinct? Do you assume it's a performance issue or a measurement issue? >> I think the answer to that question depends on the metric. If that metric is incredibly well defined and solid and it's industry standard and you're just like, "I know this thing is grounded in very strong data," I'm going to go dig into performance. But, there's uncertainty on the edges of metrics, even ones that are pretty well defined.

32:17 And so, oftentimes we do go to measurement first to really just try to get an understanding of what is happening. Is it performance or is there something upstream of this that's causing this this issue? Zapier, as an organization, has been product-led growth, self-serve. Its entire DNA is based on that. Only in the last few years have we moved into this sort of upmarket transition, selling to like tail larger organizations traditional sales. When you're establishing a new program, funnel definition and how things are qualified and treated and attributed, it's the wild west. I mean, everybody's got a different opinion. I have seen many cases where we have finance data revops trying to define things differently. Qualified ops go down period over period and you immediately go, "Oh my god, did a campaign shut off? Did we have a rep that dropped the ball? Did we have something in some channel that's gone awry?"

33:05 And then double click a few times and you're like a lead you're like, "Well, leads are the same. I'm guessing this is a qualification definition change." And then you go ask five people and you find out yeah, that's what happened. There's a bit of a sort of take a breath when you when you see something change, depending on the metric and how nascent it is or how well established it is and how well everybody understands it. Those things can help point to maybe don't go sending people down a wild sort of goose chase. Take a breath, try to understand what's happening, you know, start to look upstream of that to see if you can find what's going on.

33:33 >> That's a good process because I'm sure you've had situations where teams tried to fix a performance metric when the real issue was just how it was defined originally. >> We have, you know, within partners so ecosystem and channel. When I came over to that team, we had to define a whole bunch of metrics. They were metrics that we had never really had before. So, it was sort of net new, let's let's build this thing. And so, one of them was around partner ARR and attributing what our channel program was driving in terms of partner influenced ARR.

34:01 >> We used to call it influenced revenue at the last place I was at. >> Very similar. So, we built it, we called it, we set a target, everybody's like, yep, locked hands, we're going. Let's do this thing. Then we got a quarter or so in and we were way way off. And we're like, "Oh no, what happened?" And so, we start chasing this and we're like, "Okay, is it because of a certain set of partners that is like not performing?" Or let's go launch some demand gen campaigns. We think maybe top of funnel isn't as strong as we thought it was. Let's like we're chasing this thing. And literally trying to scramble to find what what what went wrong. As doing that, me and sort of the finance team stepping back and going, "Okay, while you're doing that, let us just take a closer look at this." And we went back and we realized that our modeling assumed the goal that we'd set was based on sort of them touching any customer at all versus net new. And the reporting that we were doing, the metric that we defined, was just based on that new. This is like production, how many customers you brought in through these partners. In reality, what we found was a lot of the influence revenue, you know, as you as you put it, was coming from existing customers where they either A already had a relationship or B established relationship through some of their network, and they were actually upselling these customers into higher tiers and helping them build new sets of automations and workflows. What we also found in that process was they were doing a pretty good job of it. And and actually we were finding sort of like correlated net revenue retention improvement and all sorts of great outcomes from their interactions with these existing customers versus some of these these sort of like net new production customers. Going and chasing and fixing the wrong thing sort of burns credibility in the metric, in the in the process, in the goal setting thing. Like we could have thrown a bunch of resources at expensive outbound and then realized later like, "Oh, this actually not the thing that's most valuable to us." Created a process in which we started doing some lead passing from our existing sort of customer base to solution partners. And that program served pretty well so far. So, yeah, it's a good example of a good intention metric sort of gone awry because the measurement wasn't quite clear.

35:56 >> You've said that most strategic debates end before someone does the math. What does a good enough model actually look like before a decision gets made? >> It's easy to build a very complicated model and sort of try to solve all the problems in the model, but the reality is is there's only a few things that are going to change the outcome and cause you to take a different approach. And so, knowing which of those variables matter most and what has to be true for the decision to change is ultimately sort of where you should draw the line as opposed to building these very full end-to-end sensitivity, you know, scenario And obviously we play a very big part of that as far as being that first step in helping teams understand what that model should look like. One of the things that I think is an interesting AI use case, I personally work on it. So here you go, right? Got my hands on the tools building myself.

36:45 >> A man of the people. >> I have to admit, I feel very left behind in some cases. So I'm trying my best to keep up. But we had gone through this exercise helping to build these sort of minimum viable models. And we met with all of these teams who are driving initiatives for 2026. So I told them to say, "Okay, explain what you're doing. Tell us what you're trying to influence." I've been doing this at Zapier now for for over 7 years, and this problem hasn't gotten any easier.

37:09 It's a difficult thing to get sort of a product team, engineers, folks who are building to really ground it all the way back to sort of like the one or two metrics that really matter. And so we went through the whole process, we built all these thesis docs, we sort of said, "Okay, here's the metrics you care about, here's the leading indicators, here's the output, here's how we're going to track it, and goal it, so on and so forth." I took all of those documents across all of those initiatives, and I fed it into an AI tool, and I gave it access to our strategy docs, our budget, how we frame investment lanes at the at the company.

37:38 I want you to help me build what is what I'm going to call the initiative sizing coach. And ultimately the idea is take all these types of questions that we had to ask. I gave it the transcripts from our calls using Granola. I want you to build a coach that ultimately I can give to anybody on the team, and it will walk them through like, "Tell me what you're trying to do. I'm trying to get teams to you know, click on this button more often, right?" "Why do you want to do that?" "Well, because if they do this, they expand." And so on and so forth. It will walk through and ask them sort of probing questions that go a layer deeper, a layer deeper, a layer deeper.

38:07 When they give you some like way lagging indicator, it'll go, "No, no, no, no, no. Come way upstream. Tell me like, what's the very first thing the customer is going to see? Why does this matter?" And then it asks for data inputs, and you sort of give it all the context that you have. And then at the end of it, it creates the thesis doc, and it also creates a prompt that you can load into whatever tool you want, whether it's like cloud for excel or you know any other ones, and it will sort of build a very rough draft V1 model to like at least give you little assumptions and sort of output. It's okay. I've sort of exported the the thinking, how do you know a model is good enough before you start getting into decision making to know like what the variables are that matters. It's like that's the process.

38:45 There's really only one or two variables that actually matter here. Let's just like build a model around that. >> We were talking to the VP of data and analytics, Chris Biden thing, at superhuman the other day, and he was he was saying, "Look, when you run a data and analytics team, they come to you basically with the question or the ask that's not always solving the ultimate thing that they're looking for, and it's your job to ask those probing questions." Henry Ford used to say that if you ask consumers what they wanted, they would have asked for faster horses.

39:13 Steve Jobs would say, "You don't ask the customer what product they want, you ask them what experience that they're trying to get to." >> No discredit to the product and engineering folks who are building amazing products, but there's a reason why we have a finance department and they build the the stuff, you know. I can't go in there and ship a whole bunch of code and deliver all these great things for customers. The thinking is different, and and I will tell you that in that process, many times in which we would get answers that were very clearly outcomes, and it's like you could do a thousand different things differently and still get this outcome. We're going to have no idea whether what you did was successful if this is the only thing you're looking at. Like we got to keep double clicking until we get to that.

39:50 So, to your point, faster horses, not a good answer. That is ultimately our job to sort of get to the bottom of. >> You mentioned outcomes driven, you've also talked about escalating on magnitude, not certainty. How do you define magnitude in a metrics context? >> It's all relative. >> Relative to the ultimate business impact, relative to what? >> Relative to the business impact and what's like the risk or opportunity. So, maybe as an example, we have a metric that looks at something very simple, just like failed payments.

40:17 >> That should be simple, binary, right? >> And you put sort of thresholds, high and low, or whatever the case is, and and we'll alert us, like a good eventing system should, if it goes above a certain threshold. And we had an alert come through, and it said, "Oh, this is trending a little bit high." But in reality, when you looked at it, it was like, "Ah, it was just barely high. It was almost within noise sort of thing."

40:36 And so, in that case, that metric was doing its job in the sense that like this small change was a big enough magnitude to flag us. Candidly, when we saw it, we didn't pay much attention. Another sort of day or so goes by, and it pops up again, and when we dug in, we had found that some code that we had shipped had removed payment methods from a handful of accounts. Didn't impact the customer or the customer experience or anything in any way, but it would impact the charge that would come on their next renewal. The magnitude of the risk of what that introduced was not at all visible in that metric. What that caused us to do was to go, "Wait a second. It's a good signal, like it's a good way to sort of pay attention to the to sort of some of the leading indicators, but there is something happening much upstream of this is probably a better one. Let's go look at all accounts and the attached payment methods, not at the rate at which they're sort of being processed, but rather the distribution of ones that are missing." And it was like, "Oh, wait a second. We deleted a handful of these. We got to go fix this right now." And all of a sudden, the magnitude became, you know, much more severe. In the case of like to your comment, the impact or the ultimate outcome or business impact was significant in the sense that we would have had customers trying to pay us that couldn't. And so, it's sort of this interesting dynamic where it's it's relative. In that case, that magnitude was properly defined, but not indicative of a problem.

41:54 >> Nearing towards a close, I do want to touch on M&A and partnerships for a bit, because Zapier thrives on partnerships. You have an agreement with, I think I wrote it down, 9,000, more than 9,000 companies. Do you have any good stories about negotiating any of those partnerships? >> Yes, it is 9,000 plus, and it's a lot, because you ultimately end up with with 9,000 certainly you can't give all of them any real attention. You have to sort of decide where you want to spend your time. And it kind of speaks to our last conversation that we had way back when on this like two-way door concept.

42:24 We had a partner who was growing in popularity and we had a handful of enterprise deals that were using this partner, but they weren't on the Zapier platform. Generally speaking, what we do is we ask the partners to build the integration. We say, "Hey, look, you got to build this. You got to maintain it, keep it up to date." There's certainly a lot of advantage it's good for us as well. Like it's a shared sort of ecosystem. And in this case, they didn't have a live integration. They didn't they we ask when they do build it you have to sign a developer agreement that sort of conforms to our terms and such. And we'd ask them and they kind of were unresponsive. Like, "Yeah." Didn't didn't really give us much attention.

42:57 But we had a bunch of enterprise customers come through and they were like, "Wait, we use this partner. We'd like it to be, you know, available on Zapier." And so we do have the capability if it's public API to build the integration privately for a customer. Long as it's a public API, we can do that no problem. We do it all the time. And so we built a few. We had that sort of continued to escalate with more enterprise customers coming asking for this. We kept building it. And at some point going, "Should we go ask the partner like, hey, could you go do this?

43:24 Got more people asking. This is a good thing for you and again for us." We sort of said, "Well, I don't know that we should because the reality is all we're going to do is open the door to a no when in reality the benefit is there both sides. So we kind of continued to build a few more. And then once we got to 10 or so customers with another 50 asking for it, I just emailed the CEO and said, "I've got 10 customers using it already. I got 50 more who want to use it. We built it. I can toggle a flip a switch on our side and it goes public.

43:49 You cool with that?" And he was like, "Yeah, go ahead." Not the whole sort of know when to ask for permission versus forgiveness, but recognize when you can sort of back up the request with a little bit of evidence to go along with it. That's sort of I think is an easy way to negotiate a partnership. >> You've also done at least five acquisitions within the last five years I'm counting, right? How do you think about the M&A side in doing the math in a process? What what does that look like and what's the minimum model you need to feel comfortable?

44:14 >> Yeah, not too dissimilar to the conversation we just had on sort of trying to be a little bit narrow in how you define the the sort of the thesis and build the model. Ultimately, we're going to look at the things that everybody's going to look at the the core the core ones, you know, growth rate trajectory, the components of that growth rate, cohorting that out, customer cube, understanding retention and how that's improving with newer cohorts, comparing that to comps in the market, trying to come up with the sort of range of what we think is a reasonable price, looking at the structure and in ways in which that aligns incentive and what the market would demand in terms of how we would structure it from a cash versus stock perspective. We've got a few a lot of smaller ones, a lot of aqua hires, explored some larger ones, and as you would expect, you see patterns in the data that will lead to a whole bunch of other questions. And we were chatting with Target and we saw decelerating growth in a segment and we saw this interesting pattern in the funnel where activation was starting to drop off. You know, that obviously prompted some questions and it sort of kind of kind of came back and said, "What's going on?"

45:13 In that scenario, I got a story around, "Well, we made these price changes and we recognized after the fact they were the wrong ones and now we have to undo this." And so it kind of comes back to the same conversation around like, "Is it the metric? Is it the measurement? Did something go awry or the way that you're attributing something?" So you kind of have to double click through those things and get to the answer. In this case, we were able to isolate those cohorts of sort of pre post price change and look at, "Okay, let's look at the retention patterns of the ones who came in then and after and then compare." And you could clearly see and delineate and say, "Okay, you're right, if this hadn't been the case, you know, the funnel would have looked like this, activation would have looked like that. If you're going to reverse these changes to this date, we can put that back in the model and assume that's the case." But ultimately, yeah, it's try to find the right place to draw the line, understand these these changes that can shift your perspective on on the potential and the sort of like management case versus your own internal case. But yeah, we stick to the the standard ones and and try to build something that's mostly based on those drivers. Again, that would change your mind if it wasn't true.

46:16 >> Would change your mind if it wasn't true. It's so well encapsulated. So, let's say you do get comfortable with the performance data. How do you think about the balance of cash versus stock in a deal? >> If the business is being acquired to run completely stand-alone, if the business is IP only, if it's full integration, those things might not sound like reasons for cash versus stock consideration, but the reality is is the way in which you are trying to align incentives changes with each of those.

46:44 Obviously, stock is a natural incentive and it might serve the purpose of that retention versus let's say cash plus a designated retention program as part of the acquisition. So, there's a bunch of different ways you can sort of square it up. We'll kind of look at that alignment and trying to get incentives in the same place. >> Candidly, it comes down to cost of capital. We have been profitable for entire history. We don't have a lot of the prep shares that would sort of force your hand one way or the other. And so, we do have the the the benefit of of saying, well, we can fund a good chunk of this or all of this with off the balance sheet.

47:15 >> Yep. Or we can go raise or we can go get traditional or hybrid debt. Kind of starting with that incentive alignment piece, we're looking at what the markets are allowing for, you know, today. Growth markets are pretty dry, right? You know, unless you're AI native on a 100% year-over-year tear, it's looking at the cost of traditional debt and really comparing that weighted average cost of capital to, oh, what if we fund this internally? And then factoring in the the different sort of dilution considerations, so on and so forth.

47:42 That's ultimately how we try to balance it. The other thing that's interesting is you also have to consider, well, how do we anchor our price in that deal? And what comps do we have? And again, Zapier unique position, we don't have financing rounds to point out to say, oh, well, look, right here, 6 months ago we were at this. This is what you're going to get." We kind of have to go, "Well, there's a 409A value. It represents fair market, and there's these other things."

48:04 And makes it a lot of tricky. >> Maybe a good question to end on here, because this has been a blast. What do you think is the CFO's role ultimately in M&A? >> One of the most important parts is to show up early in the process. >> Okay. >> Don't wait to be like the diligence person. It's very easy, founder-like company, you know, product-led company, we really care about customer sort of like uh success and everything. It's easy to kind of get attached to an idea and really believe in the vision and get emotionally sort of committed. And then you dig into the financials and you dig into the reality of it, and it starts to turn the other way. And kind of reversing that, "Okay, slow down." is harder when you're later in the process.

48:43 And so, you got to be up front as part of it. Helping to screen the opportunities. How I would frame it is like the cost of being wrong. Put a little fear into folks. Yeah, a little bit like, "Hey, this is cool and exciting and great, but let's talk about what could go wrong." Both internally, cost of distraction, and externally, this business goes the opposite direction or things don't trend the way that we expect, or the synergies aren't as great as we're hoping for. Try to be early, try to be part of that screening process. Don't just wait to the very end and say, "Hey, we want to buy this.

49:10 Here's what we're thinking. Help us get a price and do some diligence." >> It's always so fun to catch up with you. I really appreciate you doing this. >> Yeah, it's been fun. Thanks, EJ. >> Run the Numbers is a Mostly Media production. Yelling in intro by Fat Joe. Artwork by Meg D'Alessandro. Show is executive produced by Ben Hillman. Nothing said on this podcast is intended to be business or investment advice. It's the sole opinion of me, a guy [music] who feeds his dog way too much ice cream and has a history of net operating losses, lol. If you like this podcast, hit subscribe and give us five stars. It'll take like 2 seconds, and our algorithm overlords love it. Drink water, call your mom, and have a great day.

49:46 >> Peace.