Sacra Logo Sign In

How AI is transforming B2B SaaS

Jan-Erik Asplund
None

Intercom, Zapier and Brex have each made "bet the company" moves into AI.

To learn why, we teamed up with Sandhill Markets to host a panel conversation featuring Zapier co-founder and CEO Wade Foster, Intercom co-founder and CSO Des Traynor, and Brex CFO Benjamin Gammell.

  • Zapier co-founder and CEO Wade Foster
  • Intercom co-founder and CSO Des Traynor
  • Brex CFO Benjamin Gammell.

Transcript below.

Our three key takeaways from the conversation:

Adam Hardej:

One of the first things that comes to mind for me is how early and how long ago for you, as someone leading a company and involved in these companies, did AI really become a serious part of the strategy?

Wade Foster: 

Yeah. For Zapier, I'd say we started dabbling with AI pre-ChatGPT—kind of in the GPT-3 era a bit—but it was just dabbling. 

The launch of ChatGPT and the subsequent fast adoption there definitely changed the speed at which we intended to utilize this because. It wasn’t just promise anymore—there were real products in the market. ChatGPT definitely changed the way consumers thought about AI, and it changed the demand for AI.

I would say, pretty early in 2023 like February, we were seeing a lot more organic adoption of teams wanting to put it into the roadmap and then, of folks inside the business starting to use it for their day-to-day tasks. Probably April of that year, I had a top down mandate in a bunch of areas to say, “Hey! We're going to go do this,” and disrupt some roadmaps in order to go take on what I felt was a really, really important opportunity. 

Candidly, it's like one of the biggest paradigm shifts that I think we've seen in tech—certainly the biggest since mobile—but I think it'll probably end up trumping that when this is all said and done. That's kind of been our timeline and pretty much everybody inside of the company now is very like, “We know that we are both using and building AI into our products, but we're also trying to be adopters of it ourselves,” if 50 percent of the employee base has a Zap that uses AI as part of their day-to-day workflow.

Adam Hardej: 

Des, was it similar for you or different?

Des Traynor: 

Only slightly different. So obviously you referenced chatbots at the start. Chatbots went through three eras, if you like. There was the original phone tree style button-based bots of like, “Is this sales or support?” That was the first generation. 

We started taking it really seriously in the second generation, which was still before the LLM revolution, if you like but that was when you would do things like for instance, let the customer write a message and you'd use magic AI to work out what the rough intent was, and try to match that to an answer where you were still doing a lot of training for every question answer pairing. We had a product in the space called Resolution Bot. It’s still live, still doing millions of revenue today, but that was our first real investment in AI. That was 2016. 

Then, November ‘22, I got a message from our V.P. of AI and he was like, “Hey! We need to talk. This is huge.” And I was like, “What is?” and we played with ChatGPT for like five hours that evening. Very quickly, I think I had a call with the Owner-CEO the following day and I was like, “This is something that we need to get on.” And pretty much the day after ChatGPT launched, we had our first product on the market in January. Then, we got something else in May. And then, in June, we launched Fin, which is sort of next generation LLM powered chatbot. And that's like, we've gone all in, changed the work structure, quadrupled the size of the AI team, and just thrown everything we have on it because I think, in customer support, we're in a real kill zone. 

It’s like, AI is clearly going to change support. We're clearly not the business doing it. We're clearly going to die. Everyone gets that so we're very, very much like going after this as aggressively as we can. 

Now, Fin's now pairing millions of answers and doing millions of revenue. It's got thousands of customers. That product would not have existed were it not for this new AI revolution. I second, or even advance what Wade said. Like, I think in Intercom we say, this AI revolution is probably going to be bigger than the internet, full stop. That's kind of where we're at.

Adam Hardej: 

Ben, similar?

Benjamin Gammell: No, I mean, I agree. I’d say, for us, from a product perspective, we started to engage more with this next wave of AI more through the partnerships or the products and services we were using as a company. Like using the Intercoms of the world, for example. That was how we probably got exposed to it, first. Just seeing the work that was already being done by other companies. 

That being said, when ChatGPT 4 launched, we recognized that, it maybe not quite as dire as Des feels on Intercom, but we also have this view that for Brex to continue to be innovative and push the frontier of what we're doing, we also need to be thinking about how can we leverage this for the products and services we offer.

For AI's impact on finance, it probably was a little slower in terms of when people were thinking about “Okay, how can AI move finance?” because you just think about finance as the sort of thing that is very structured and it has to be right. You can't be wrong. It’s like there's a lot of guardrails around. I would say, broadly speaking it's probably kind of a one or two month delay for companies like Brex versus broader ecosystems, definitely companies like Intercom. That being said, we really kicked it into gear probably towards the end of Q1 last year and really thought, “Okay. Where does it make sense for Brex to play? Where would there be a competitive or compelling value proposition?” It was like we didn't want to just throw a whole heap of things against the wall and hope something stuck. We just wanted to be very formulaic about, “Where are we going to invest? How are we going to do so?” 

We didn't have an AI team, for example, pre ChatGPT, but subsequently, we  spun one up. It was like a dedicated team focused on “Okay. Run through where we think this sort of new way of doing sort of work will apply to Brex. How do we develop our roadmap around that? Is it probably a slower adoption?” I’d say versus the Zapiers of the world it was very much in full swing. 

Adam Hardej: 

Something you all mentioned and which is really interesting is the ChatGPT moment and how you think it’s, “Oh! Cool.” It's this new tool and there's a paradigm shift but the ability for that moment to sell internally by getting the internal team on board of “AI is worth spending time on?” and externally—can you touch on what that was like? How important was that? 

Wade Foster: 

We were playing around with GPT-2 in 2020, and people talked about it, but nobody was saying, “We have to go put this on roadmaps.”

If you go play with GPT-3, it's good. The models are solid. Sure, it's not as good as GPT-4 or even GPT-3.5, but it's solid. It does good work. You can build very useful capabilities on top of that. 

But the reality is just because it existed didn't mean that it galvanized like an entire ecosystem to go do that stuff. So, I certainly think that ChatGPT did. 

It was the catalyst that lit a bunch of people, caught everyone's attention even outside of tech. It had launched in November, and I was visiting family for the holidays. People not in tech—like I had an uncle who leads a Bible study—and they were talking about and using ChatGPT to write the prayer. These are not tech forward folks, and you're just realizing that “Okay, this is gonna be a part of the future.” So we're either gonna be a part of it or we're going to die to use this thing. I think some industries are maybe further upstream and ahead on that curve. Some are lagging. But make no mistake, it doesn't matter what industry you're in, this is going to touch you and it's going to fundamentally change how it works. It's just going to. Some might take longer to figure out but eventually, it's going to impact everything. 

I think that it is helpful when you lead a large organization because inside of a company—at Zapier, it's about 800 people—one of the hardest things is getting them all to row the boat in the same direction. But when you have a moment like this, it's super clarifying. So now, everyone's like, “Yeah” and you stop arguing about whatever small thing it was that you were arguing about because you're like, “Oh! Is it going to be this or that or what have you?” and you start to go like, “This is dumb. Why are we arguing about this?” It’s like this thing doesn't matter. You have a moment like that and it really is much easier than to get 800 people to say, “Okay, we're gonna go do this.” So, it is definitely helpful to galvanize folks and it's also exciting. 

ChatGPT definitely felt like the future was here, that we were living in a sci-fi world and for those of us who work in tech, that's kind of why we showed up here. It's like, we like building the future and being a part of that. So to actually have a moment where you feel like you can be a part of that, that’s exciting.

Adam Hardej: 

This is actually like a part of the paradigm shift.

Benjamin Gammell: 

A part of it. Also, you're being accessible, right? And anyone could interact with them outside of work, inside of work, depending on how sophisticated or not sophisticated you were technically, which meant that to your point, Wade, it was like everyone got it very quickly in the organization that this can have a material shift.

If it was something that required a lot of technical expertise to engage with, then sure, maybe the engineering team would be like, “Hey! This is going to revolutionize everything and it's going to make a big difference.” But does our sales team agree with that? Does our marketing team agree with that? Operations and such? How accessible it is as a technology makes it much easier to have that galvanizing effect because people can engage with it versus like when everyone was using machine learning, which is obviously adjacent. From that perspective, that's much harder to access and understand the impact, whereas this is much more accessible.

Adam Hardej: 

And Des, you probably experienced it directly on the buy side. Tell us, when spinning that and using it, did you realize that on the buy side there were people accelerating “Give me some AI”?

Des Traynor: 

We felt it from every side, so I agree with Wade and Ben. The missing hero in the story is the person who designed the user interface for it because they made it so easy and so accessible and that’s actually what changed people's ability to interact with AI. It’s like the playgrounds before where it was shift and return and all this workflow, it was janky. But once this thing dropped and people started playing with it, it became very clear that it was real. That was great for me because it galvanized for us in three different ways. 

One, all of our team played with it and they were like, “Holy shit, this thing is real.” All of a sudden, everything else in the road map looked so damn irrelevant like, “Ooh! We're going to build a dark mode, a command cage, it doesn't matter.” Then separately, in November-December, all businesses were still a little like, “Oh! I don't know. Hallucinations are apparently a problem and blah, blah, blah.” But there was definitely interest. 

There was a growing awareness that something's changing here. The really cool thing was, for the longest time people’s perception of chatbots on the internet was terrible. Everyone was just like, “Oh! Those things are useless. Skip past them.” Then, I think, ChatGPT was the first time people were like, “Oh! You can actually say the thing you want and it mostly gets it right.” So it changes people's willingness to interact with a chatbot. Then the thing that really suited us once we had Fin was that it also raised the standards of what people expected. They used to expect some clunky, derpy derp of, you mentioned the word “password.” So, here's the only article I have about “password.” That was the nature of chatbots, pre 2022. 

Now, even if you look at the types of queries that people ask of the Intercom messenger, it's like they're talking to it with a lot more confidence in its ability. And they're starting to believe in a sense. 

Now, I will just say the biggest sort of domino to fall is, when Google and Apple launch these things as on-device software that everyone can access, that will be the perceptual cliff that we fall off. 

All of a sudden, all the resistance goes away and everyone's just living in an AI world. That's the moment, like when tools like Apple's Siri or Apple or DLK, Google thing become mainstream, that's when AI just becomes a part of everyone's life, full stop. And we're changed forevermore.

Adam Hardej: 

Do you think everyone will continue to think that ChatGPT invented the chatbot? They didn't, but they owned that moment. Or do you think that it'll fade into a space where everyone’s building their own things, there’s a lot of competition, and open source software?

Des Traynor: 

My guess is it'll go the way processors went or something like that. You don't use the best available processor in every device in your house. Sometimes you use the best if it's a massively complicated task and sometimes you use a cheap one. Cheap in our world is either low cost to run, low battery, high speed, or whatever.

But I just think it's a case of right-sizing the thing you're using for the purpose that you need to solve. So to put it another way, I'm sure Wade or Ben can say as well, sometimes, you use GPT 4, even though it's more expensive and a little bit slower and sometimes, you use GPT 3.5 because it's cheaper and faster. It really depends on what you're trying to do. If you're trying to do something that requires a lot of high inference or high deduction, you're probably going to reach for 4. So the spectrum of possibilities is there and I think, the thing that I don't have great line of sight on is, “Does open AI want to be like the AWS of AI where they sell it to everyone and basically they’re like a really high end sort of hosting company while also being a research lab while also pushing consumer products while also allegedly building a hardware product?”

I don't really know how that company grows up. There's so many directions that it can go towards like potentially trillions of dollars but it's not obvious to me. What is obvious though is, there are now loads of options out there between Mistral, Lama, Claude from Antropic, and OpenAI's models. That's really, really great for us because it means that there'll be competition, various different models trained for various different use cases or price points or whatever. The spectrum of options is open for us as people who will now be depending on these things, which is awesome. But, as to where I go, that's anyone's guess. ChatGPT could be the future of Google, but so could Perplexity or anything else as well. 

Wade Foster: 

I probably have a similar viewpoint. See Google. Obviously they got their wedge by having better search, but the magic was, they invented AdWords.

They built this incredible business model on top of search and when we look at LLM specifically, the sort of in-state business models there’s not obvious to me. I think to Des's point, we're going to get more knowledgeable on what are the specs that we care about and how we're going to use them for different purposes.

The person that's going to win is the end consumer. The consumer is going to get incredible experiences that they otherwise wouldn't have had. If you're an investor, you're probably asking, “Where's the economic value going to accrue to?” It’s not obvious to me yet. There's probably smarter people out there that have better and keener insights. To me, it just feels like the consumer is going to get some awesome stuff for the foreseeable future because of just how much investment and how many people are building unique, remarkable experiences in their pocket of the universe.

Benjamin Gammell: 

I very much agree with Des and Wade. I think there's an element of—when you actually frame the question of build versus buy—when I think about how companies and consumers will interface, a lot of it may not be going direct to ChatGPT as it is, today.

That's obviously primarily how people are interfacing with the product and AI, today. But when like Intercom, Zapier, Brex, other companies develop their own solutions leveraging that tool, maybe you're not interfacing as much directly with AI and maybe it's more via these ecosystems that are being built on top of it. I agree with it that it's unclear to me where all the economic value is created and I suspect, it will actually be a situation where all the boats rise with the tide as companies benefit from cost reduction opportunities. They'll also benefit from revenue opportunities.

Consumers will, hopefully, benefit as well from a better system, better service, and better products. Then obviously, the infrastructure players and the ones providing it will also obviously benefit. So, it will be spread across the industry and I don't think it will be just like OpenAI. I don't necessarily think that OpenAI will become the next Google-like monolith. It will probably be dispersed a bit more. 

Adam Hardej: 

Ben, can you talk a little bit about whether you’ve seen and felt a difference on internal processes outside of like an AI button? 

Benjamin Gammell: 

Yeah, I would say for a lot of companies, then, at least Brex is an example. I would say the more near-term opportunities are probably on the cost reduction side and then, the revenue generation side. For just the average business, when we think about the number of engineers we need to develop code or the number of developer hours that are needed to shift a code base from one to another, it’s business as usual work for us as a company, but these are getting much faster. You can get a lot further, a lot quicker, at least within the initial development of code base, such that, we actually do think there's pretty material cost savings for us as a company from an R&D perspective. When we think about it on the revenue opportunity side, it's for Brex, we were trying to be very careful about where we thought we would play and where we wouldn't play in this space.

From a finance perspective, I would say, closing books fast is an obvious area. Forecasting improvements is an obvious area. Where we went to market from a product perspective is really around helping our customers review expenses faster, and it was leveraging LLMs to essentially help us better complete memos and also itemize receipts. So just tactically to give you an example, if a customer takes a photo of receipts and they have alcohol is a policy that they don't cover for the employee's expenses because of LLMs, we can flag it, obviously like no receipt has alcohol. It has “Aperol Spritz” or whatever. 

Through LLMs we’re able to better identify those types of spends and then highlight that to a customer. For Brex, we think of that as an add on, a benefit of a better product. We don't think we're going to necessarily build our entire product around that or from an AI perspective, at least today.

Adam Hardej: 

There was a recent sound bite from Sam Altman who was interviewed about this idea of the one person, billion dollar company. Wade, within Zapier, you’re already at the forefront of that type of thing. What does it look like? Is it just more extreme? Is it growing faster? 

Wade Foster: 

I don't think we're at a spot where you're going to see entire functions just basically disappear overnight because an LLM can do every single thing that a human can do. 

They're generally not taking jobs, but they are generally doing a much better job at absorbing tasks—very specific tasks. So inside of a company like Zapier or many companies, you can see individuals who are leaning into using these tools and who generally are able to automate those sets of tasks or do them much more efficiently as before. 

What will be really interesting is to see how startups tackle this, because I do think startups are not encumbered by a bunch of legacy choices and they can just build with this mental model from the ground up.

What I'm observing from folks that, I would say, have the most deftness with these tools is that they're able to work across a huge variety of tasks that they maybe, in the past, wouldn't have been able to. 

You could take a generalist who is pretty smart about a lot of things and now, they can masquerade a little bit like they're in legal, or in marketing, or in sales, or in support, or in cost, and they can use these tools to basically raise the floor.

The reality is, in most of our companies, you generally don't need to be an expert at all of those things. You just need to have a certain level of competence at most capabilities, and then you need to be great at one to really distinguish yourself in the market. 

What I think will happen is, for these small startups that are being built this way—and this is how I would do it if I was doing it, today—you just wouldn't see the headcount grow as fast. You would be hiring folks who have a deftness at using these tools, and as a result you're just able to accomplish way more. You just wouldn't be hiring like fleets of folks to do certain stuff but It doesn't mean that they won't still have more employees and they won't be building companies. 

I do think there will probably someday be a single person billion dollar company. I do agree with that and I think you're just gonna see a little bit of compression here. That's my read on it. I would be surprised though if anyone today has just like mass laid off a massive function because an LLM has been able to do the entire set of capabilities there. 

Des Traynor: 

Yeah, no, that makes sense. We see obviously customer support is an area where people are often keen to shed headcount. We just generally see it's even a really, really great rollout, it's not like they have a big all hands and “Oh! We've decided to fire all the support team.” That's not what happens. What happens is support is like churning enough as everyone knows. You see any given support team loses like 30, 40 percent of its headcount year to year because they either get promoted or they move on to another job. What you're seeing now is like the backfill not happening as quickly because for some of our customers, Fin's doing like 50, 55 percent of complete support volume resolution. But you still need to manage the AI and you still need to have somebody who's in charge of keeping the docs up to date and keeping Fin under some sort of governance, so it speaks to the right people and not to the wrong people. You just got to see the jobs that merge, converge, and compress, and elevate the way I describe a lot of them.

The support team becomes the team of answering questions for the first time and then, the last time. So you deal with it and then, you generalize it through the AI and you never see it again. That's generally like what Wade spoke to. I think we're going to go from no automation where most businesses are to step level, to task, to workflow, to go level, to all the way to say outcome level. The outcome level might be very distant and could be years away. I could just want to have great support. The step level. Things might be like “Click a button to summarize the conversation.” We're somewhere in between those two, right? We're way past summarization. Fin is doing the job of half of the team. It's also reporting back on how it's getting on. It's doing a lot more work and we have a lot more plans there. 

But what I'd advise anyone watching or listening is just think about where tasks land on the spectrum from “this can never be automated” or “this should never be automated” all the way over to “we can actually automate this entire outcome”.

Ben gave a great example of detecting if something is a legitimate expense, then, we can probably automate the shit out of that. However, processing the entire month's payroll, nah. Maybe a human should have a look at that before they run it. There's some spectrum here that you just need to keep an eye on. I just bear in mind GPT 4, 5 and all that. They'll just walk us along that spectrum over the next couple of years. 

Adam Hardej: 

We lived in an era where headcount was how you tracked a company’s growth. With AI, it seems like we're heading towards a post-headcount world. Have you seen that, Ben?

Benjamin Gammell: 

I would say for us, actually, we're a little insulated from this specifically from a pricing perspective, because how Brex makes the majority of our revenue is usage. We make interchange revenue when people use our cards to spend. And then deposits are obviously assets under management. So it's not about headcount. 

Pricing, because the amount that is like software revenue in terms of direct per seat, per month type cost is relatively low. Where we do see it occur and this is what you were hinting out in terms of this post-headcount world is, when we think about our target customer base. It is shifting, right? We can't just rely solely on headcount as a proxy for the size of a company. You need to start looking at things around—obviously their revenue, but that's hard to know as well as their spending and that is also hard to know.

It’s more like a “Let's call it upfront discovery” that's necessary for us from an ICP perspective so when we go to market and sell our product, it’s just to make sure that, “Hey! There's this company that's a 20 person company but because of how they're operating or because of their aspirations or whatever it may be, they're actually operating and almost behaving in a way that is more like a mid market or growth company. Therefore, they're a great fit for Brex, for our enterprise sort of software solution.” It's just more around upfront discovery. 

I think we've benefited historically from the fact that we've served startups. I would say startups have this aspiration of being larger companies and so they, they've had this sort of growth trajectory and tear such that we haven't run into the issue of having too few heads for our product to exist or be valuable for them because they often grow into it, eventually. But I definitely think it is shifting—how we should be thinking about our pricing more broadly—but again, I think it's less, less impetus for us right now by virtue of just the makeup of our revenue. 

Adam Hardej: 

Wade and Des, I know Intercom, I've paid for plenty of seats on that and experienced that directly in Zapier—a shift towards usage before. This isn't the first time we heard about how usage pricing has come up already. Has it accelerated things or changed the way you're thinking about it at all? 

Des Traynor: 

I think we both did the same thing recently. Fin charges per resolution—similar to how Zapier moved to usage-based pricing with Zaps, right? As opposed to seats.

Wade Foster: 

We've always had usage-based pricing but it was all under a subscription. What we added was a pay-per-task element that allowed you to have more flexibility.

I didn't really drive that change. That was more of a change to better address customer pain points where customers would have these step ups where it's like, “I have to go from 10,000 tasks to 20,000 tasks. What if I need 11, 000 tasks?” “I feel like I'm overpaying for the next 9,000.” So we said, “Okay, well you can pay a subscription for 10,000 and then you can pay as you go for the next 1,000.” And the pay as you go rate is a little bit higher than the committed rate, but not so much higher that it feels like the customer is getting a bad deal. It allows the customer to just optimize their bill in a way that the higher flexibility fits them. It wasn't really driven by AI, but AI in general is just causing more of us to ask, “What are we paying for?” I want to pay for the job to get done. I want to pay for the outcome to happen versus to pay for a person to sit there. 

Des Traynor: 

We just charge a dollar when Fin does the job of answering the question. In theory, you could say, “Oh my God, but you also sell seats.” And we do. But if somebody decides to hand back a seat and deploy Fin, the way we see this is, maybe the seat cost them $39. I think that's our cheapest seat. Maybe it was $100, but Fin's definitely going to answer more than 100 conversations for you in any given month. So we're going to make the money back. 

But on the flip side, if the customer doesn't need the seat, chances are, it's because there's no one in the seat and as a result, that salary will move around as well. So I think we've moved back to the world of automating outcomes. 

If it's the case that you want to automate away your support because you don't want to have to employ so many people in so many time zones and so many languages or whatever, there's a way to do that, but it's no longer salary. It's just usage-based pricing. 

Adam Hardej: 

Within VC, B2B SaaS as a category was always one where you could do the simple math around X contracts, X value, and churn to assess the business. We're very much getting out of that world now. When you think of forecasting and quarterly goals, do you see what’s coming as a benefit or as a concern?

Des Traynor: 

I'll answer quickly. I would say this whole world has to change. It just does. We're usage based revenue specifically when it's just really closely tied to usage and you're not buying credits or a package of coupons or whatever, but you're actually living charge usage. 

Forecasting and all that's just different. It really is the thing we care about. And the thing that we forecast often is the assumption that our own customers, if they like that, then their support volume Is in some sense predictable. But there will be spikes, of course, and there'll be dips during holidays and stuff. In general, what we look for is stickiness in the value we provide, not stickiness in the monthly contract because I think, as we shift to usage, the idea that somebody can't click a console button, especially as we've all moved to PLG and we're moving away from sales led in general, the idea of like big bang contracts that customers can't get out of is going to go away in a world where AI can do loads of things. I can probably do loads of things on a per usage charge. So if it’s not good AI, people cancel or they get someone else to do it. And if it is good AI, then you have to just ask yourself, is this use case sticky? 

There's some people who are building features very much in the same way that large enterprise megacorps build features: “We're going to summarize your blah, blah, blah and we'll do it automatically and we're going to charge you an extra $9.” There are no features I worry about where no one was ever asking a human to do that work anyway. So now you're just trying to make money off something that no one would have done. So I think you have to really look at the stickiness and relevance of the use cases. Price against that, and I think, you can start to forecast against that as long as you know your industry. 

Wade Foster: 

I'm curious, Benjamin, for your take on this—how you look at seat-based software, and as your company grows, you end up paying for so many seats of this thing, it’s like they're just vastly underutilized. So you just start to ask, “Am I getting a good deal out of this?”

It's hard for me to look at most of the software we're paying for and say, “I'm getting a good deal out of this.” It just doesn't feel that way. Then you've got these annual contracts, you've got this da da da, and the world just sort of was set up in this way that it just worked when you were like a tiny company. But as you grew, you just started to feel more and more dissatisfied with the way you're buying the stuff. So it felt like a lot of these companies were built on a little bit of a fragile thing. Then again, what AI did was that it galvanized a lot of innovation. 

So yes, you have the innovations coming from AI, but you have people just looking at, “Okay, I'm building a new company” and not only “I'm building a new company with new capabilities,” but also “I'm just going to look at where the incumbents are weak.” And “I'm just going to see if I can attack them that way.” 

I think pricing is one element where a lot of the incumbents have just pushed their pricing power and especially over the last couple years, they really pushed it because of how tight the markets have been. So it just created more pain with customers. A lot of us as consumers of software are just looking for alternative ways to do this stuff. And usage-based pricing just feels more aligned with what I want in a business. If you're solving a problem for me, I'm happy to pay you for it. But I hate it when you're going to try and charge me for something that I'm just simply not going to use. That to me felt like that was always there. AI was just another lever that created an opening for many, many companies to say, “Hey! We're going to just try and meet customer needs much better.” That's Capitalism at best. It's like, “We're here to solve problems for our customers.” I don't know, Ben, I'm curious how you see it because you're right in the thick of this. 

Benjamin Gammell: 

Yes. I'll answer it in terms of both breaks, but then also I guess my role as CFO I think, is on the right side. We've always been usage based and that's been our entire model. It's much more recent that we have software—contracted, sort of—revenue. 

To answer your question, Adam, it's tough to forecast accurately, like usage-based pricing is much, much harder than contract, obviously. But I think it's in a way that maybe people underestimate a little bit how I would, broadly speaking, say to a startup is very much like “You need to have a deep understanding of the drivers of your business and your customer base, fundamentally.” 

Des, to your point about understanding what is the support volume of your customer base and how that ebbs and flows because you can even have false triggers where customer volume support drops suddenly so you’re like, “Oh my God, this customer is churning.” But no, they're not. It's just a seasonal aspect of that customer's business. I think, moving towards usage-based pricing means you, as a company, have to have a much better understanding of your customer base. What is driving revenue? What is driving the usage so that you can understand where your customer is growing, where you may have some churn risk, and you can adapt and adjust accordingly. 

From a CFO perspective, wait 100 percent. I figured you'd say that. So much on software. It's kind of insane. There's always this sort of productivity software, and that especially is really hard for me personally, as a CFO, because how do you quantify the benefit of a seat and usage? We have a software that I—don't worry, it's not Zapier or Intercom—but we just killed the software tool the other day because it had good usage. Half of our company was using it but it was doing this thing that really didn't move the needle. People were probably using it because it was a “nice to have” and I was like, “Guys, there's a free solution that we can use. We don't need this. It's gone.” 

As a CFO or someone leading finance, I think usage-based pricing is easy to get behind because you can draw much more clear parallels between cost and the outcomes. Blanket per-seat plans are a much harder pill to swallow, especially to your point of how prices have been over the past several years.

Des Traynor: 

I think it's interesting, like late stage B2B SaaS has just been a little bit of sleep at the cash register. That's my comment earlier, but people building dark mode and Command-K shit and all that, while also jacking up prices is a little bit like the later stages of that sort of empire. Now we're building a new empire and it's going to be usage based, outcome based, and the language that Ben does speak, which is—how much are we paying for? Why are we paying? What's the volume and what's the value we put on that?

Wade Foster: 

That's what's making it fun right now. Like you said, it felt like B2B SaaS was just sort of cruising along doing its thing and got a little sloppy. I put us in that bucket, too. But we get into building these software companies to build new products, versus just build dark mode, command K, like whatever. It's okay, I guess if that's what you really want or if that's the most important thing. Then I guess we'll give it to you. 

Adam Hardej: 

I have a large question that I'll pose. Is there a new energy or does it feel like an exciting time to be at the helm of something that's big enough to be broad, but you're not an incumbent yet? You're not like a public company and people are breathing down your neck quarterly. Has it been an exciting time? 

Wade Foster: 

A hundred percent, yeah. It's so much more fun to do this kind of thing. Sure. There's nerve wracking bits because we're not entirely sure how the competitive dynamics are going to work out or, pricing is harder to forecast or, there's all these things that are hard about it, but it's way more fun to be building this. Building a company that's tackling this stuff that actually has an impact and drives things versus the other way around, in my opinion.

Des Traynor: 

I totally agree. It's both unpredictable and exciting. And then also just technically, because we're all hacker nerds types so, it's just fun. The last time I remember this is, do you remember Web 2.0? It’s like when the JavaScript libraries were coming out of nowhere, Ajax was released, and Google Maps came out, and del.icio.us and Flickr, and every single day it was like what else is new? What else is possible? 

You can animate things on the website and we were getting so excited. It's that sort of childlike joy that has come back to looking at software products. I wonder what sort of dope shit am I going to see today that I never would've thought was possible. Whereas there was a while when it was like the way you scan a lot of receipts, well, check it. You forget everything but I'm like, dude, I've scanned plenty of receipts. This isn't that interesting. 

Wade Foster: 

Everything was derivative. 

Des Traynor: 

Or they were taking the existing tool and finding a really weird niche or vertical for him. It was like, I don't care about dentistry for farmers.

Adam Hardej: 

It sounds pretty lucrative. But the interesting part though, and the thing that you guys all touch on, is this new kind of state of software, for example, and building it. They've been doing a lot of interesting, unique stuff for, right? That's always been their thing, but one of the things you mentioned about seat based pricing and, you know, the “I don't really need it” when you get bigger, but that's software margins, right? That was kind of the VC. That was the promise and saw from an investment perspective. It was like 95 percent margins because I copy and paste, and I charge you per seat for the same thing I was doing before. It’s exciting but is it more cutthroat? It feels like, like when, you know, when the customer wins and the customer can change really quickly between like, “Oh! This is solving the problem for me. No, not anymore. I'm not on a contract. I'm on usage. See ya.” Is it where the economic benefit falls? Is that like looming at all or is it you don't even care and it's just serving customers better and that's that? 

Des Traynor: 

I would say a few different things. 

One is, there is some, how would you say in this world, illegitimate revenue that doesn't make sense anymore because either the startups are going to come at it, or people are going to realize they don't need to pay for it, or they demand usage-based pricing, or whatever. So I think there's a lot of that. 

This is carved away in different ways and there will be repercussions for businesses, specifically ones who aren't agile enough to get ahead of the movement and instead, get caught asleep at the cash register. The thing I say or the value is—what's the margin you have to make—that is going to be on the outcome you deliver. So right now, for example, we're charging a dollar for support resolution. There is plenty we could charge; we could charge a lot more. 

As in, an average person is definitely employed and if they do like $40 a day, that's a good day. They're also paid more than $40 a day. So there's a lot of wiggle room there to play in. But I just think there's still credible, strong SaaS level margins to be made in charging for the usage for things that specifically deliver valuable outcomes. It is true that we're coming out of a period where a lot of that wasn't happening. As a result, I think, a lot of revenue might sort of dissipate and then, AI will be somewhat deflationary. I think the idea of being able to nail a really specific niche, like, “Oh! We're building a project management for architects” or something like that can enable you to probably charge a lot of money because you've got zero competition for a very small software footprint. I do think AI massively shrinks the amount of work required to build a clone of that software.

You're going to see other types of competitive angles emerge where the ability to copy software is just going to get easier and easier and easier. I just think like you'll be left with moats of security, reliability, relationships, brand, platform, integration, all those things will still matter but we're the only people bothered building whatever, like project management tools for architects. That's not going to cut it anymore. This is where the one person 100 million company might come from. But I think the margin exists as long as you're willing to find an area that's valuable enough and then work out what the person is currently paying all in all the human costs. The overhead is fully loaded. You can definitely make a lot of that money back and still come on a great margin. It's still going to be a low price. 

Wade Foster: 

Solving customer problems never goes out of style. Are there going to be changes to business models? Are there going to be ramifications? One hundred percent and that'll get figured out, that'll get sorted through but, at the core of this, it's just about making something that people want and something one cares about. 

If you do that well, there's going to be a business for you to have somewhere, some way. The structure of that business might look different on the other side but I think that's something you figure out when you solve for your customer's needs. 

It's scary for incumbents because you've gotten so used to making your money one way. Now you might have to do it some different way, but at the end of the day, that's what you are. That's what we're all in business for; it is to solve customer problems. And if you forgot how to do that, then, somebody else is going to figure out how to do it instead. 

Benjamin Gammell: 

Yeah, I think the only thing I would add is, if you're relying on contracts as a way to save your revenue, then you're going to be a melting ice cube, right? Because contracts will come up, you'll get ripped and replaced. To Des and Wade's point, you need to be solving customer points, like pain points all the time, continually throughout the period. 

The shift, whilst yes, contract revenue is a lot easier to predict, and there's some cushion there probably that, that maybe the B2B software sort of market has benefited from, but I think, aligning your product to a customer's needs and how they use your product, will mean you still have all the moats that Des was pointing out. You're still hard to rip and replace. 

If everyone in your company is using the same solution, whether it's usage-based pricing or contract based pricing, it doesn't really make a difference. It's the usage of the product that helps you retain that customer. That's the pricing construct. 

Adam Hardej: 

Are you thinking about how you can expand your offerings or are you thinking about how to go deeper? Has that changed? Again, because you maybe can spin up a product faster than you could before. So is that an allure? Is that a distraction?

Des Traynor: 

The opportunity is there, but we wouldn't pursue it. We're very hell bent on being the best AI customer support platform.

Wade Foster: 

We're pretty singularly focused on our mission, which is, making automation work for everyone. Certainly, within those bounds we're going to explore all the different mechanisms that exist, like how we go about it is what's different. 

When we look at our staff and the people we want on the team, we want folks that have a deftness with these tools. We want folks that know how to utilize this stuff. And we're generally not going to tolerate folks who are operating the old way because they're just not going to be as good, creative, or as novel. We're investing in our people to learn how to do these tools, giving them time and space to do it. And when we hire folks in, we're looking for that. That's going to create a bit of a shift in the market because I do think we're not alone in that choice. We may be early, but that's what's happening here.

Benjamin Gammell: 

My sense for Brex is similar to what Des said a while back. We want to continually ask the question, “Is this a problem that we need to be solving for?” “Is Brex uniquely solved and positioned to solve for us?” Sure, there is a world of things we could go do because of this but for us still remaining super focused on our core products, our core customer base, is important because whether you can do something or not doesn't mean you should do something or not. I think we're pretty regressed around that. 

Adam Hardej: 

That makes a lot of sense. I guess maybe as a fun ender then if you weren't similarly in hacker news and product time, what cool stuff's going on? Anything that is specifically like jumped out at you? I used Perplexity the other day to write some of the memos for some investments and it was fantastic. Have you found yourself any—whether it's a big known tool or a smaller tool—that you've really enjoyed? 

Wade Foster: 

A smaller tool that I came across recently is HeyGen. It does these AI powered video creations so you can make an avatar of yourself. It's pretty impressive but I will say this, the funny thing is how quickly this space is moving. A year and a half ago, you thought summarization and creation was this greatest thing. Now, as a consumer, I'm like. “That's all you're doing? It's funny how fast it is.” 

Des Traynor: 

I'd say one company I invested in and that I've just been blown away by their execution is Synthesia. They do generated video and generated video characters for producing videos where you give it a script and it'll give you talking heads that look very reliable and believable. It works and it's like an open ended script. It’s like these movies don't have to end, they can be shot from multiple camera angles and it can be “choose your own adventure style” stuff.

We've been thinking about applications of that for all sorts of training, onboarding educational content and all that sort of stuff but I think, the very nature of video or books for that matter, ee now need to think about these things as never ending, multiple camera angles, multiple different paths through a story, a video, a song, through anything, like the entire creativity that's going to unlock through like AI. Just the actual generative aspect of it is breathtaking. 

We're scratching the surface right now in the same way I said the other day like the first movies were shot from a single camera pointed at the stage because that's what a theater play was. That's all anyone knew. It took a while for someone to go, “We could try two camera angles and we could try different angles and soundtracks.” That's where we are. We're still shooting one camera. When we think about all of these creative art forms, yeah, I think the next few years are just going to be bananas. It's possible Taylor Swift's third album from now is released and I hear it in an Irish accent and you hear it in an American accent. It's possible that there's a “choose your own adventure” through the songs, lyrics, tone, and through all this shit. We're going to see how it all plays out, but it's going to be crazy.

Benjamin Gammell: 

I definitely like the film spaces. It'll be super fascinating over the next couple of years in terms of what you can do with AI, especially around animation. One area different from an industry that I find fascinating is cyber security. It's one thing that we think about fraud—obviously a lot in the company—but this is a different element. What cybersecurity companies are doing around AI is just super interesting as this field, just by virtue of the nature of the human element involved in cybersecurity and how AI can help there. There's some really interesting companies doing a few things in that space that I just find very mind blowing. It always reminds me that everyone knows far more about me than I think they do, so, it's very humbling. 

Read more from

Read more from

Brex revenue, growth, and valuation

lightningbolt_icon Unlocked Report
Continue Reading

Read more from

Zapier revenue, growth, and valuation

lightningbolt_icon Unlocked Report
Continue Reading

Read more from

Hugging Face revenue, growth, and valuation

lightningbolt_icon Unlocked Report
Continue Reading

Hugging Face: the $70M/year anti-OpenAI growing 367% year-over-year

lightningbolt_icon Unlocked Report
Continue Reading

CoreWeave: the $440M/year cloud GPU startup growing 1,660% YoY

lightningbolt_icon Unlocked Report
Continue Reading

Read more from

ShipBob revenue, growth, and valuation

lightningbolt_icon Unlocked Report
Continue Reading