If you would like to earn CPE credit for listening to the show, visit earmarkcpe .com slash FPA.
Download the app, take a short quiz, and get your CPE certificate.
Finally, if you enjoy listening to FP &A Today, please go to your podcast platform of choice, click the subscribe button, and leave a rating and review of the show.
And now, on to the show.
From Data Reels, this is FP &A Today.
Welcome to FP &A Today, I'm your host Glenn Hopper, and I'm really excited about today's guest. Matt Hudson is the new Chief Financial Officer at Grammarly, where he brings a rare combination of expertise in finance, product management, data science, and growth strategy.
Before stepping into the CFO role, Matt spent more than a decade at Coda as an early team member, helping grow the the company from beta into a million -dollar business, a multi -million -dollar business, that is.
Over his tenure, he led across product, go -to -market, and finance, ultimately serving as CFO and playing a central role in Coda's acquisition by Grammarly.
Earlier in his career, Matt held roles at Google and YouTube, including serving as chief of staff to YouTube's head of product and engineering.
He holds degrees in finance and data science from the University of Michigan and has built a reputation as a data -native finance leader, someone who blends explains analytics, experimentation, and strategic thinking to scale both teams and systems. Matt, welcome to the show.
Thank you, Glenn. Looking forward to being here.
Yeah, man. It's been a while since you and I first talked and I've really been looking forward to this episode and I know you are super busy right now, so I really appreciate you taking the time to come on.
Of course, of course.
Let's dive in. I mentioned it before the show.
I am very excited about having you on the show because I think you and I are kindred spirits in the way we approach the Office of the CFO And I think that looking at your background, it's, you know, it's not your traditional come in through public, you know, big four accounting audit coming up through the CPA ranks.
And it's much more where you're seeing a lot of new CFOs come in today.
And so your career, you spanned product, data science and finance.
And I'm wondering, in all that, what led you down that CFO path and that transition?
additionally, how did your early experience influence how you lead in finance today?
Because it's a different approach. Yeah, it's a great question.
I certainly have a pretty uncommon background for folks that are in this role today.
I think for starters, this was never a preordained path for me.
I think there's a question I often give to folks coming out of school or when I'm interviewing, which is, what do you want to be when you grow up?
And I always preface it with, I've never known that answer.
So it's okay not to have one.
I bounced around my whole career.
I think the most common theme is I like being an operator.
I like building businesses.
Even back in school, I initially thought I'd go into astrophysics or engineering and just didn't find something that was that exciting.
And finance, coming out of business school and undergrad felt like the most technical thing you did in business, which is kind of why I gravitated there.
But I didn't even go the traditional finance route coming out of undergrad business, not going into banking, not going into consulting and instead buying a business, a furniture manufacturing company and doing that for a few years.
And then picking...
Google had started a rotation program that was modeled after the General Electric Financial Management Program.
And I got to bounce around finance there and try out accounting and controls, treasury, FP &A.
And it was there where I really cut my cloth and found a team that was very quantitative in nature.
Today, you would call them a data science team.
Back then, we just called it a quant team.
And it was all mathematicians, PhDs who are responsible for doing predictive modeling for revenue across all the Google product lines.
And I had a lot of fun on that group.
And that's kind of what led me down a product management path and then the startup path.
And then once you're in a startup, you can wear whatever hat you want.
And I did product management.
I did go to market before kind of going back to my analytical roots at the end before the Grammarly transaction.
Yeah, that with you at Google, my first finance role, I actually started in marketing.
I was in telecom and I ended up moving over in first job after business school in marketing.
I ended up moving over to finance after a couple of years.
And I was in telecom.
I've never worked with a team that had that many nerds, that many smart people that were thinking.
And I really I was in finance, but I kept gravitating more to the engineering, you know, in telecom.
It was a lot of network routing and all the everything that went through that.
But these guys were also really good at Excel.
So I learned early on, you know just all these excel tricks and we would sort of compete with each other and everything and it really influenced my approach to finance and it sounds like at Google you had a similar experience and being around that that level of math nerds and data folks can really settle in early.
Yeah and I'd say one other thing that's very unique about finance so I've spent years in product management and running go -to -market teams etc and the really unique part about finance is within a company it's the only one that has the vantage point into everything and you kind of have a reason to stick your nose anywhere so when it comes to building businesses and having a data orientation to it it's the one function where it's always appropriate to go look into marketing go look into sales go look into engineering like you can kind of go anywhere with it which is nice yeah and you had that mindset
you leave Google was that did you go straight in Dakota after Google yeah so I was within the Google finance team for a while I spent three and a half years at YouTube which was owned by Google but was somewhat run as like a not fully independent subsidiary, but a different team, different location, different headquarters.
So it was separated enough and then went straight to, we didn't call it Coda then we called it Krypton.
Um, but yeah, we had, we had raised the series A, um, and we're in stealth for the first four years or so.
Uh, but I was there on day one with the, with the founding team.
Wow. So you, they're day one, scale it up into a widely adopted platform.
form. So now you've got big company, huge team experience around all the math PhDs, and then you're in this startup.
You're getting to wear a lot of hats.
What lessons from that rise at Coda, business and financial, you know, growing with the company, are there lessons that you picked up then that you carry today and that you're even now seeing you're applying them at Grammarly?
Yeah, I mean, this is probably an hour's worth of conversation.
conversation, I'll try to keep it brief.
I'd say it's probably worth commenting a little bit on context for both Coda and Grammarly, and just the acquisition itself.
So Coda, for folks that are unfamiliar is a freemium SaaS business used by individuals, startups, large enterprises to streamline their workflows.
It's an all in one document platform.
So think of it as a modern take on Microsoft Office, the whole productivity suite blending together docs, spreadsheets, databases, apps into one flexible surface.
And you end up seeing teams start from meeting notes, going all the way up to custom tools without writing any code, but using a common set of building blocks.
Very spreadsheet -like in terms of the formula language and how you can build pretty amazing things in it.
Grammarly probably is better known, but for folks that don't know, Grammarly is also a freemium SaaS business.
It's claimed to fame as being one of the early Grammarly NLP or today you'd call it AI writing assistance that helps a lot of people communicate clearly and effectively across all sorts of different apps.
So you install a client on your device, like a Chrome extension or a Mac app or a Windows client, and you get help anywhere you write.
Grammarly is used by a lot of individuals and it's built a really impressive business, over 700 million in revenue.
So in terms of scale between Coda, Grammarly, YouTube, YouTube, Grammarly, for me, reminds me a lot of the YouTube days.
So I was there when it was around, I think I joined it when it was around 140 million in revenue and it scaled up to about three billion by the time I left. So it's kind of through that really big growth phase to your question on like what learnings come across it.
I'd say Coda, one of the biggest pieces, was just thinking about SaaS businesses generally.
YouTube and Google were fully ads businesses.
experiences. And there's a lot that you learn around product telemetry and measuring a subscription business from first principles, like cohorting subscriptions or thinking about ARR and NRR.
And I'd say that Grammarly has this nice intersection of it's a lot of those same principles and mechanics, but at YouTube -level web scale, because of Grammarly's business just being so large.
So I think very early on, some of the first things we're doing are trying to stand up a lot of the forums and the stats meetings or metrics meetings that we had patterns of at YouTube, but doing it on the Grammarly side.
While we're on the business, I wanted to ask about this.
So Grammarly just secured a billion dollar non -dilutive growth investment.
I think first off, I guess for our listeners who may not be familiar with what this means, can you explain what this non -dilutive investment is, who the investor was and actually, you know, seeing that deal, I'm wondering how common is that and is it becoming more common now?
Just kind of a little little bit of background on that investment and the type of investment that it is?
Yeah, certainly. So as most SaaS businesses go, you end up having a very similar cash flow pattern where there's a big investment up front, and then it pays back a little bit over time.
And so the firm that led this was General Catalyst. The fund is called CVF for the Customer Value Fund.
And it's a financing vehicle that co -invest in a company's sales and marketing spend up front, up to a certain threshold, and it gets paid back over time as a profit share of each respective cohort that they fund.
The value add for it, and what makes it super appealing for us is really simple in the sense that it improves the cashflow dynamics of a SaaS business.
Maybe to give a very silly comparison, that lofting business that I ran when I was in college had a negative cash conversion cycle, which was lovely.
We'd get cash from customers before we had to pay suppliers.
SaaS businesses don't really have that dynamic.
You end up spending a bunch up front and then earning it back over time.
This flips that a little bit.
This allows you to have a lot less capital intensity up front and maybe to pick a very common way that finance folks will understand.
If you think of this whole equation as a net present value or just kind of cash flow equation and think about IRR, this dramatically improves our IRR for every investment in new cohorts because where you don't have to put as much cash up front and we still get most of the return over time.
And you keep the equity.
That's right. You keep the equity.
And compared to most traditional places, you're going to go acquire capital from a bank or whatnot.
There's no anchor on the business.
There's no lien. And you're utilizing an asset that traditional banks don't look at.
They don't look at customer cohorts with predictable attributes as an asset that you can underwrite against. CVF allows you to do that.
To your question on how common it is, that's a good, I think it's a somewhat newer product in market.
I want to say that they're to the tunes of 50 or so deployments.
I might even be one of the few CFOs that's used it twice because we used it at Coda too.
Did you consider, was it even on the table to do a traditional round, an equity round instead of that?
Or was this, you were going to do this or nothing or alternative financing?
financing. I got advice from someone many years back when we were first looking at this at Coda, where equity is really useful to fund your R &D.
It's your most expensive cost of capital.
It's something that you want a lot of leverage on.
You want to see big returns on.
Your sales and marketing expense, spending equity dollars to go fund that, it's kind of like an expensive way to go do it.
So CBF, I feel like, offers a very good sales and marketing option to do it.
it's not a huge commit up front.
It gives us lots of flexibility on how much to draw and when and the timing of it.
So it's very helpful that way.
And it's much, much cheaper than using equity.
And you mentioned the cost structures for a SaaS business, but I think also now, well, Vramerly is AI driven and AI driven products have their own cost structures, model inference costs and GP usage and all that.
How are you adjusting your margin models and planning frameworks to reflect these kind of changes.
Yeah, that's, this is a common, I was, I was at a, a conference the other month and it was a common question that came up of people being worried, worried about the COGs and worried about like an explosion in R and D because of all this stuff.
Maybe two different perspectives.
I think one is I have a philosophy or a perspective that this is all going to come down over time.
It's not going to quite, it's not going to be as big of a crunch as we're playing it out to be right now.
I think we'll find optimizations and ways that you offload certain aspects of the job to something that's much cheaper.
I think the second is, and maybe one other comment, I don't think I've seen huge explosions yet in cost. Like we just haven't incurred that or seen something like that yet.
The second is from like a ROI perspective, my guess is this is all gonna end up just getting folded into the way that we traditionally measure and monitor ROI.
There was one tip I was talking to a few months ago and there's a question of where is this going to get categorized?
Like in the world where you've got, you know, agents that are running around and you've replaced key workflows within the team with something that's agentic or whatnot?
And like, how do you, how are we going to capture it?
And the analogy I gave was you'll end up treating agents as headcount.
It'll be, you'll allocate them to different lines of business and you'll have agents in R &D, agents in S &M, agents in G &A, and you'll just track the cost the same way.
So you'll end up looking at ROI exactly the same way.
How much value am I normal measures against the cost that I'm investing in them.
And they'll all fold into the rest of the common parts of the P &L.
Yeah. I feel like this is a sci -fi episode of FP &A Today.
I mean, it's amazing that that's where we are, where we're really having this kind of conversations about agents.
And I know I have this whole soapbox I get on about people calling things agents now that aren't.
But I know true agents are coming.
And in the interim, there's agentic workflows and all sorts of tools that can, to the end user, they might as well be an agent.
they look like one.
But it's the productivity and efficiency.
I was talking to someone months ago about, you used to be hired as an FP &A analyst. You would kind of come with what you would bring would be essentially all of your models, all of your Excel skills.
And so you're getting hired, but also all your experience and the models you've built.
And talking about a future where when you're hired somewhere, it's you and all the agents that you've built, sort of the way that we're going to look at it.
So it's pretty, pretty amazing times.
Oh, that's a, that's a very, it's, and in that world, the idea is that they're portable, which is an interesting one.
Uh, I was, I was listening to someone talk.
We're on the topic of, uh, shapes of teams going forward. And how do you, especially if agents, um, and a lot of the workflows that get replaced end up being the ones that folks that were entry level were doing before, like, how does that end up shaping the shaping?
What teams look like broader instead of a pyramid, it looks like a diamond or it's, you know, big in the middle versus a big on the bottom.
them and um one of the analogies going back to the point of like how do you measure roi and think about costs for agents and how do they how do you think of them if you think of them all as head count and just count the number of them you'll still have a pyramid it's just that at the bottom it'll be a bunch more of these agents that you're using to deploy on certain tasks or workflows and coming out of school your job is not going to be doing those tasks yourself it's going to be managing a set of 15 agents that are doing those tasks and you like immediately moving like What is prompt engineering
other than giving people clear instructions, the context necessary to complete a job?
It sounds a lot like management.
And it's going to be managing agents, not managing people to start your career.
Yeah. And that's one of the conversations that happens over and over is, well, if agents are taking all the entry -level jobs, then what are the entry -level people going to do?
But I think you just nailed it right there.
It is. And it's funny because when we think about prompt engineering, it sounds like something that you would have to have more of a CS background, but really it's almost the liberal arts or psychology background, even, or just, you know, industrial organizational psychology or just basic management.
It's breaking down a task into its component parts and managing those parts of the task.
And I think it's I don't know there.
I don't want to be overly optimistic, but I think there is a lot more doom and gloom around AI maybe than is warranted.
If you look at every other tech revolution we've been through, you know, 120 years ago, we were all 80 percent of the U .S. economy was in agriculture.
You know, look where we are now and there's jobs we couldn't have imagined.
But I think you're, you and I are aligned on that with what we kind of see the future being as agentic powers get stronger and more capable.
This will be the same product, like a similar productivity shift of what you saw with, I mean, for finance people in particular, the spreadsheet, you went from calculating it all by hand manually to now having it programmatic computers that also used to be done by hand, which was a lot of work.
And it allows you to to go to the higher level strategic stuff.
FP &A Today is brought to you by DataRails, the world's number one FP &A solution.
DataRails is the artificial intelligence -powered financial planning and analysis platform built for Excel users.
That's right, you can stay in Excel.
But instead of facing hell for every budget, month -end close or forecast, you can enjoy a paradise of data consolidation, advanced visualization, reporting, and AI capabilities.
possibilities plus game -changing insights giving you instant answers and your story created in seconds find out why more than a thousand finance teams use data rails to uncover their company's real story don't replace excel embrace excel learn more at datarails .com as a as a data guy I really want to hear your thoughts on this and at the size companies where you are and we've got listeners that are everywhere from SMB all the way up to, you know, the biggest enterprise companies out there and all at different levels of data maturity.
But I, you know, we've been talking about digital transformation for 30 something years now.
And the idea that we, you know, I don't even know what digital transformation means anymore, but to me, it's, it's evolving to keep up with the technology that's out there and, and at the foundation of that is data.
But I still talk to businesses of all sizes that don't have their data house in order.
And I'm wondering what you're seeing and what you're thinking.
And for these agents to work, we have to have that sort of foundation layer of data.
And I'm wondering what tips or advice or insights you might have looking at the different companies where you have the amount of data that's there and for people who maybe aren't caught up yet, how you would guide them to think about data and what they need to do ahead of this sort of agentic future that we're headed towards.
Yeah, I mean, I think maybe two or three reactions I'd give.
I think one of them is I would distinguish stuff from, like when we talk about data, I'm often referring to things that are like heavy analytical workflows or about deriving insights from stuff versus when I talk about agentic workflows and using agents to do stuff.
I think I'm more relying on a perspective of the world that's about human workflows within software.
It's not as much like magic answers to complicated questions.
I feel like that is part of the equation that's probably the furthest off.
Being able to ask a question and get a full faith and fidelity answer back, I think will actually be a long pole in this environment.
I think on the data side, the best practices for how teams get set up, I think are no different today than they were 10 years ago when things like dbt and thinking about warehouses versus lake houses, thinking about how to use structure things in a star schema using highly performant column IO oriented data stores separating storage from processing that that all I think is still true and as true today as it was before I do think in the data realm there has been a shortcoming in folks end up thinking about a lot about the middle of the stack like the warehouse the lake house where you store stuff
how much store how much data you can store etc and then the last mile often stops short it's like Like we get to a dashboard or a report.
And what I think that there really needs to be is a lot more innovation in what I'll call data apps, which are consumer grade applications for data products.
Like things that feel, like maybe to pick a really clear example, every single company has a single pane of glass problem for their customer usage of their product.
Like you wanna look in one spot everything about this customer.
And you want that to be an experience for like a frontline salesperson that is working with them.
And it's a combination of what's in the CRM, all my notes about them, how they're using the product.
And as a salesperson bouncing around between a dashboard that's kind of like in a language I don't totally understand because it's a bunch of charts that I'm not super used to, and then going to my CRM to try to do stuff, none of that is a particularly good product experience.
Like if you put a product manager hat on, what you really want is a data product that feels like a way to explore the organization and click on people, not refresh the filters and change the select list on a dashboard and hit run.
You'd want to interact with these data products that feel like a first -class application.
I think there's a huge gap right now on that side of the stuff.
On the AI world, and where you're the agent or agentic workflow world, I think where that really is going to come together is, today there's like, I wouldn't say a hard fork, but there are at least a couple different camps emerging in the AI space.
I think there's one camp, which is very much I'll call like the black box world, where it's like you submit something and get something back.
The other side of the spectrum, I think, is, and this is, I think, the philosophy that we've got at Grammarly, is one where we expect a lot of these AI agents to work where people work and feel like a supplement rather than a replacement.
So it's about, kind of similar to what the spreadsheet did, it's about giving someone a lot more power to go execute on a lot of stuff rather than removing the human in the loop entirely.
Those, I think, are going to be totally different worlds or constructs, and we'll see over the next couple of years how things really play out.
Yeah, and when I talk to a lot of companies, if they have, if they happen to be large enough, and they've real are committed to data, and they have a data science team, and they have machine learning engineers, you know, they're gonna they're gonna build tools that they're gonna use within their company that are agentic, or at least agentic workflows.
And they're gonna be able to operate with that, but when other companies apparently have FP &A in a lot of cases, And it's a stretch to think about them on their own building out some big AI plan.
And I wonder, what I end up telling them is really focus on your data right now because the tools that you're already using, your CRM, your ERP, whatever tools are in your finance tech stack, they're going to have generative AI worked into it.
And the foundation of these tools is going to be the same.
same. There's still going to be a thousand tables in your NetSuite instance or whatever, but how you interact with those tables and thinking about going through and building your custom reports, instead, it's going to be a conversational layer.
And we're seeing in the FP &A space, data rails among them are already starting to integrate the AI genie kind of tools where you can talk to your data and all that.
And at Grammarly, are you guys focused right now?
Are you building any sort of new interfaces using AI or are you seeing it in any of the tools that you're using as far as applying AI beyond your basic?
And we'll talk a little bit about machine learning, too, before I let you go.
But any generative AI use cases at Grammarly right now?
Yeah, I mean, I think it's a good problem.
I mean, there are certainly a handful of cases.
One philosophy I have a strong one on just in terms of like, how do you engage the finance and see them to the broader company is being heavy dogfooders of your own product.
Meaning that you use a lot of what you build.
I think there are a handful, going back to like the dichotomy I was just describing a moment ago of like the data world versus AI and like the agentic flows and the agentic flows often being about workflow rather than being about analytical insights or answers.
I think there's several areas where you end up seeing a lot of broad -based workflow tools.
I think a couple somewhat silly ones that come to mind for me the first one being any time you have to go do deep research on a topic like I think in a traditional fb &a team you're getting ready for the board meeting and you want to look at comps public market comps all right well that normally would have been like two days for someone to go and like copy and paste stuff from a source somewhere and instead it's like oh let me queue up a deep deep research topic to go pull in this stuff with citations I can go check it and spot check it it's not the end of the world if there's something wrong in it
but it It saves me a lot of time to get that first baseline black pen done.
The other one that I found really useful in my role in particular is getting feedback on presentations.
If you're going to do a board deck, I heard this from someone a couple of months ago, and what they'll do is they'll walk through their board script.
So they'll have the enterprise version of ChatGPT or whatnot.
They'll load their board deck in it and say, hey, pretend you're a board member and ask me questions.
Take a look through this and grill me.
And then it might find parts you're missing or things that you're doing and ask it questions that you can now get ahead of and do dry runs.
I think for the finance team in particular, I still feel like there is a gap in terms of a lot of core finance flows are still really hard. And a lot of the AI tools or services or solutions today don't address them particularly well.
Maybe to give a really quick anecdote, both because of my background on that team back in the Google days, I still am very hands -on keyboard, write a lot of code, do a lot of stuff in R and Python, and was an early adopter of GitHub Copilot.
So I've seen from the early, early days of what this collaborative code editing looked like before Cursor really took off and all this stuff.
And it was magical.
It was a huge benefit, a huge plus.
And I think that they did two or three things really right that are independent of the quality of the model.
I think one of them is the user interface was done in a way that it It allowed me, as someone that knew what I was doing, to edit and correct things that were wrong.
So it would write code and I can go change the code.
Or it would suggest code and autocomplete and I could easily dismiss it by keep typing.
That was very useful.
Or I could like select a block of code or ask a question about that block of code and say what is this doing or like why is this working this way.
Those are three okay to be wrong, I can still go change the output and have it be useful.
To translate that to finance for a moment now, Now, what I want out of an AI agent that I'm using in my workflow is not someone to answer the question for me.
It's someone to like, maybe to pick an example, we talk about vibe coding is a cool thing right now.
Why is there not vibe modeling?
Like take this spreadsheet I've already built, I want it now for this company or these new set of inputs.
And what you're going to go build are not static values, I want to change the formulas.
And I'm going to go change the formulas in the spreadsheet.
Now I'm going to go do it there.
There's not a lot of that today.
Or I'll pick another one that seems like a prime use case for something that should be really easy to do.
And that's like flux analysis or BVA prep.
So close the month.
Tell me what's different.
What moved? Give me the summary of it all.
That seems like something that should be very easy of prepping versus reviewing month close entries or something like that.
Yeah. And to get that flux analysis, to have it useful, because you can, you could dump, you know, just convert your financial statement month over month into CSV and upload it into, into ChatGPT or whatever and do some great analysis.
But without tied directly into your GL, it's like, yeah, I can see there's a variance of, you know, 8 % over last month, but I have to then go back to my system and find out what it is.
And I could do the same thing in Excel.
So getting that value.
you. And I like I know why Microsoft, we all know why Microsoft put 15 billion dollars into open AI and has continued to invest in AI.
I mean, you know, they want to bring back Clippy.
They want Clippy's revenge.
They want it to actually work.
And it's on one hand, I want to say I'm surprised that Microsoft using Copilot.
I mean, they're just they're not there yet with their chat GPT and the others are kind of eating Copilot's lunch. But you know where they want to get is that true integration into Excel where you don't have to be able to write the 10 line long nested if statement and do all the complex stuff that week that you and I came up being proud of.
look what I can do in Excel and all that.
Like I see that, you know, they want to get it to that conversational interaction or iterating.
Like it's, I get to ask that question all the time or people will upload, you know, try to upload a massive spreadsheet with all the formulas and the formatting and everything up in the chat GPT.
And it just completely craps out because, you know, it blows out the context window and it's just too much for it to handle.
I don't know. I mean, I expect it to be there and I know there's a lot of platforms out there that are trying to...
And Google has tried with building AI into Google Sheets as well.
But apparently, it's a harder engineering problem than you might imagine.
I'm sure that there is a lot on the back end technical side that make it difficult.
I don't think it's intractable.
I think it's possible.
Ironically, I'd say the hardest part about this, in my mind, is going to be on the UX side.
How do you do it? I think the key insight that I've had...
Clippy. Clippy. Yeah Clippy.
But that was like a great example of imagine in GitHub Copilot instead of like auto completing code it like was prompting me like it was like a modal like that clearly won't work.
I do think there's something though about what is the right UX to provide suggestions and to provide these kind of like interact decorate work with users in the surfaces users are already in with this kind of like extra intelligence which I think is a hard problem.
Part of the reason we're so bullish about what we're doing with Grammarly we view it as a very good platform to go do this But I still think, back to the point on finance teams and adopting AI, I just don't think we've seen them yet.
I think one of the analogies I heard from someone a couple months back was, if you think of modern computing, there is a period of time where everything was done on the command line, and not a lot of people could use it, and it wasn't really that useful.
and then it switched to the graphical user interface and everything changed.
And now a whole new category of people could use it and do stuff with computers.
I think it's the same thing for this.
Right now we're in the command line era of AI.
We're talking about chat interfaces and the pinging back and forth with stuff, painstakingly copying and pasting your GL into chat GPT and having it regurgitate stuff back out.
That is a horrible UX.
No one wants to do that.
But we just haven't gotten to the next UX paradigm yet.
Yeah. All right. It's easy for me to get sidetracked and go down a tangent with generative AI, but I think it's important.
There's a lot of people who never, even in finance, who never thought of or really considered AI at all, but you're an OG with AI.
You're a machine learning guy and you're a data native CFO. And I mean, a lot of people are, obviously it's been the trend for a long time, but going back to your background and history and everything, being a data native CEO?
What does that actually mean for you now in day -to -day practice?
And how does it shape how you approach planning, forecasting, decision -making, and all that?
Yeah, I don't know if I've used that exact phrase before, but I certainly prefer getting a SQL statement than getting a CSV.
I think that is much more descriptive to me of what's going on.
Certainly, I'm non -traditional in the CFO mold, but I think the most salient historical pattern I've got from the last 20 years is that I spent a large time outside of finance.
Very practically, I think that ends up meaning that I show up differently in three distinct ways.
I think one of them is, and this is less about having the data background.
It's more, I'd say I learned a lot more having the operator experience in product or in go -to -market because it allows you to sit down with some of these other executive partners and solution based on firsthand experience versus just judging it based on reporting your data.
One quality we used to emphasize both in the the YouTube days when I was responsible for the data and data science team there, and in the CODA days where we ended up merging our user research and data science team so we had qual and quant together, was that data itself has a way to lie to you all the time.
And sometimes you have to look at the data, you're trying to get to truth, and you have to supplement it with qualitative input too.
So back to how do you partner with a finance hat on with all of these other functions?
It goes back to starting with empathy, understanding what the data is saying, but trying to see what their real goals are, is now the right time to be looking at these numbers, or should we be looking at a different one?
And maybe the very last point I'd give is that it's meant that my background has had much more...
Every couple of years, I end up on these learning curves where I'm ramping up very quickly on what it means to run a function like sales or marketing.
And that means that I end up approaching a lot of these problems with first principles thinking rather than a hardened playbook.
Thought here, as you were talking about that in the teams that you've led.
I frequently preach, and I get pushback from this, especially from the OG FP &A professionals.
I really believe that if you're coming in to FP &A today, and that's your aspiration to take that path in finance, it's not just the traditional good at Excel, good at modeling.
I think you need to lean heavily into, you don't need to be a full -blown machine learning engineer, near, but data science chops and not just, you know, many people in FP &A can get around in SQL and they can write queries and all that.
Now, that's kind of, that's pretty much become table stakes.
But when I get pushback, it is, well, you know, in large enough companies as well, there's a data science team.
I'm a customer of theirs.
I get it. But to me, it's like, I'd rather not be a customer.
I'd rather get to that route.
I mean, if somebody were coming into FP &A right now, would you advise them, especially kind of knowing where we're going in with AI and maybe increased automation, is it required or recommended that they become pseudo data scientists?
I mean, I think it's a very personal question.
It's probably an overstatement to say that every single person that will ever work in FP &A needs to know that stuff.
I think that's certainly not true in a lot of parts of the process and what we end up doing day to day.
Maybe to pick on, I think you have to look at the whole shape of the problem and pick which parts of this you'll end up being an expert on doing machine learning or becoming an expert in statistics and a lot of the underlying mathematics that go into predictive modeling or ML or anything like that is certainly one category it might be the last category I would pick maybe maybe to pick with the first one that I'd start with when I was running the data science team at YouTube we used to have a statement which was the hardest problem we ever faced was asking the right question it was an implementation it
wasn't getting the the data.
It wasn't cleaning the data.
It was starting off on the right path and knowing what to ask.
That's true in data, data science and finance, like asking the right question is hard. It does not require you to be an expert in machine learning and know all the different ways that you can do clustering.
It's probably useful to know what solutions you have once you've got a question and know what's possible, but you don't necessarily have to be on the ground floor to implement it.
The second thing I was going to say, if the first one is asking the right question, the second one I'd say is I often describe this as the data gene and it's about knowing in your mind like the shape of data that answers certain questions like what do you need this thing to look like to be able to visualize it the way that you want to be able to measure it the way you want and again that's not a question of like how statistics work works or understanding like the details of how neural nets work but you kind of need to know what shape it needs to be in up up front so that you can go answer the question off
of it. And then the last thing I'd say, which is the spot that has the biggest variance in terms of people being able to do well, is inference.
Once you've asked the right question, you've got the data stood up in the right way, and you're now looking at something on the other side of it, are you able to interpret it in a way that is meaningful, that is pointed in the direction of truth, and you'll be able to go change the direction the business is headed using it?
All three of those things, I think, are way more important than, you know, actually writing code or knowing how to use very specific ML models.
Maybe one other comment I'd give, you were making a statement a moment ago of if you're on a big team, you know, there's a data science team, my business partner with them, I ask them for stuff.
Early on in my career, someone gave me advice that there's kind of two ways you end up scaling yourself.
you either scale yourself with people and you lead teams and that's how you get leverage or you scale yourself with technical skills and you learn how to code you learn how to get access to stuff and for the first 10 years in my career I did everything in the second camp it was all about technical skills learning as much as I could if knowledge is power then access are the keys and if you can't get the data if you can't pull it well guess what like you don't necessarily have a seat at the table everywhere and so I spent a lot of time I'm focused on how do I make sure I can get the stuff I want
know how to use it be able to I'll know how to ask the right question because I have a vantage point to see what's important to the business but now I need to be able to go get it all and do something with it that is probably a convoluted answer to your question of what advice would I give to folks right I think it's a little bit personal like it's different not everyone is like well suited to have the data gene or whatever and like you should know I have a proclivity that way or I don't and I wouldn't beat myself up over not having it.
Yeah, that's a great point.
And I actually really like what you said about scaling.
And I think for me, the reason it is necessity is the mother of invention and maybe because I was wired to go in that direction a little bit early, but I was in smaller companies usually turn around situations where I was coming in without a lot of resources, private equity groups, wanting detailed reporting, modeling, and not being able to do it and not not willing to work, you know, the extra 20 hours a week on top of what I was already working to do that.
It was, I had to lean into figuring this stuff out myself because it was my, my way to go in.
But now that, you know, the way you broke it down, it is, it's almost like a left brain, right brain.
I don't know, but there's an orchestrator that is knowing these are the questions to ask and then having the team that can go answer them in different ways.
And then whoever, you know, your, your data engineer is not typically going to be your best storyteller or presenter or whatever.
So having those different lanes where if you've got the resources that you can have different people doing that, that does make sense because people are wired for, you know, things they're going to excel at.
And it's funny, working with engineers all the time, you know, they're always, when they try to put together like a UX for their prototype website or something, I'm thinking that now I can see why you're better in front of the blinking cursor than designing the webpages and everything.
But that, yeah, that makes total sense.
And maybe the last point I'll make on it too is that in many cases, there's lots of, I think we've focused on FP &A largely from product analytics, business analytics perspective.
There's many parts of it that using advanced tools would just be the wrong approach or the wrong tool for some of the problems. As an example, me trying to understand or modeling out headcount growth across functions on the team, I'm not going to move to R and write out a big model.
that those scenarios like you're not going to do that.
It's kind of a silly, silly use of time.
On the flip side, if you're in a company that's got large scale marketing programs, and you can get into the click data, that's much better to like switch to some of these more advanced tools or tools that are built for those bigger jobs.
Even something at the like transaction level, like if you're able to plumb transaction level stuff into the GL, you're looking at stuff that way.
Now it's like, all right, I can see perhaps using one of these tools to go do some first pass analysis.
I know a lot of like the big four, where they've got technical teams that do accounting or do audits, like they'll staff some folks to look for, you know, signs of fraud or forensic accounting and those kinds of tools.
That might be an interesting use case, but it's still a far cry from many of the problems that FPA teams face day to day in terms of just planning the business and helping think about scenarios.
It's a different different toolkit.
Can we talk about your, your team FP &A and sort of the broader finance function at Grammarly and maybe how that's, how that's changed since you came on and, and it's not just you coming on, it's the shift in, in expanded focus, bringing Coda in as well.
And I imagine that there's a lot of shuffling and moving around.
So what's, what's the basic structure of your team now?
Yeah. I mean, it's, it's somewhat uncreative.
Maybe I'll just start with, start with that.
I think many finance teams end up having a very similar pattern.
So as context, Coda coming in was a pretty thin team.
We ran very, very lean.
We had a controller, had someone that was leading all of our FP &A work, and that was it.
We were roughly a 200 -person team at the acquisition time across the whole company, and most of the finance team or the functions within it were outsourced.
We used a PEO. We used a third party to do a lot of our bookkeeping.
so when it came time for integration there wasn't a ton that we had to do for people it was more on the system side to plug things together uh the structure of the team today is kind of what you'd find in a lot of we're we're roughly a 50 person finance team and the structure is very common for what you'd see in a lot of places i'd say two or three broad groups there's the accounting controls group that includes everything from you know revenue accounting traditional accounting accounting, tax, treasury, payroll, AP, AR, all of that.
A lot of our, I often think of it in terms of, it's our looking backwards team.
It's like accounting for history, making sure that we've got everything in the right spot.
And then we've got our FP &A group, which is roughly mapped into three big groups.
One of them is our corporate finance team, all of our consolidation, giving us the full P &L view.
And then two business partner teams that are aligned to the sales motion.
So a self -serve group and a managed group.
And then the last group is kind of our corp dev and IR team.
And there, if the FP &A team is very focused in the present, you know, this year, this quarter, the corp dev and IR team is kind of thinking like longer term, thinking about our capital structure, thinking about M &A, those kind of questions.
And do you have, are any of the members on your team, hardcore data science folks or?
org. So Grammarly had a, so at Coda, I ran the data science team as well.
And coming over to Grammarly, they had a very large data org and a very large established data science team.
So that currently is all within engineering.
However, one comment I'd make is that Grammarly as a product is very data intensive.
So a very big part of that group is about the foundations of the product itself.
And so whereas at Coda, I think our data science team was much more on the, I'll call it but like the insights and value add side versus core product underpinnings.
At Grammarly, a lot of them are staffed towards core product work.
At YouTube, we had a very similar dynamic where the data science team that we had carved out under that brand, actually, I think we might've called it the quant team while I was there too.
That was, again, more oriented on, in a supporting capacity of driving insights, partnering with all those feature teams. And we had a separate group that did all of like the core, the hardcore ML for like recommended videos on YouTube and stuff.
That was fully staffed in engineering.
Gotcha. And I'm wondering, our listeners always like to hear from folks what KPIs are important to them.
And I think it's an interesting time for Grammarly as the business broadens into this more of a collaboration tool.
But can you tell us without giving any secret sauce or anything, but what KPIs are important to you?
Like, what are you looking at most frequently and what's maybe changed from how grammarly, what used to be important to them now, as the shift happens, what kind of things you're looking at?
Yeah. I mean, it's maybe I'll start with the boring answer and then maybe I'll, I'll switch to something that's a little more creative or interesting.
The boring answer is I feel like a lot of the core things that you would have measured 10 years ago are the same things that are still important today.
You're looking at usage, whether it attracts new users or within in an acquisition.
You're trying to understand what's going to lead to conversion, paid conversion after someone's been using it for a while.
How to think about spread.
If someone starts using a product, starts paying for it, do they bring in new users?
Is there a K factor that you care about?
And then lastly, retention.
You look at the most powerful common denominator is your net dollar retention.
You look at whatever cohort looks like, whether they're trending in the right direction over time and understand just how sticky these revenue lines are.
I don't think AI, other than it's a new product experience, and you want to think hard about users seeing it as differentiated value against other things in the marketplace, has really changed that.
I think maybe there are small things, like when you think about usage, is it something that they're actively using?
So the equivalent of a long click in the search result versus something that was generated but not used.
There's something interesting there.
But again, I'd say those feel to me very much like traditional product metrics.
Maybe stepping back to the less boring or more creative aspect of the question.
Are you familiar with Richard Feynman.
Richard Feynman he was a theoretical physicist. Yeah yeah.
Project. So it's funny I was telling someone this the other day.
YouTube and all of its supreme intelligence recommended one of his like 1950s 1960s lectures for me.
And I'm a nerd so I watched it.
And one of the students was asking him a question.
So he's known for, and I'm going to get this part wrong, so no one should fact check me, electrodynamics in quantum theory.
Don't know what any of that means.
But someone was asking about why we needed new theories versus Einstein's theory of general relativity or whatnot.
And the answer he gave was that if you've got two different models, two different ways of measuring the universe, or in our case, measuring a business or a product, and they They both produce exactly the same result.
The question might be, why do you have to have different models?
Can't you just pick one of them and use that all the time?
And his response was that it's not because the prediction is right.
It's not because they produce the same result that's important.
What's important is that when you look at the different models, it leads to different questions.
It leads to different ways of thinking about how does the universe work or how does this business or product work?
I actually think we're at a very interesting inflection point as a company right now where where we're going from the COTA world, the Grammarly world, into this new space where all these things combine, and it is an important moment to rethink what does this model look like?
What is the right way to think about how the product works?
What are the key drivers?
What are different ways you could go develop a revenue model or a growth model that represents all the moving parts and all the important mechanics?
I think right now we're kind of in exploration mode, thinking about what are the different ways we might wanna model this, but it is something we're actively going through It's fun, fertile ground.
Yeah, and it's got to be an interesting time for you just thinking about – Microsoft has had – I'm doing air quotes here – collaboration tools forever that are definitely, in a lot of ways, lead a lot to be desired.
OpenAI just announced yesterday whatever they're thinking, and that's going to drive a wedge probably more so between them and Microsoft as they roll it out.
But OpenAI is talking about doing collaboration.
And I know Grammarly has a dedicated, committed user base of which everybody in my family is one of them.
But, you know, going into the collaboration space, knowing that you're going up against these big heavy hitters there and then trying to think about where this goes in the future, it's got to be an exciting time to be a CFO right now.
I mean, I think it's just an exciting time to be in the space, to be in software.
where, again, I think the CFO role gives you a vantage point in all parts of a business, which is fun.
But it is certainly a fun time to be part of this part of this space or part of this environment.
As I was walking through and thinking about this, I was wondering if any of that product background is sneaking into your your CFO thought and actually sneaking in is not the right way.
I mean, I think that's a great way to to sort of round out the CFO office.
But are you finding yourself with your your defaulting to having your product hat on sometimes?
times? Oh, certainly.
I think especially when you start talking about how do you measure the business.
When I was a product manager, I'd say there are a lot of different ways you can think about what a PM looks like, like what their strengths are.
I wasn't the design PM.
I wasn't necessarily the process working PM.
I was always the data PM.
I always had kind of more of an instinct for it.
Somewhat natural coming from my background.
So I think that in looking at all these problems, this goes back to the modeling thing we were just talking about.
A lot of my my instinct slash perspective starts from what are the ways that you'd actually model out the mechanics of how this new product experience is going to work?
What are the ways it's going to happen, etc?
To your macro prompt and open AI getting involved, I think that there is a gut check happening across the industry and trying to figure out where in this chain value is going to accrue and then value get captured.
If it happens at the foundation layer, at the app layer, and we're going to see all sorts of interesting investments and movement.
But it's a fun time.
Gosh, man, I can't believe how we're almost coming up on time and I've got a bunch of other questions I wanted to ask.
I'm going to try to guide us in for a landing here so that I don't know if our listeners are sitting in their car or waiting to go into the office.
I don't know if people still go to the office, but we're going to try not to keep them in the car much longer.
So I had a couple of questions.
I guess maybe I'll just throw out, I'm going to wrap these together and you can address them as sort of one.
But I always like to ask, especially you really have your finger on the pulse of AI and tech and where we are right now.
So two questions I'll put together.
One, if you're looking forward five years out, knowing where the technology is heading, how you might see the finance function evolving, and I'll piggyback on that, and maybe this is unfair, but piggyback on that.
For FP &A people who are in the space right now, maybe they're looking to build out stronger data and AI skills, as you look into your crystal ball and see where we're headed, what would you advise them to focus on right now?
um it's a great prompt I mean I think uh in terms of how does the finance function or the fpna function evolve uh my guess is it'll be a very similar pattern to what we saw in prior generations with the advent of computers or spreadsheets and ultimately what it ends up producing is more time to focus on you know the interesting strategic choices or questions I certainly am hopeful that we'd end up seeing a bunch of the more mundane or or regular work becoming less and less something that consumes the days.
There was someone that I was talking to that was describing one of the measures that you have of your broader finance org is how much of the org is dedicated within the month to month close in reporting.
If you can minimize and squish the amount that comes from month close reporting and just turning over, you maximize more time left over for strategic thinking, interesting problems. problems, I imagine all of these tools are going to end up helping with that distribution and shifting towards more time to spend really driving inference and understanding rather than just turning the crank on making sure all the books close and that we're reporting out budget and stuff like that.
Second part of your question was where should folks apply focus?
Yeah. Yeah. Sort of to ride the wave and be ready for it.
It is a night and day difference talking with folks that that are curious versus not.
Finding individuals that are interested in adopting new tools.
Trying to find or experiment with new ways of doing things.
Boy I would certainly be on the front of the line of try out new tools, see how you can learn them.
I might even encourage towards, instead of using packaged tools that might have like somewhat limited capability or perspective and try to go towards more of the raw tools to see how you can string them together.
Things that are composable.
Maybe it's a different way to say it.
You want to try to use things that are adaptable and then you can see applying the same skill set that you develop using it across many domains versus just being locked in or stuck in the one spot where you can use it.
Love it. Love it. Great guidance there.
Okay. So lightning round two questions we ask everybody.
So I'm going to dive in here.
The first one, what is something that not many people know about you outside of your work interests?
risks? I mean, it's for folks that have worked with me, they probably know it at this point, but one of my favorite parts about working at YouTube was that it was within walking distance to Chili's.
If Chili's was a franchise model, I'd probably be in it.
But I started, that was probably my first quote unquote real job was being a busboy at Chili's.
Dude, I was at Chili's.
I started as a busboy.
I did prep cook. I was a waiter.
And by the time I left, I was a bartender.
I'd made it to the top of the chain at Chili's.
I was there for quite a few years.
But I love that we have that in comedy.
If you want to find hard -working people, start by closing Chili's at the end of the day.
Oh, yeah. And you would go home smelling like fajitas.
Mitch Hedberg had a line of he wished there was a fajita cologne.
I was like, go work at Chili's.
You'll wear some fajita cologne.
For the longest time, I made my wife take me to Chili's on my birthday.
That would be my celebration.
That's great. That's great.
Okay. This, I'm curious to hear your answer on this one and I need to start logging these.
I say it every time we don't, we don't log them, but it would be very interesting to see what is your favorite Excel function and why?
I have a confession.
I have been a hardcore Google sheets user for at least 15 to 17 years.
Wow. Well, because you were, yeah, because you were there.
They called it tricks internally when they were first building it, but I was like one of the first users of it and I've been using it since.
And ironically, Ironically, the formula that I use the most within Sheets, Google Sheets, is not available in Excel and it's called Filter.
And it's basically, I mean, now they might have equivalent versions of it, but I find it to be, using Filter along with Named Ranges allows you to write very, very readable formulas that can be quite complicated.
It'd be equivalent of doing a huge block of sumifs or countifs or whatever.
But it's, yeah. So anyways, Filter would probably be my top list. If I have to stick to Excel, it would probably be Offset or Indirect.
Yep. Yep. So no, no, we're, we're, we're equal opportunity here.
So I just, I, it's, it's funny how few people, uh, actually I talked to someone at Google a couple of months ago that they, that he doesn't use Google sheets.
He's in finance at Google.
Yeah. I mean, for a long time we could, I didn't even switch from it until I had moved out of finance because we would have to use S base or Hyperion, which didn't have a Google connector or Google sheets connector.
So we'd have to do it in Excel.
Well, Matt, this has been an absolute blast. I I really appreciate you coming on and thanks so much. And best of luck to you and Grammarly.
I'm super excited to see this new collaboration direction.
Yeah, that sounds great.
Can I give one final plug?
Absolutely. We are hiring on the finance team at Grammarly and in particular are actively looking for folks that have this finance analytics background.
So engage very much in this space or this domain.
I actually love that you threw that out there and I bet a lot of our listeners are immediately opening their web browser and finally, where do they go?
just go to the Grammarly website and Jobs.
Is that the best way to find the positions?
That's right. Or you can find me on LinkedIn too.
Happy to route. Okay, great.
Great. Well, Matt, thank you very much. Thank you.
