Name: Finance Forward 2025 - Rethinking Scale: The AI Finance Playbook
Uploaded: 2026-01-22T03:56:05.251Z
Duration: 40 min 8 s
Description: Finance Forward 2025 - Rethinking Scale: The AI Finance Playbook

Transcript for "Finance Forward 2025 - Rethinking Scale: The AI Finance Playbook": So, one fun fact, this is the Unicorn Club. They're all unicorns. They're doing something right. And, we're gonna talk about AI. So I know we just talked about it a little bit. We're talking about how we're using AI today. Maybe as a quick primer, if everybody just gives a quick intro of themselves, what your company does, I think that'd be helpful. And then, anything you wanna share from, like, a size perspective so when the audience hears your answers that might might be helpful, like head count or something. Sure. Go for it. Martin. Right. Martin Abramson, CFO of Vercel. We are an infrastructure company for web applications and AI. About 650 employees today and about 250,000,000 of ARR. I'm Kelly. I'm the global controller at Speak. At Speak, we put fluency in the palm of your hand by delivering a fluency partner that is AI. Our primary customer base is in APAC, EMEA, and South America, as you can imagine. We have about a 150 employees today. Hi. I'm Fraser, ops and finance leader at Posthog. We're a all in one developer tool for building products, supporting customers, and managing your customer data in one place. We're about a 170 people at the moment, about 45,000,000 ARR. And, yeah, thanks for having us. Amazing. Well, I was talking to Fraser this morning, and he said he asked the team about, like, how they're using AI. And then he had a 150 replies on Slack. So I gotta ask. I'm just gonna go with you first on that one, and then all of you please fill in. What were the themes that you heard, like, just internally across post hoc? Where were you what was what stood out? Yeah. So I think a lot of people have quite strong opinions in postdoc on everything. But on AI specifically, a lot of it is about leading into one of our big kind of values is talent density and trying to increase that. So trying to make sure that our employees who we think are excellent can become even better. So providing them with the tools, a lot of our engineers would use code code or something like that. That's kind of very basic example, but we're trying to make people more efficient. We kind of think about our product as helping developers do what they do best, which is building products. So we want them we think about an engineer's day as log again, checking support tickets, looking up dashboards, figuring out what's going on, and by about 02:00, coding. Now we've hired these engineers to code, so we wanna bring that time shift forward. So using AI inside postdoc, we're able to reduce all that by having agents go through all that data through their dashboards, checking what is going on. And then eventually using their favorite tool like Cloud Code or whatever they use to prepare a pull request for them to check and then be able to release that. So it's tightly controlled by excellent people, but it's making the best people move even faster. Kelly? Yep. So we're a nine year old series c company. And for up until just a few weeks ago, the accounting team has been only a team of two. And so always been, what can we do to leverage our talent density as much as possible? So the way we thought about application of AI across our kind of tech stack is, first, a deep investment in the tech stack itself. Are the tools you're using AI first, AI native, AI forward. Because if you're not building that tech stack, that layer, that foundational layer that is, you know, enabling and ingesting and delivering the AI technology, then you're always going to be fighting against that. So that's step one. Step two is that we're leveraging agents as much as possible. I think right now agents are a bit prescriptive and bit rules based, but I think that that technology is really ripe for further investment. And then we think about it from a search perspective. So this is a little bit broader, but how can we leverage some of these LLMs to really do some basic research or even deep research when we think about our technical accounting memos or how we're going to prepare and internally use software work paper. And then we think about it from insights. So how can we leverage I think Campfire's Lam is a perfect example. How can we leverage our existing data to develop insights about our business in a really meaningful way and one that we can do in a fraction of the time that we used to be able to? That's how we're thinking about it. For us, it's been very much a journey. So we've always been pretty underinvested in G and A. Like, we tend to focus most of our headcount and resourcing to to engineering. Sales kinda get second pick, and G and A very much gets last pick. So this has been out of necessity. And it started out, you know, what I call, you know, pretty simplistic, deterministic stuff like, you know, GL coding, expense report approval, stuff that has, like, one definitive answer, like, is this travel booking in policy or not? So that was kind of wave one. There are plenty of tools that can help you do this today. And now we're pushing through the envelope more and more on this and trying to get into what we call nondeterministic complex use cases. So think, they're not stuff that have, like, one definitive answer. So trying to think through, like, why are sign ups up in the last thirty days? Right? Like, trying to do variance analysis of, like, what exactly is going on here? Like, why didn't churn increase? Hopefully, it didn't. But, like, what if what is happening to ARPU? Like, when you can't tell the machine to go say like, we use Ironclad as an example. You can easily query Ironclad and ask for complex stuff like, how many of my contracts have net 30, net 45, net 60? And it has to do some thinking, but at least there's, like, an answer to it. The challenging part now that we're working on even having, like, our own resell agents are these much more complicated use cases, like, trying to really understand, like, the business insights. And this is where we think we can if we can do that really, really well, then we can make, like, our Stratfin and FNA teams meaningfully more efficient with their time so that we don't need to go hire, like, an army of people in order to extract that insight. Like, the insight is there. All the numbers are there. All these systems now connect with each other, but you're still dependent on humans to try to figure out, like, the why. And if what we increasingly find is, like, you know, the expectations keep getting higher and higher. So, like, the asks and, like, the, you know, the why why why, will take more and more and more time. So if you can use AI for some of these workloads, I think you can add a tremendous amount of value. I think we're all familiar with G and A being last in terms of one quick story. I was at a public company. My boss was a CFO. And we had bring your own beer happy hour for the finance team. And legal was next. The section next to us was doing Omakase that night. So there was this clear, like, disparity in spend. So very familiar, Martin. In terms of, like, buzzwords, I know there's like AI maybe is the most hyped, over hyped, you know, agents, a lot of hype right now. But how do we think about like going from buzzword to the team into like actually using it and within finance or if you wanna talk more across the org, I think many of us are like, how do we shift our team into into using AI? Maybe it's just because, like, we're forced to because we don't have more headcount, like the the traditional austerity of the finance team. Maybe it's we feel AI is far enough along, or maybe it's just like we're just pushing the team or maybe someone's pulling the team along. Like, how do we how do we think about our AI journey from that regard? Yeah, I think for post hoc anyway, were quite considered with our approach to AI. We've actually taken a step back and not sort of pushed it immediately. And we're now releasing post hoc AI and having more invested a lot more time in it now that we know what we think the use cases, which I explained is trying to help people improve faster. And we've asked every team in the company to think like that. And for finance, we're actually bringing our finance team in house. So our accounting in house from being external to internal and AI is actually at the heart of that. So how do we leverage AI to make sure we can follow the principles of talent density to have excellent people that are able to own wider areas and we call them functional tasks rather than core things that are going to improve Postdoc, make Postdoc a better company. That's what we want our employees to be thinking about. We want the team members to be thinking about what makes things better at Postdoc tomorrow than it did today rather than a task that just needs to get done and can't be kind of done in a, I guess, creative way and is more like a functional task. So trying to make sure those functional tasks are reduced as much as possible with AI. Campfire is a big reason and using Campfire will help us reduce a lot of those functional tasks whilst also allowing senior people in the business to come in and ask questions of our financial data and our financial information, and they can then go and start using that to pull different levers in the company. So it's about reducing the number of people that we eventually need and making the people that we bring in to take over this in house far more efficient and far more productive, ultimately making Postalc a better company, not just that individual task or the finance team or the operations team or G and A. So it's about making the business improve rather than necessarily get that individual task perfect. I think anyone that's worked in core accounting in the last ten years knows how manual and time intensive so many processes are and the old school, I guess you'd call it, UI of some of the incumbents. And I think everyone was ready and ripe for some disruption in the technology that we're using because so much of your day, going back to talking about your engineers, was spent doing manual tasks or tasks that weren't adding a lot of value. So I think the moment that this technology became available through our vendor partners primarily and bubbled up to our teams was everyone was really, really excited about the adoption. I think that the two biggest misconceptions right now are, one, around agents. I just think that there's a lot of hype there, I think the technology is just not quite as nimble as it could be. And I'm excited to see that continue to evolve. And then the other thing is people think, oh, AI is here and now everything's better, and I'm working two hours a day now. I think that with AI, there's tons and tons and tons of little micro changes you make to your processes, your systems, your tooling. And all of those changes start to bubble up and really impact your processes, your workflows, and therefore your time. But it's all those micro changes that you need to consistently be making and investing in upfront. I think a perfect example of that is just like talk about a staff accountant, like going back to cash matching. Right? If you can invest a little bit of time upfront to develop the rules for your cash matching, then every single month, all those transactions are auto matching. So it's a good example of, like, a small investment upfront that's gonna pay dividends long term. And the more you can build on top of those and on top of those, the more all of your processes become more efficient. Great points. I think the only one I would add for us is forecasting. So I'm sure they're the same problem of you know, we have an enterprise sales team, they somehow keep sandbagging. I can never get good answers for that. And I love them. Chris, if you're listening, you're doing great. But it's really that partnership has been really, really difficult because they're not sort of willing to commit to numbers. They always come in low, and then, you know, we start managing, hey, with the board. Hey. We think this quarter might be light. And then quarter comes, and, like, we exceed, and you sort of look like you didn't really know what was going on. So this actually started first. We have now a team called go to market engineering, and they built something called a deal bot. And what a deal bot did was look at all the reasons why, we did not land a company. So goes to sales rep, you're like, oh, the product was too expensive or like, you know, the champion used to work somewhere else. So, like, that's why we didn't win. And you actually have this deal bot that will go through, read all the documentation. It has all the call notes and everything. Recommends like, oh, you never got in front of the economic buyer or you were late to the game. Gives you this brutal honest of, like, what actually happened in the deal. And they they have something called demo days at Rosalvy or Friday, so they showed this off. And the finance team basically said, wow. We should do the exact same thing on the pipeline. Like, which of these deals do we actually think will close? And we're not just gonna base it off of, oh, this thing is, like, marking, like, stage one, which if it is, you know, in October 27, it's not gonna close by the thirty first. But the deal bubble actually tell you, like, actually, in the call, they said they were making a decision on Wednesday, and, you know, you should hear back any day now. So this is really a stage four, not a stage one. So we can get much more accurate forecasting data through this this deal bot. So as a result of this, we've gotten much, much better at predicting which enterprise deals will actually close. And, you know, more fun for us is when we see the deals lost, deal bot of, like, getting some real feedback back to the sales organization of, you know, why we didn't win certain deals. And this has helpful because we've been on this big price reduction, mission this year, which we'll continue to do, of course. But we've been learning through this deal bot that that may be not the only reason why someone didn't sign up. Want to give a shout out to Shopify and other finance teams here. Their CEO and founder Toby, famously issued an AI memo. And it said, like, before you hire a person, you have to prove that you cannot use AI for the task, and he published it. And whether you agree or not, I think the question is like, how are we rethinking headcount hiring? Particularly, many of us are in annual planning right now. How are we thinking about headcount planning in the age of AI? Or have we made any changes at all? Is it just too soon to actually see it in in how we forecast? Yeah. I think we're not quite thinking about it as extreme as that, but more of a, what are the limits that someone can go to individually when leveraging tools with AI and what they've got in front of them? I think we kind of have an idea of like someone should almost be about 150% capacity before we actually get a second person in. That is what 150% capacity looks like today is very different to what it looks like two years ago. And that 100% is very vague. Some people push that to like 250%. Some people are like 110%. They're like, oh, that's too much. So we've got to try and figure out what actually that looks like with various different tools. Far, engineering output, we can use some tools like Weave, which are not perfect, but they can look at productivity of what people are doing for sales, very similar to the deal bot that Vercel have. We have a lot of activity and reviewing and understanding how much are we pushing towards those deals closing as well. So for now, I think it's more about empowering our team to understand that they can push the limit of what they can do a lot better and a lot more and they will help their team. Right? Have 35 small teams at Postdoc, which ranges from two person teams to I think our largest team is maybe one of our sales teams, which I think is like eight. So when you're a very small team, you can notice these massive productivity increases when something is actually useful rather than something just being kind of there and not actually doing anything. So we really empower the teams to put in our head count planning actually from we ask our engineering teams, do you need a new person at the end of each quarter? Do you need two new people? And the number of people entering teams requested was much lower at the end of Q3 than it was the end of Q2, just with the cost of Cloud Code had gone up massively. So people are feeling much more productive in their teams, which is great for headcount reduction. Yeah. I think we were taking a bit of a different approach. I think we have very little interest or appetite for any incremental headcount in g and a. And the first question is always, why can't this be automated? Or why can't our existing team or tech stack or workflows really be retooled? Why can't this be automated? And so we, yeah, stretching to the bounds of kind of what one individual person can do. But I think the way we think about it is like, you know, it used to be you had one accounting manager who was your corporate accounting manager or your technical accounting manager or your AP manager. That one person now can do all of those jobs because they have, you know, so many automations now built in or, you know, agents that can perform a lot of the manual transaction processing. And so you have one person that's doing so much more. And so I think really what we think about in terms of additional headcount is at some point, that person has met that capacity because there continues to be things that can't be automated yet. And so once that person reaches that, all of those things in the margins, then we think about, okay, maybe we do now need one more headcount. But I think what this is doing is that it's making everyone's role more multifaceted. Right? Everyone is becoming more enabled to be more responsible and have bigger roles. And so I think it's a really great opportunity right now. I know, you know, the accounting profession has had a bit of a downturn in terms of interest among graduates or or undergrads. And I think this is a real opportunity for us to sort of reinvent or redesign the role and the function. So I think that this is actually a really exciting time that shows that we can not just put person someone person someone, excuse me, someone through, you know, years and years of having one small role at one large company doing kind of the same thing over and over. You know, maybe they're not answering the what, but it's still very important that they are able to answer the how. We use AI everywhere, not not just in finance. I think a couple of the areas where we made, like, pretty big leaps is, go to market is one example. So we used to have 10 people that were inbound SDRs. They would look at basically someone hitting the contact sales thing on the website, evaluate the leads, respond, coordinate, and use of 10 people. Now we have one. And the only thing this person does is basically manage the agent. So it turns out we get a lot of people using contact sales to ask for support. So now they just get funneled away, and there are a lot of, contact sales that are just not qualified leads, bunch of people spamming us for whatever reason, so we can remove that. And then for those that are legitimate leads, they can actually respond to the customer and say, hey. You know, some more information on x y z. Click here to, like, schedule a call with me. And they think they're interacting with a Vercel SDR, but it's just an agent. So that used to be 10 heads. That's now one, and we're now moving this globally. Just wanna make sure, like, the language thing actually works well enough in in other countries. So that's one example. Like, legal, which I also run. I'll be using tools like, you know, GCAI, which is saving both a bunch of time, heads. And we looked at it and we we've shaved up $700,000, on our quarterly external legal counsel bill, which maybe tells you how big it is, after we started using GCAI, and result continues to grow. So massive cost savings. And even just like the small stuff. Right? Of like, hey. Can you just review this NDA? And you can that can be done in a couple of minutes as opposed to like Whatever. Forty five. Or or we need a quick addendum to some customer contract that would usually go to commercial legal team, and now you can just forward that entire thing to GCAI. You still need a lawyer to actually look at it. Sometimes these models hallucinate or do something like really, really stupid. Guess this is why my doctor tells me that she can use ChatGPT for medical advice and I should not because it might be an expert that should review these things. But, yes, we do the same thing. So, as a result of this, like, we moved all the, you know, inbound SDR representatives and our outbound SDRs. So we're just becoming really more efficient, across the company. And then more specifically to finance, we launched something. So there's a result product called v zero, which is built to build applications. And then we launched something internally called d zero, standing for data zero, which is the starting point for your most basic data requests. So rather than like, if you work in product or finance or any other team and you have a pretty straightforward data question that may not be in a dashboard, you can ask d zero to to do it for you As opposed to needing to, like, hey, data analyst x y z, here's my here's my prompt. Can you do it for me? And it's obviously always super, super urgent, so they have to drop whatever they're working on. But d zero is sort of meant to be able to do be like first line of defense for these kind of requests. And that has allowed us to hire fewer data analysts because the data analysts can then work on, you know, maybe d three at some point with, like, the really complex advanced requests. Just a few themes I'm hearing. One from Kelly around the role is changing from a specialist to a generalist in a little bit. It sounds like we're evolving a bit in what we do. We're managing more systems. Suppose having one person just do AP all day, every day, maybe doing AP and AR. We're all broadening our role. Martin, incredibly impressive cost savings. Congrats on that. Too much spend before. We were running a little hot, awesome job getting it down, particularly as we scale. Right? You made a good point there. We're growing so fast. We're adding zeros to some of these numbers as we grow. With with that in mind, it's like two themes there. How do we ensure AI is an enabler of the team or maybe it's just not and not a threat? I think it's like some of us that have been through acquisitions before, the finance person, you're often digging your own grave through the acquisition. When you're buying an AI tool, there's this risk that you may be digging your own grave as an employee. I'm automating myself out of a role. Anything that we've thought through from a framework? Or is this just like the world that we live in? Or do we really push the team of like, how do we have something handle this specific task? And we free you up to focus on other things? It sounds like for you, Martin, it's a bit of both or maybe more of just like we're going to ruthlessly look at headcount. But from the group, is any of it just shifting what people are doing as well? Like how do we get folks comfortable? Yeah, I mean, think we've kind of taken an approach that if AI is replacing people in postdoc, it means you probably weren't doing it very well to begin with. I thought like obviously So avoiding it, but if it's too late, it's too late. So it's for the people on the team, we want to enable them, as I said, to just be very productive and think about how they improve the business. So if they can find a use case that's not necessarily within their tight focus, that's going to improve the business. That may then as a focus for them to reshift what they're doing and do that and push that towards making postdoc better rather than necessarily the small team focus that they had before. So everything we do when it comes to whether it's involving AI or not is about pushing postdoc forward. It doesn't necessarily need to be in what you're doing. We have another value, a step on toes and you're the driver. So getting people to think about how to make post hoc better doesn't necessarily need to be within the very tight remits that we want people to focus on. Focus is a really important part. And when you focus AI on something, it gets even better. A wide remit for AI is not very good. It will make more mistakes. Will, if it's not tightly controlled, it can hallucinate and get things wrong. So I think having people understand that tightly controlled iterative processes with AI are what we believe is the best use case right now rather than unleashing AI and help thinking it's going to be this big boom shot fix for a department or a team helps people in their role think, great, and I'm going to use this to the best of its ability because it's going help the business. If people are thinking about themselves in this process, then they're probably thinking about it wrong. I think everyone wants to be more strategic. And that's what we all want to spend our time doing is that upper level thinking. And so all of these automations and all these improvements to our processes and our systems give us leverage to be able to spend more time being strategic and providing value to the business. And so I think everyone is more thinking about how much of an enablement moment this is for everyone to get out of some of the mundane repetitive work. And the work that clearly, just like reading an NDA, it's like, oh my gosh, I've seen 100 of these in the last year. That's not something you wake up in the morning and you can't wait to go do. And so all of this is, I think, just bringing us to a more sophisticated approach when we think about the work that we do. Yeah. It's almost like the Industrial Revolution. There would be people going from doing the work to, like, operating the machinery. I don't think AI is capable to, like, run cross functional projects or, like, you know, sort of PM a big initiative, you know, solicit, you know, customer feedback and sort of deeply understand issues and use cases and then sort of get that implemented in an organization. That's the stuff that's really, really difficult. Sure, the analysis and some of that work can be done. But yeah, we'll probably have fewer people, particularly in G and A, than we would have had if we were sort of at our stage fifteen years ago. But I do think the jobs that we do have and we'll continue to add more headcount, of course, will be meaningful, more interesting Yeah. Than the jobs we had, at that period of time. So, we actually have a little bit of this problem with, like, the Resolve core product as well because we effectively sell it. When you purchase Resolve, you don't need, like, DevOps team because we do that for you or, like, site reliability engineers. And some of our customers would have more of, like, a fixed mindset. Be like, wait. Then I have to, like, go fire this team. They don't wanna do that. They wanna have jobs. Like, no. No. No. They can now work on, like, value added stuff, not just making sure that, like, your systems are up and running. Right? So if you can sort of get people to think about, hey, you can actually have, as you said, a meaningful more strategic and interesting job and someone else will take care of the mundane stuff, this is kind of a win for you. Increasingly, I'm seeing sort of people lean into this as opposed to being scared of it. We're gonna open up for q and a in just a minute. But before we do, I just wanna ask the group here, any parting thoughts or words of wisdom as we think about AI and on the finance team? Anything that, like, you hadn't said yet that's, like, probably worth bringing up? Otherwise, we'll start to take some hands. No. I think that's all I've got. Okay. Some question. Right? Let's do q and a. Go for it. Right here. Yeah. I'll just I'll repeat the question for the group. How do think about build versus buy? And so go from So some things are like more specialized, like GCI and Harvey are like pretty good tools for legal. I don't feel like we need to kind of reinvent the wheel on those ones. So let me buy. Like, use Ironclad for all of our contracts. They have, like, an AI feature that we can use on top of that. That's kinda, like, easy plug and play. Then we buy. We build when things are, like, unique and custom to us. So, like, the deal bot we built, Moonshot AI, I think, we're using as an underlying tool for it. D zero, we built ourselves. You know, connects into, like, our Snowflake databases and, like, extracts data. Don't think you need like the data team built it, which like they're smart people, but they're not, you know, PhDs in AI to build something that's like super complex. There are enough, tools out there including open source stuff that you can kinda use to to build some of these things, connect in the data sources, select the LLM you wanna use to actually drive it and sort of play with it, from there. Yeah, our approach has basically been if something is super easy off the shelf, including Campfire, use it. If you need something super, super specialized to you, you kinda have to build it. We similar approach. I think what we thought about from our finance tech stack was let's spend the time to invest in the tools that are themselves investing in this technology, and then really lean in to what they are doing and continue to put pressure on them to develop the agents and automations and the integrations and then, you know, provide a lot of feedback to them. That's that's our approach. Yeah. I think we're very similar. We can't take a selfish approach of this, is it very functional? Is it something that we just want to get a task done? Or as we are a developer tool and we want to build products for people building products, if it's something that's going to improve our company and get better, then likely our customers are going to want to use it. So we should build ourselves and then potentially think about releasing it to our customers. Also we really want to empower. We have a very technical commercial team, everyone in our customer success sales team, they've all probably former engineers. So they're all building things. Our sales team built their own quoting tool, which takes in all the information that we've got from previous customers and then is connected to Salesforce and it's it's which now we're thinking of releasing to our customers that are a bit later stage. Yeah. Amazing. I think we have time for one more question. Russ, go for it. And the question is how has your criteria for hiring changed in the last three years? I think for us, we have a large focus in our hiring process. We're actually hiring in finance right now for people that have driven automation processes. How have they themselves automated something in the process that was a functional task and basically show the skills or the mindset of when analysing what tasks should be automated and which ones are like the core brain that we will and are strategic enough that we don't want to lose focus on it. So the assessment of that is not like massively different. It's just sort of similar, but it's more kind of what we're asking, whereas previously it would have been maybe more deep on the technical side of doing something, but as now it's like, how do you think about replacing that with a tool of some kind? We see a lot of our customers, like in interviews, hiring people that are what's now called like systems forward or even like a forward deployed accountant or finance person. People, to Fraser's point, are, like, they're just constantly looking for ways to to, like, automate a task. I think there's a saying that, like, sometimes the laziest employees are the best employees because they, like, find a way to automate something because they're lazy. It's almost like have that lazy mindset of, like, take a task and, like, in between one of our investors, Somya, the controller from OpenAI, she would always push her team there of, once we closed, we would use the window into the next close to find whatever task took the longest and then, like, offload it to some sort of automation. And so, like, having this mindset of, like, we're constantly rebuilding and rethinking, we've seen work really well. Anything you wanna add before we wrap up? I can add a quick thing, which is we found interviews, like face to face conversations, were pretty low signal. Some people are great interviews because they're, like, charismatic, might not be very good workers, and vice versa. What AI has allowed us to do is actually to ask for more meaningful things part of, like, a deliverable case study. So we load people up with, like, v zero credits so they can, like, actually go build something. And through that, they can also use LLMs. And we typically give them, like, a big dataset or, like, some problem to automate, and they can do that in, like, you know, three hours. Right? If you had done this pre AI, you would have asked them to do, like, six months worth of work. Right? But you can now give them a relatively small problem to solve, give them the tools to try to solve it, and see how they go about doing it. And it's been fantastic for us to sort of test out, like, are they leaning into using tools in AI? And do they also understand what a good work product actually looks like? We can have slightly less reliance on face to face interviews. We still do those. They start you just come to the case and you're done, but it's a combination. Amazing. Well, let's give them all a hand.