E9: AI's role in transforming finance - with Chris Mossa

Not everything has to be scalable right off the bat. It's far better to figure out what works in a manual environment. Do it by hand first and then figure out how to optimize that thing for scale.

It's very hard to have this balance between you want to build for where AI is going to be, but at the same time, especially in financial services, there's not a lot of room for errors, hallucination.

There's a lot of discussion around how AI is going to disrupt the finance field, and it is disrupting it in many ways. It's building some really great financial models, FP and A tooling, analysis tooling, but it's not great yet at telling a fully front to back story. It's going to get there.

What's one trend in finance and accounting you believe will shape the next five years?

Oh, geez, we spent the whole time talking about it.

You're listening to the 10x Finance Podcast, quick candid conversations with the people shaping modern finance. Hosted by Albert Ghazi.

Hello, everyone, and welcome to the 10x Finance Podcast, where we dive into the real challenges and opportunities shaping modern finance teams. I'm Albert Gozi, Co Founder and CEO at Aleph. And today, I'm joined by Chris Mosa, CEO at Graphite Financial. Chris, great to have you here.

Thanks, Albert. Great to be here with you.

Chris, before we get into the hot take of today's episode, can you give a brief overview of Graphite for those in the audience that might not know about the company?

Yeah, absolutely. So, we're an outsourced accounting finance, really a general back office partner to early in growth stage, mostly venture backed businesses, really being a partner to them across accounting, finance, tax, and people operations.

Alright. So Chris, let's jump straight in. What's your whole take?

Well, you know, it's controversial in some circles, non controversial in others, but you know, what we've seen is, you know, there's a lot of discussion around how AI is gonna disrupt the finance field. And it is disrupting it in many ways and it's building some really great financial models, FP and A tooling, analysis tooling that with our clients. Where we see it stop short, though, is that second and third iteration, that interaction with kinda the real world, the replacement for, you know, maybe a financial analyst, but not a VP of finance yet. You know, it's not a partner yet in, you know, in the leadership spaces in terms of how to tell a story.

It can interpret results really well, but it's not great yet at telling a fully, you know, front to back story. It's gonna get there because the pace of change from last year to just now and from six months ago to just now is mind boggling, but it's not there yet. And what we're trying to do is really wrap our service around what AI can do as fast as it's evolving and really bringing on the service layer where the AI stops short.

I think it's very interesting and I think like, you know, in general, one of the mental models that I always have, especially with AI and how fast things change is, it's like, it's very hard to have this balance between you want to build for where AI is going to be, but at the same time, especially in financial services and like the domain that we both operate there's not a lot of room for errors, hallucination. So how do you think about that? How do you like try and keep yourself up to date? How do you find where that line is on different workflows?

Yeah, I mean, you're one hundred percent right. You know, in our world, the penalty for being wrong is pretty steep. And so the checks and double checks and reviews that we've built around our application of AI and our customer work and our partner work has actually, I don't want to say it's added time generally, but it hasn't purely disrupted the amount of time that we spend in a way that, you know, you see these sort of, you know, ROL kind of ten X time returns on people who have introduced AI workflows. I don't think we've quite seen that.

I think we've probably seen more of like a three to one where things have gone, you know, lower level, automatable, repeatable tasks and the early building happen in a blink of an eye much faster based on really smart prompting. But the additional services we're wrapping around kind of interpreting that, but also reviewing that work doesn't make it quite a ten to one yet. Right? It's we're still seeing it at kind of like a three to one return because the penalty is so high.

So, we have, you know, a very high bar for what we're putting in front of our customers. So, if a human's not doing it and going through the same checks, but the model's doing it, we're still going to provide the same checks on the human side until we're more comfortable.

Just like bring you to something concrete for the audience out there, I think you were talking before we started recording about financial models as one example and things like, Hey, Cloth4XL might like get you a very clunky, but like a first draft or something, but like that's not gonna can you can you talk to me a little bit about how you think about that as an example?

Yeah. I would say the first drafts are really good. Now everything is rooted in how good the prompt is, and if you're working with an agent, how good the kind of natural language and the dialogue is. Right?

So it requires a tremendous amount of information gathering early days or early in the process to be able to sort of structure the prompt to drive to the output that you want and then iterate on that. And that first draft is really, really good. You know? But a financial model in our world is something that it's a living, breathing tool for our customers.

And so what we find is that once you get kind of beyond that first draft, that first model that hasn't interacted with actuals yet, it hasn't interacted with reality, it doesn't need anything interpreted yet. It's really good. But it starts that experience starts to degrade a little bit as it requires kind of more more human experience. It requires, you know, the ability to see what actually happened and interpret that in light of the model beyond, hey, this is up and this is down and it's because you spent more in this area.

Well, that's great. Well, but why? Well, we spent more on legal fees because we received a notice from X, Y and Z partner that were being sued and, you know, we don't like, it doesn't have enough of the context yet to be a true kind of storytelling partner to give the answer behind. It can tell you the what, but not always the why.

I have two things that I think you touched on that I would love to double click and go a little bit deeper.

Yeah, sure.

And both I very much agree with. The first one is about the importance of prompts. I think earlier this year, we actually released like a prompt library. I don't know. I actually don't know you contributed.

We can ask you for I haven't seen the prompt library. I haven't seen it So, disclosure, we're a customer.

We're customers.

We'll ask you about the no, it's not it's like a marketing asset. It's in our website. But basically, we ask a few people, you know, good prompts. How do you promote that even, you know, you're a finance person at heart, you're also like the CEO of a company where a lot of finance people How do you think about knowledge sharing, you know, having the, you know, everyone from like the best prompter that you have in the company? How do you think about things like that?

Yeah, I'll be honest with you. I don't think we've done enough here. I think we've, you know, we've partnered with you. We've partnered with a lot of other tools as well that are building in AI native worlds.

Right? And they've dramatically those tools have dramatically improved our ability to service our customers. What I don't think we've actually done enough of, and I think we're learning this afterwards, we sort of rushed into the environment and figured out the tools that that were gonna accelerate our service. We didn't actually level up the kinda, you know, knowledge base of our team fast enough to know how do you be really effective in this world.

It's not just, you know, a new tool and you turn it on and it just does better work. It can be that. But the way that you interact with, with Aleph or any of these other tools, it's a totally different style of working. Right?

I mean, when you see some of the prompts that our teams have sort of learned over time in order to get that even that baseline model, I mean, they almost read like a code base. Right? Because and they have to be so complete and so accurate in the prompt. There's a diminishing return that can happen.

If I spend, you know, if I spend two hours collecting all the data and building a prompt that gets me that first model, well, I might be able to stand up the first couple sheets in a model, right, by the time I I I've done all of that. Now that's not the right way to look at it, but we have to be really intentional about that prompt. We have to be really intentional on what instructions are we giving the model to be able to do the work that we want it to do. I think we have more work internally to educate people how to do that better.

And maybe that's a good content idea for the market. Maybe we should, you know, we, SLF, could facilitate some of that and, like, you know, help crowdsource some So, our marketing team hopefully listens to this podcast every episode, so I'm sure, you know, they will pick up on that and we'll discuss it. The other part I wanted to go into, you know, I don't know when this podcast is going to be released, but I'm doing a vibe coding webinar next week. So I've been thinking a lot about my own vibe coding.

I think like one of the things I was thinking that resonated with what you said is there's almost this disjoint experience today between what you're able to byte code or where you're able just like from your way into existence and what's like an actual financial like, you know, what's an actual output that you can use repeatably? And I think we've seen that in engineering and we've seen that in our domains and like it's kind of fine for, you know, prompting to be a way of getting a first draft where the objective is more learning than scaling. And then you might need to throw it away and then you might need to build again from the ground up in something that is scale.

Does that mindset resonate of like, you know, the objective of what you're prompting into existing is much more learning than scaling it?

I think scaling it's actually very, very difficult because, know, at least the work that we do so specific to the customer, right? Because we're a professional services environment, you know, we're not working on the same company all the time. So when we're trying to do this, the model doesn't always have all the historical context. We sort of have to build it.

And once we're in flight on a specific customer, it learns and it gets smarter on that. But we have to sort of create that from the beginning and create all of that historical context, right? Here's the last board deck. Here's the last strategic management presentation.

Here's the, you know, twelve months or twenty four months of financial data. Right? But we're starting each client instance, we're sort of starting again. So the scalability problem for us is real.

Right? Each one of those requires almost as much work in many cases as it would be to to do it manually. Now it's gonna keep getting better. It's an investment that we should make and we are making.

We think a lot about scalability. Like, a question I always put to the team in anything that we do is if it doesn't work for a thousand customers, we shouldn't do it for four. Right? Like, we're not in the bespoke solutions business.

Each partner has a bespoke team and we're building just for them. But in terms of the systems that we use and the platforms that we use, edge cases degrade the work over time. And so we really look to, you know, and again, not to to pat you on the back, but you guys have built something, you know, really pretty special. And that's why we do a lot of work upfront to figure out who are we gonna partner with because whatever we partner with has to work at a thousand customers.

Right? Not four, not six. So, a nice problem to have, right? We're running a business of a nice size, but it's also a problem and it's a challenge and it's something we have to think very intentionally about.

Nice. Chris, as we move towards wrapping and we try to keep this episode short, we have this rapid fire set of questions.

Yeah. Okay.

Three questions. Are you ready?

I'm ready.

First question. What's one mistake you see finance teams make over and over again?

Finance teams who don't stay close enough to their accounting teams and really understand how the numbers became what they are before they start to do their work and carry it forward. Those two teams have to work shoulder to shoulder or else the end product is seventy five, sixty, eighty percent as good as it could be. But it's something less than one hundred percent.

Very much agree. What's one piece of advice you would give finance leaders scaling their team?

Yeah, it's a great question because I think a lot about this. Not everything, despite what I just said a minute ago about operating at scale, not everything has to be scalable right off the bat. It's far better to figure out what works in a manual environment, do it by hand first, and then figure out how to optimize that thing for scale. What I've seen too many times is somebody sort of rushes into a, you know, a bright shiny thing because they think it's scalable and it doesn't end up it doesn't.

Next, they don't really understand the operating principles behind it. So work it out. Don't try to be scalable right away. Work out what you actually need in a hyper manual environment and then figure out how to optimize that thing for scale.

Last one. What's one trend in finance and accounting you believe will shape the next five years?

Jeez, we spent the whole time talking about it. I'm gonna go back. I think, you know, I think the junior, I think junior finance roles are going to become storytellers faster. I think mid and senior finance roles are going to become, you know, leadership storytellers faster because what technology and AI specifically is gonna do is it's gonna unlock a tremendous amount of time that today gets spent doing repeatable updates and tasks that will completely go away. And if you really enjoy this work, what you're not going to be able to do, right? What you're not going to be able to do is hide behind the busyness of work. You're going to have to find a way to be able to interpret and tell a story and use that work to show real value to the business at every level.

Really great. Chris, it's been a pleasure to have you on. Thanks so much for the time.

Thanks so much. Great to see you, Albert.

Great to see you.

That's it for this episode of the 10x Finance Podcast, bringing you sharp, real world finance conversations powered by Aleph.

Learn more at get aleph dot com. That's g e t a l e p h dot com.

‍