[00:00:00] INTRO: Welcome to Unpack Pricing, the show that deconstructs the dark arts of SaaS pricing and packaging. I'm your host, Scott Woody, co-founder and CEO of Metronome. In each episode, you'll learn how the best leaders in tech are turning pricing into a key driver for revenue growth. Let's dive in.

[00:00:23] Scott: I'm excited to chat with Matan. Welcome to our illustrious podcast. With this one, what we like to talk about is the art of pricing especially in modern, fast-growing businesses. And Writer is one of the fastest-growing businesses I've heard of.

So I would love for you maybe to do a quick intro about your background and how you got into the world of pricing and then also how you landed at Writer. 

[00:00:47] Matan: Yeah, for sure. I actually started my career in government. I worked for about four years for the treasury department on like, healthcare and social security policies. I spent a bunch of time building like, you know, insurance plans for the government and pricing those, and incentive structures and so on and so forth. After that, I was like, okay, I want to try something else. And I went to work at the algo trading hedge fund which took about a month for me to discover that I hated my life while doing that.

And then took another 11 months before I left it. At that time I moved out here to California. This was in 2012. And a friend actually introduced me to John and Patrick Collins from Stripe. This was right after their B series and I joined their working on the risk team. So this was in the early days of, you know, payment and like what would become eventually like more like fintech and so on.

So, I kind of like stumbled into that space by chance. Spent a bunch of time there and after that, just like stay in the fintech space. I was at Gumroad, if you're familiar with that company. And then at Square and funny enough, through my roles, I started tackling more and more I guess, products in the developer space actually.

So Square, I led the public API and the platform team there. And then moved to Flexport where I was leading the marketplace team which also has a large pricing component and algo pricing element. And then moved to Brex and most recently before Writer, I was head of product for a company called A.Team, which is a talent marketplace.

And at some point, a friend introduced me to Waseem, who's the CTO at Writer. Founded a company called Baseten which is also in the AI space. And I started meeting Waseem. After a bunch of meetings, we decided that we really like each other and we don't just want to hang out when he's in town, but also want to work together.

And I joined Writer basically to build what's called AI Studio, which is, if people are familiar with Writer, we are a full stack LLM provider for focusing on global 2000 and so enterprises. We have, I would call what the core app is more of like, a SaaS offering with everything that comes with a full stack.

So our models, the LLMs, the rag layer, the guardrails, and also the UI to interact with it. And AI studio is the piece where you can actually build applications, whether it's with no code, low code or full on APIs, and then deploy them either into the Writer environment or anywhere else you want in your enterprise.

So what, that's what I've been doing for close to a year now. And also, not surprisingly, with new models coming and offerings, there's a lot of pricing and pricing considerations in that space. 

[00:03:42] Scott: Awesome. Okay. Well, since we are a pricing podcast, and you come from a deep product background, how do you look at pricing from a product manager lens?

Like, where does pricing fit into the job of being a great product leader? What is your strong opinion there? And then how, how do you see that changing over time, especially in these AI based businesses? 

[00:04:03] Matan: Yeah, that's a great question. I think… So first off, I think pricing is… a lot of it is an art and like there's a lot of trial and error.

Mostly errors until you nail something. I think you're also somewhat limited and constrained, especially in areas where you have competition, so you can't do whatever the hell you want. Obviously. Where I think it's really interesting is there's a lot, a bunch of layers to it.

One is where you are in the product lifecycle. So that's kinda like, you know, if you are like, early company, you need to get users. You might even give things for free. Or like, even pay for users at first, especially if it's like a B2C. If it's a B2B, you might say you get, like, X year, X months for free, just come try, give me feedback. So it really depends on that cycle. Where you see a lot of trouble in pricing, or in the pricing world when it comes to product, is I think a lot of times it's an afterthought for a lot of PMs. And I think, well, that's a problem because that's essentially what you're making.

That's your P&L. And, you know, a lot of times P&L is like, okay, my job is just to crank out features rather than like, okay, features... building the features actually depletes the easy part. Once the feature is live or the product or whatever, how do you actually... there's the grind of how do you make money, but obviously you get engagement and usage on that experience.

I think that's a lot of times something that tends to be forgotten and PMs are like, oh, this is the PMM's problem, or this is the sales team's problem. No, no, that's like, if the problem doesn't serve, we all have a problem because we're out of a job soon. But it's also, like, to be very clear, it's not like that the PM sits there and makes the decision by themselves, it's like, it takes a village, so there's a lot of considerations.

There's obviously, for example, I might want to undercut the market and, like, buy something very low, but then they're like, oh, our investors, there's a consideration on margins, there's a consideration here. So there's, like, a lot of these considerations that come into play. And sometimes your opinion is what gets heard, sometimes you negotiate and reach a different number, sometimes you get overridden by, let's say, a founder, and that's every... All of these are fine. So to me, it's just another facet of the product that unfortunately, a lot of times, tends to be forgotten.

[00:06:11] Scott: Why do you think that's happening? What is the kind of fundamental mistake that product managers are making? And then maybe, why do you believe, like, why have you not made that mistake or kind of embraced this, like, kind of idea so strongly?

[00:06:25] Matan: So I have made that mistake a lot of times, let's be very clear. I think it's like, it was more like a trial by fire of after you make a mistake multiple times, you either learn or yet there's a bigger problem at play. I think in some sense, AI is easier right now because no one knows what they're doing around pricing, so you kind of, like, fake it till you land on something, so that's kind of easy. You know, when I was in fintech, especially in payments, there's only so much you can do.

Like, if the dominant player is Stripe and you're not at Square, you're not going to charge more than Stripe. Unless you really provide outsize value, which you might, or you might decide, you know, even though I provide the outsize value, I still want to capture a market and so forth . So I've always worked in areas that people price shop and look at the competition and payments API as hard as they are, like you can migrate off of it, it's not fun. But like migration and like MasterCard and Visa, especially now are making it easier and easier to migrate off the processors. 

So like the competition there is really hard. So you can't price gouge kind of thing. So you, you're kind of like constrained by the market. But I do think where I see this mistake, especially in SaaS components, is that it's, it's really hard, you know, like if you think about the way SaaS is priced, and I'm sure you guys have somewhat similar fixed plus variable pricing the fixed piece is really hard because you bundle a lot of things in there.

[00:07:49] Matan: So, what's like... let's say I have a new feature. Some users are on existing contracts., So do I roll them into these users or do I charge like an extra SaaS component? But then as a PM, I want to have usage on my offering 'cause I want to show that there's value in what product that I get. So there's that inherent tension between like, I want people to use this shit I build, but then the commercial side is like, no, we want to get a higher ACV or whatever.

And it's a healthy tension, but like, I think a lot of times, pMs by default kind of like fall to like, no, I optimize for usage where go to market was like, well, we need to make money because like, that's our commission and whatever. So I think that tension is like always 

[00:08:29] Scott: there. 

Yeah, I agree.

Sometimes you want to create the tension and you've worked at some places that I feel like kind of really understand. Like, I mean, they're very big, successful businesses: Square, Stripe, you know, Writer. How... how do you advise, like, leadership teams to kind of create a tension that's healthy and make sure that, neither side of, like, usage or... or commercialization has too much power?

Have you seen any, good models for making sure the tension is properly balanced or how do you even know if [00:09:00] things are properly balanced? Like, what are the indicators that things might be out of balance? 

[00:09:04] Matan: I don't even know if I have an answer to that, but I'm looking, for example, when I joined Writer...

So Writer has a fixed platform theme, then we launched AI Studio, which introduced the concept of usage-based pricing, very similar to other AI players which was kind of like, okay, still to this day, the norm and like in the market, because we were faced with a need to just like pressure from the market it's like, Hey, I can test this other company's APIs. Why can't I just like, I just want to mess around and see if it answers my need. And in our approach that was like, Hey, there's a pressure from the market, let's go do this. We'll price it as following, you know, we know our underlying cost structure, let's price it, we're not going to price it more than let's say Anthropic, OpenAI, or Gemini, whatever.

But in some cases we'll price it in between depending on the evals and how our models are performing. But we're also maintaining the right to that we understand that this market is moving really, really fast and we can't get it right. And we will probably need to change our pricing, the whole pricing structure.

Now, the problem here, and this is like actually somewhat nuanced. Our assumption was, as we go out to the market, is that compute costs are going to go down. So we're not, when I say change the pricing, we're not going to go up with our price model. We're only going to go down when it comes to token usage.

Where we, what we did, that's the known known. The known unknown was like, we actually predict that within a year or two, we'll have a completely different price model because the market is evolving so quickly. We talked about this before, we're actually going to change the platform the way we're pricing because it just does not cover the ICP that we're targeting which is the Global 2000 CIO and their needs. This usage-based pricing is extremely problematic when you come to enterprises. And we're going to evolve off of it. So, now there's a whole pain of how do you migrate contracts and so on and so forth. So for a duration of time, we'll have legacy contractors, net new contracts. But that's part of the price of figuring this out.

And this is not unique to AI. This has happened in cloud. This is happening in a lot of industries as new technology comes in. And we try to fit into an existing paradigm of a mindset as a new technology comes out. And that is actually the harder part and like say, Oh, we're going to charge $5 a million token versus two and a half dollars versus $10. That's more of like, okay, let me take like X percent margin and I'll be fine with it. Plus what's the competition. 

I think the issue is like more really understanding your customer persona and their pains and being like, actually saying something which I would argue and this came from our sales ops and together with product which were like the pricing model we have just does not work. What the market does for our ICP does not work and does not scale. 

[00:11:53] Scott: Would you mind elaborating a little bit on like what about the existing model like this like usage per token per million token model? Like what wasn't working. As you know there's not that many people selling to the global 2000 like.

But what about that enterprise, that large buyer, made the kind of per token pricing not work so well? 

[00:12:12] Matan: So I think it works when they want to do like experimentation, a bunch of engineers in your organization would just want to mess around, maybe put it on a credit card. But at the end of the day, when you sell to these large companies, you have like somewhere, someone has like a budget line item. And the problem with... like, if you look at credit cards, the space, let's say Stripe or Square or PayPal, whatever. It's actually really easy to migrate someone like, I'm not even talking about the technology layer, I'm talking about the commercial piece because you're saying okay, give me your last whatever... 10 year transactions. You look at the bins. You look at the countries. You look at this. You look at that. And you give them some blended pricing. And you say, okay, given the patterns I'm seeing here, I'll give you, like, the rack rate is 2. 9 plus 30 cents. I'll give you interchange plus whatever. Five, 10, whatever, depending on the [00:13:00] volume and the mix. So if you have more American Express, I might charge you a little more. Versus if you have all debit card, which is like significantly less than a credit card.

So it's quite, quite, quite simple. There's a somewhat certainty. Where the problem with AI is, so I come to a CAO and he says, Hey, Matan, this is the use case we want to tackle. We have these types of docs, this type of information, and this is what we want to do. It's like, okay, great. And then he comes to me and is like, how much is it going to cost me? I'm like, I don't know. Because the problem is, tokens is this, you know, on average, you have three tokens a word in English. Great. Like you can run a business unit with averages to think. So you can, okay, I have an idea. It's going to cost me a million dollars. And then you start using it. And all of a sudden prices start going up to like $5, $10, $20 million. And then I go, holy shit. 

So then your champion, which you need a champion in the enterprise is now on ,the stake because they're like, I just signed a contract and like, I thought it's going to cost X and it's costing me 10X. And that's a real problem when you see this across the board. So then two things happen in AI. One which is less related to pricing, but has some elements of it. You know, they do like a 10 million commit, let's say, and you see this with Anthropic, OpenAI and other, then you hear that they've only deployed two, three apps into the wild, into that production. They're in POC land. So they're nowhere close to consuming their tokens. That's related to, they don't know how to roll this out, they don't know how to do this, but they're also scared shitless. 

The other extreme is they build all these apps, they roll them out, and then they wake up one day and they're like, why the hell is it costing us so much? Like, what's the efficiencies I'm gaining here? What's the ROI? No one can answer that question unless you agreed on those things up front. So for us, this became a real... a real big problem because we're like, we want you to start using the platform, but we don't want you to be scared shitless. And, and that's the change. The change is essentially we're moving to a place where we're going to have a fixed platform fee and tiers that give you access to certain features for the full stack including observability, including agent management, all that stuff, and the model access, but we're going to bake in as part of the platform fee, like a monthly token usage that you get out of, like, for every month.

And that monthly token is actually quite generous. I think when we calculated, we basically looked at the existing tiers and added a pretty big buffer there that you can eat into. 'Cause our assumption is as you use us more, your usage would grow. So we don't want you to like do one more app and then like, Oh shit, now I need to pay. So then you have this buffer that allows you to really experiment, play, and I'll be scared that you're going to wake up with a massive bill cause it's baked in. 

Now the question is, what happens if you're about to breach that? And that's a separate conversation. That's a conversation that the commercials will, the go to market team can handle on like, buy another packet or increase the platform fee or there's a bunch of levers that you can play with here. Buy another commit. That conversation is actually easier for us because as we work with customers and build applications, we upfront talk about what is the ROI? What are the metrics that we want to drive? And that's part of the conversation that's baked in to the sales process of Writer. So then, if you consume, let's say, a million tokens, that was your cap, or a billion tokens, whatever, a month and you're hitting that limit, our conversation is not like, you need to buy another billion. It's like, let me show you what that billion is used for. Let me show you the value you drive. And that's like, as a champion, I am arming you to go to your procurement, to go to your boss, to go to your board. It's like, hey, look what's going on here. I need more capacity. I need more budget. And it's just like a much easier conversation to have.

[00:16:39] Scott: Yeah, I mean, what you're saying is exactly, you're basically, you're solving the procurement problem by giving people predictability, but you're allowing variability in workload and you're allowing like incremental organic value discovery that then your sales team's job is like, yes, let's get the initial commit, but really I want you to be.

I want to make sure you're actually using the command. I want to, if I'm giving you a billion tokens, I want to get as close to the pin as I can. And if I think you're on a trajectory to go above it because you brought on this new workload or this new app, the frame of the conversation is not go get more money. It is more like, look, oh, wow, you found this new value. 

How do I go arm my champion to go ask for more budget to kind of unlock this incremental value? And then also the other thing, the other nice thing about this model is it lets the product bake in the org for like many months or quarters before you even need to really get into the details around like the new workloads and stuff like that.

It's like by the time they're crossing this, like, you know, token threshold, you understand, and they understand, is this actually hitting in production or not? And if it's not, and it also gives you time to course correct, right? If in like month one they're only hitting 10 percent of their commit, but the next month it's like 30, then they're at least on a good trajectory and then you kind of can budget against.

[00:17:57] Matan: And they actually don't pay for, to be very clear, they don't pay for the commit. They pay for platform features, meaning... so the commit is kind of like, that's on the house. Kind of like when you go to Costco and you get a 1 plus 1 deal. That's like, that's a plus 1. You get it as part of this expected to allow you to play.

But what you're actually paying, you're paying for let's say higher rate limits. You're paying for better security. You're paying for observability and control, so you pay for enterprise features as you go up the scale of the platform fee. And you're getting the token increase, it's kind of like that's on us.

So you can start messing around over e, 

[00:18:33] Scott: Exactly. It's like, almost like the token limit is just like... it's like a premium feature or it's like in the subscription world. It's like you get a certain number of seats and then if you go above, the next plan gives you twice as many seats. And it's like they're not nickel and diming you if your seat goes up by one. It's like doesn't it's fine. It's an entitlement. What we're seeing a lot is exactly what you're talking about which is essentially the future looks like these entitled subscriptions where there's like tiered entitlements where they're basically using usage caps as an entitlement.

And then some companies, you know, to your point, you know, some companies then have an overage concept in a more enterprise motion like yours. It's like overages are more of a sign to have an account management conversation rather than trying to collect money. It's like, who cares?

[00:19:17] Matan: And I'm not going to stop your usage on overage. I will never stop. Cause that's like, if you're... Exactly, exactly, exactly. If you're a global 2000, I'm not going to stop you. It's like, I want you to consume. We just need to have a conversation. Like, you can go into debt, so I do some credit risk on you, but I need you to like figure this out if you don't want to have stop issue.

[00:19:36] Scott: Exactly right. Like, you're like, wow, okay, you found this new hot thing. Let's sort of quantify this value in a way that you actually can value. And then, you know, yeah, to me, it's very interesting because I actually think the future of the world, it's like, when people say usage-based, I think they kind of default to this, like, pay go, pay overages, but I actually think the world, especially in the enterprise, looks a lot more like what you all are trending toward, which is it's a hybrid of subscription with usage-based entitlements.

And the usage is useful because you know, this may not be the perfect proxy, but it's a proxy for value that is like something you can have a conversation about, and it becomes this scaffold upon which you can start to build ROI conversations. And so, and you know, I mean like, it's no surprise like chat GPT, this is how they do it, right? They launch a new product and they give you one like, you know, two Sora files on the whatever pro and then 500 on the like professional, whatever it is. That's the kind of thing that I think is actually... I think a lot of people ask, like, how is usage percolating in consumer? And the way I think about it is, is actually just this entitlement stuff.

It's like, you don't want the customer to have to think in terms of the granular usage stuff. It's more like, actually, I just want to buy a subscription. I want the simplicity of that. And then I want to be able to up level when it's appropriate, when I've discovered new value. 

[00:20:56] Matan: Totally. And I think you asked me about the products in here as a PM. This allows me to give you access to every feature and cap it on usage. So, like, I'm not gating. For example, what we were doing historically, is we would cap on, like, our RAG service, and I'm like, why would we... like, we agreed, like, why would we do that? This is a core differentiator. We have built in RAG in the platform, that's really easy to use, and our RAG solution is different than a vector database, yada, yada, yada. It's a selling point. Like, why would I limit people from using it and seeing the power of the platform versus, for example, say, you know what, you can have as many knowledge graphs, you call them... as many knowledge graphs as you want. But where we put the limit is maybe on the number of files you can store on the platform.

So maybe you can store up to 50 gigabytes, but if you're like an enterprise plus plus, you can store up to 500 gigabytes or something like that. But you can still use the functionality in the right functionality with AI to see the power as a means to attract you. If you need more capacity, well...pay. 

[00:21:59] Scott: Yeah, to me, it's very interesting because, you know, I worked at Dropbox and we did a ton of growth stuff and our entire job was like, how do you engineer a product so that it grows, you know, it's like this incremental growth and in a weird way, I just see like all of these AI companies are kind of rediscovering.

It's like essentially growth techniques that like Facebook pioneered, but applied to product and kind of how do you design these things so that, you know, you give them whatever. A small amount of a free tier, but then, how does the usage of that then promote that into like a paid tier, et cetera? So it's super fascinating. 

[00:22:31] Matan: I remember Dropbox in the early days was like, invite another fan, get 500 megabytes. And like you were chasing the storage that you don't need. 

[00:22:40] Scott: Well, exactly. And it's like, people are like, well, why didn't you just charge for every gigabyte or whatever. I'm like, because if you understood the economics of this business and you actually charge that way, this business would be worth nothing.

We, you know, we had to design the exact gigabyte threshold coupled with price very precisely to make sure that 99 percent of users were free to serve. And then the tail users, those cost a shitload of money to the business, but there were just so few of them that it worked itself out. And so the designing of the entitlement limit, in that case space, was actually the entire game of the pricing model. And it had to be simple, right? 'Cause consumers are buying something and ultimately consumers just want something simple. And I think that kind of drove this lesson home really hard.

I'm actually super interested. You mentioned something, which I feel, I think everyone feels viscerally, which is AI functionality and value is changing super fast. How do you recommend product teams kind of operate in this moment where the value that you can provide, like, especially powered by these AI models could grow pretty quickly, could stabilize? How do you even like run a product organization in that kind of volatile environment? 'Cause it feels, it feels very volatile and I'm curious, like how you run your teams in this kind of environment?

[00:23:59] Matan: Yeah, so I think there's two sides to this answer. There's running it inside Writer or an AI company versus running it in a company that's building on top of AI. And what do I mean by that? So naturally, I have... and it's true for all of our... Writer trains their own models, but Mira, which is our family of models, is one of the top five models according to Stanford Helm, when we look at like benchmarks.

So we have more visibility to what's coming and access to it. So we have the ability, as we do planning, to plan around that. Now that sounds really impressive, but it's also, there's a lot of greater bullshit in what I just said. And what do I mean by that? Because AI development is unlike, I would say, anything that I've experienced in the sense that you might think that something is going really well. And then you're, let's say, a month out from having a production ready, like the model or something like that, and then you wake up and you like, you hit a wall and just like this whole branch of development is just a waste of time. 

Now it's not a waste of time because you've learned, but you're nowhere close to the time as you thought. It's not like building Like a deterministic, you know, product. You're like, okay, I'm connecting to this bank or that bank or doing this or doing that. And, that's what you're seeing. You know, everyone was talking about GPT5 and GPT5 did not come. It's not that there's not research and a lot of effort there. It's like something happened in the development cycle and that's okay. That's like, it's not that it's a cop out. It's just like, there's a lot of uncertainty in this space. 

And so it could be that the model is ready, but maybe the guardrails wasn't working properly, or the cost per token was like insane or whatever. And I think there's a lot of that, that challenges. And so that's one thing to know. So like having a visibility does help as I get access, or we get access to newer things that we can do upfront. 

I think the other thing is where, what we're trying to do and what we are doing in Writer is we train our own models because honestly there's no alternative. We cannot build an enter a sustainable enterprise business on the third-party models where the models change under you on a regular basis, the labs quantitize and distill which basically what this means is they take them, like, the analogy I love to when I talk to customers is, think about it, you know, something happened, you have a windfall of money, you go and buy a Porsche because you always wanted to have a Porsche. You drive it for two, three weeks and the Porsche is amazing, you know, zero to 60 in like three seconds, whatever. Amazing drive. Everyone is looking at you in the street as you drive in. It's like, holy shit, this is a great car. Awesome. You're park in your home. You go to sleep, you wake up, you try to drive it again. It's not behaving like you expected. You open the trunk and you find out that someone switched your engine from a Porsche engine to a Kia engine.

Now Kia is a great car, but it's not a Porsche. So, what distilling and quantization means is these labs, what they do is they have these amazingly powerful models and two, three weeks after they launch, they actually look at the usage and how people are using the inputs and the outputs. And they collapse the model. They collapse it and they fine tune it around the use cases to reduce their cost structure because they cannot run these models at full capacity. This is a problem. If I want to now embed my applications or my agents in a bank, if the underlying model changes, that's a huge problem. 

The other thing is the whole issue on copyright. I cannot answer how the training data was structured in these models. So there's a bunch of these reasons why we train our models that gives us that. But these view models, honestly. Price per token is going to zero. And over time, it's just accelerating and accelerating and we're seeing this on a daily, weekly basis in AI.

So our value is not just in the model layer that I can guarantee that they're forward and backward compatible, that you don't need to change your prompt strategy every time we have a new model, but it's in everything that's built on top of that. So it's like this full stack solution that you don't need to come to work with duct tape to kind of like get an AI solution in your company, but you can come in and like really start driving change from day one.

And I think this is critical because where there's a lot of struggle with AI in any enterprises, everyone wants to use it, because they use it consumer, and they love it. But the CIOs are really struggling on what it means to deploy this in scale in the organization. So, what we're trying to do, and what we've been doing with AI Studio is, I wanna empower the CIO persona, the IT persona to not be an AI detractor, but be an enabler to be the champion in the company that distributes it and manage this in scale, because that's how these companies work.

And that's very different in other foundational shops that started as research to consumer where they make the bulk of their money. And now they're trying to dabble into enterprise. And as you know, Scott, from your guys' work, it's not enough in enterprise that your core product is great, whether it's modeled or not. There's all these secondary things that have to happen to even go through InfoSec, to go through the CIO, to go through procurement that have nothing to do with your core functionality. But without them, there is no way that a Global 2000 would use you. 

So our view is essentially that, you know, as we think about this product development, it's not just about models or then observability. It's about security, scheme, thoughts, thermal, transparency in pricing. The models have to be able to like be forward and backward compatible. And there's all this complexity that comes into it that are not yet, but I'm sure they will be more and more so in the mindset of these like other foundational shots. And that's okay. 

So what we're doing is as we layer these experiences on top of the model layer, that's where I'm able to extract value. Obviously the models are great and without good models or great models, you won't be able to even do what's on top, but I can charge because I give you like all these other features and experiences that are built in this foundational model layer.

[00:30:12] Scott: Right. In a sense, it's like the thing that someone would have to replace you with is not just access to the model. That's like part of it, sure, but it's everything else. It's the control, it's the certainty, it's the uptime, it's the security, all of that stuff. And in a way, it's like, they're trusting you to handle this infrastructural pieces and provide them a service or a product that is more stable than building on like the baseline model coming out of an LLM foundation model.

[00:30:41] Matan: Yeah. I mean, it's similar to Stripe or even to you all, which is like, can I build a subscription? Sure, I can. Do I want to? Absolutely not. That's not my core competency. It's a really hard problem. Same with Stripe. Can I go integrate directly with Wells Fargo? Sure. But it costs you money. You need to know what ISO messages mean. And also, by the way, you need to have PCI and all that stuff that costs money. No, I want to focus on my product. My product is AI. My product is like, I don't know, a marketplace for whatever. 

So I think you pick and choose where you want to invest your time and effort and I think that's like a key component here for us, which is at the end of the day, we're here to be the platform where enterprises run their AI in scale. And yes, you do that by using models and our models are great, but the value is everything that's built on top of that as well. 

[00:31:37] Scott: Yeah, it's almost like the model is almost analogous to like a data storage technology. It's like, cool. It's like, yes, different databases work in different ways and they have different properties, but in some sense, the value that accrues to most users is like application layer logic that's on top of it. And you're kind of turning this like thing that's like this uncontrollable beast of the LLM into like a tool that can be used in lots of different contexts. 

Actually, I'm curious Since you're working with the biggest of the big, how real is the hunger for AI in these companies? Like, it has to be real, or at least I feel like it is, but, where are the areas where you're the most surprised by how, actively people are uptaking this stuff?

[00:32:16] Matan: Yeah, so I think there's a bunch of things here. The appetite is real, the pressure is real, but also we're past the 2023 let me throw a blank check in this, like people need to see that this is actually moving the needle. So we're kind of like past the hype now. Show me, show me results. That's one thing. 

There was the hype, basically the copilot hypes, like everyone was a chat app. I actually think the best and worst things that happened to enterprise AI is ChatGPT because it's the best thing because everyone knows now about AI and it created this hype and excitement.

It's the worst thing because everyone thinks that AI is a chat app, and there's much more to it than that. And so I think that's the first thing. 

Now we're kind of like experiencing somewhat similar hype with agents that at least to this day don't really work. But they will, it's just a question of time. So there's a lot of that excitement in enterprise. We understand a lot of them have FOMO for like missing the boat on cloud and they're like, we don't want to be, we're not going to wait 10 years to do adoption. We need to be at the forefront to stay relevant. And what is surprising is, so I think like, there's a lot of what I would call mundane workflows in the enterprise that are worth a lot of money, like a lot. 

Like I'll give you an example. We work with CPG, Consumer Product Goods. It takes about, I don't know, three months, maybe even a little more to bring a physical good to market. And it's not just, I'm not even talking about the manufacturing. We're talking about everything about marketing, branding, language. And I'll give you one example for something that sounds silly. But let's say I am Acme Inc a cosmetics company, and I have multiple channels, so, I sell on Amazon, Target, Bloomingdale's, Macy's, Nordstrom, and so on and so forth.

Each one of these channels has their own way of dealing with product, getting the product, the language of the product, how many characters can you use, which images, which that. So that's just like one vector. And then you also serve globally, even within the same chain. So you sell let's say in Mexico, but also in Spain. That's not the same language, even though it's Spanish. It's a dialect. And then in Europe. 

So you have this n times n metrics of like product description. And this is just product description. Historically, a human would generate each box there. And it would take them months. Now with AI, you can do this in like minutes. And that's like compressed to something that would take months, maybe compress it by two weeks.

Now you can start chipping away at these like elements, and they're worth millions of dollars. Hours of efficiency. And you're like, this is not exciting maybe, but that's how you really start with doing transformation. And what we've noticed and what we've learned is that we don't shy from like, give me your hardest challenge.

Because if I blow you out of the water with that hardest challenge, my ability to go and expand in the business? It's a no brainer. So like, my ability to go with, and honestly, we are one of the only AI companies that do this- seven figure and more, multi-year expansion that come year over year.

There has been even seven, eight, ten, nine figure contracts in AI, but they have not been necessarily renewed because no one used the credits. So I'm talking about multi-year renewal. You have a one year contract and then there's a multi-year renewal that comes after that and that shows like a real sustainable business because you start with a the non sexy, the hairy, the really really messy use cases, and then going from there and expanding to the whole org is actually not a problem. And that's true for finance for FinServ, for healthcare. So we work in highly regulated industries for CPG, for insurance and so on... 

[00:36:07] Scott: That's awesome. I am curious in a sense, like if you came and pitched, I'm sure like a CPG company, like, cool, I can take this process from three weeks down to five minutes, whatever it is. They're going to be like, cool sure. Like, how do you discover that first hairy thing? And then how do you prove the value to them? Literally what's the process that you all use? Are you doing like a POC? Are you like trying, are you doing some kind of like, Let's find a way to get self-serve kind of like inveigled in and then, let our champion prove this stuff. I mean, I'm just curious how you're getting that initial proof of value in front of the exact buyer and then landing that big deal.

[00:36:45] Matan: Well, nowadays we have like, I would say, a portfolio of successes and needs for specific protocols that we've tackled. So our ability to replicate and to show real customers... by the way, the CIO at X is a friend of the CIO at Y, and they talk. It's a very small pool, so if you poison one, like, you're kind of like in a problematic situation, so if you knock that out of the park with I don't know, Acme Bank, then going to the next bank is actually much easier.

The problem is that initial, and then you do like a lot of POC. But even if you got it well with Acme Bank and you go to the next bank, like, doing a POC is something that makes sense. It's also somewhat lightweight. We work with them to really, and this is the critical part for Writer and for enterprise as a whole, is we help them manage the transformation.

Like the idea that an enterprise would build against an API self-serve will not happen. It will not happen. They need to be handheld because they have their way of doing things and it's been working for them and they're great companies and they want the transformation but they need your help. They need to understand what use cases are better and worse.

They need to agree with you on what the success look like and that's critical because you don't want to come to end of POC and you're like this is not successful. Well, it actually is because this is what we agreed on. And look for the KPIs and look how it's going to do it. So I think these are all conversations that you frontload and you might be like, well, but that delays the sales process.

Sure, it does, but A, suddenly these companies take a while to begin with and B, that's how you build a long, sustaining partnership with them versus like a one-off, here's a large check and then they churn. So that change management is absolutely critical. Like, I cannot emphasize this enough and this is something that does not happen enough in the Bay or in tech when it comes to when you go and deploy to these companies and start working. 

[00:38:46] Scott: Yeah, I feel like the reflexive I mean, you worked at one of the places that I think taught everyone that like Stripe where it's like, cool, everything can be solved with an API and the developer will. And I do think that when you're talking to these larger customers, actually, they really value the human touch. They value professional services. They value helping them transform because they want to transform, but frankly, it's kind of weird and very alien. And honestly, it's not weird for them to be skeptical, because four years ago, this wasn't possible. Like, two years ago, this probably wasn't possible. 

[00:39:20] Matan: And by the way, today, people say a lot of things are possible that are not possible. So there's a lot of noise also in the market of overselling.

[00:39:28] Scott: Exactly. We see this all the time in our industry, but we see it in AI all the time too, right? Actually I was talking with some CFOs recently and they taught me this term of ERR. I'm like, what the hell is ERR? And they're like, it's experimental revenue. I'm like, Oh, okay. You know, things are bad when people are inventing a term like that. And so, with all these AI companies, it's like, cool. Come back to me in a year when the renewals have hit and show me the upsells and the best companies, you know, like Writer. The upsells are like 500%, like these are not small, these are true. Like actually, wow, we really, you know, need this even more and more. But it does require almost a return to what I think of as more of a, you know, enterprise-ish motion because AI is fundamentally properly done. Is that disruptive? It is like it is rewriting the DNA of a business. 

[00:40:22] Matan: Yeah. I mean if you go 10, 20, 10, 15 years back and you think about like a bank migrating to the cloud from on prem, like, it's not a simple thing. It's a mind shift. Like what, the servers are not in my building. How does that work?

And like, it is a mind shift and like, you have to handle them. And like, we've seen this in cloud. There was like the whole digital transformation. It was a reason that a lot of companies made a lot of money on this. And it took a very long time to transition these players because it was not done... it was not done right, honestly, in a lot of cases.

So, yeah. 

[00:40:56] Scott: Yeah, I do, I do, and I think you're right to point out, I think Silicon Valley kind of, in my head... it's like they fetishize this self-serve, PLG stuff, which yes, if it makes sense for your business, love it, sure. I would also love that kind of business. But a lot of business value just can't be unlocked that way.

And then the other thing that I think they overfocus on, and to our point of our conversation is, they really focus on the product value, but then they don't think about how the product value connects to a commercial motion that actually enables you to have a business long-term. 

[00:41:26] Matan: Yeah, and don't get me wrong, I do believe that long-term as Writer, not even long-term. I do believe as Writer shows the feasibility of the solution inside the organization, we see this already more and more engineers within the organization. And I would say technical savvy business users use AI studio to self-serve and build agents by themselves. But like, I don't see a world that a global 2000 engineer counts on Writer, PLG signs up, uses the product, and then deploys it inside the organization without any touch point by themselves.

It's just not going to happen and that's okay because the ROI on these contracts is so high that you can basically do that work and build that relationship. And like, no one would go PLG into like, I don't know, an investment bank or a bank or whatever, without going through security. It's just not going to happen.

So that has to go, the whole motion has to happen, and you need a relationship management, they need all of that and you want that because you want it to be successful. 

[00:42:34] Scott: Yeah, exactly. If you study the largest Snowflake's, Databricks' customer expansion stories, it's like they, by hook or by crook, get in on some small workload, but then the product is a platform that the workloads can expand over time as value is discovered. And in some sense, the value of a general platform is that it can serve a lot of different workloads in a lot of different ways.

And the reason why Snowflake's, you know, revenue numbers look so good and so specific is because they've architected their entire business around this motion of get in, get secure, make sure everyone understands it, and then find new workloads over time and expand, expand, expand.

[00:43:12] Matan: Land and expand. 

[00:43:14] Scott: Exactly. It's like a return to that. And I think very strongly that especially the more high value AI plays look a lot like this analogy that you're using, which I think is exactly right. It's the move from on prem to cloud. If you study how that worked and how, like, especially and consider us to be in like... we're in 2005 or something like that, which is like, we don't even know the nouns of the right ways to talk about the way to integrate a hybrid on prem cloud solution. And I think that in AI, we're kind of in a very similar place where like, yeah, that global 2000. You know, I barely know how we should integrate AI into our company. It's like, how could you expect a 50-year-old company to be able to make that transition on their own? It's like impossible. So I, and especially with all the noise and snake oil in the water, too, I think it's like... 

[00:44:01] Matan: I also think, to your point, I would actually go even, you're being a little polite to the Valley, so I'll even go one step further.

We have this mentality, which is great by the way, that we think everything that was done before us is stupid and we know best. So the sign of sin is great because it allows us to dare where others didn't. But there's a reason why these companies are successful. Like erasing everything they've done is so like arrogant and misguided versus understanding that we need to come and partner with them and learn what they're doing.

And I would actually say, if I look at the arc of AI, if I have to guess, you know, we'll talk about this called pilots. Now you're going into more business critical workflows. And as these models improve, they have better reasoning whether it's self-evolving or not, or whatever. I think the next step, you capture more of these, you learn how the enterprise works. You then go to a place where you start kind of like creating the net new neural pathways or workflows in the organization and finding these optimizations. But the idea that like, for example, you can employ 9,000 agents in a bank and you find the optimization is insane. No one is going to let you do that.

It's going to kill, no, you're not going to, it's not going to work. Like, And it's not gonna work, and also they're not gonna let you even come through the door. 

[00:45:17] Scott: No, exactly. Exactly. Awesome. Well, I think that's a great place to end it, Matan. I think this was great. Like, you have great, spicy opinions, and I really appreciate it.

And yeah with that, I'll say goodbye. 

[00:45:30] Matan: Thank you so much, Scott. 

[00:45:31] OUTRO: Thanks for tuning into this episode of Unpack Pricing. If you enjoyed it, we really appreciate you sharing it with a friend. We'd also love to hear from you. Feel free to email me at scott@metronome.com with feedback and suggestions for who you'd like to see on a future podcast.

‍