Latka logo

Founder Interview

How Parloa Reached $50M in Revenue and a $3B Valuation Selling AI Agents to Large Enterprises (Interview with CEO Malte Kosub)

Interview Date
August 18, 2026
Interviewee
Malte KosubCo-Founder and CEO

Company Metrics at Interview Time

Revenue (2025)

$50M

Valuation (2026)

$3B

Total Funding Raised (2026)

$560M

Team Size (2026)

500+

Biggest Customer Contract (2026)

More than $1M per year

Historical Snapshot

These numbers were reported by Malte Kosub during the interview recorded in August 2026 and are a historical snapshot, not current figures. See Parloa’s current numbers.

Key Takeaways

  • 01Parloa surpassed $50M in revenue by end of 2025, confirmed by CEO Malte Kosub
  • 02The company was valued at $3B and had raised $560M in total funding as of August 2026
  • 03Team size exceeded 500 people, with roughly 200 working in engineering and product
  • 04Average contract value is hundreds of thousands of dollars, with some customers paying more than $1M per year
  • 05Customers include IKEA and Booking.com
  • 06Parloa was founded in 2018 and spun out of a consulting agency sold in 2020
  • 07In 2022 the team rebuilt the entire product from scratch to take advantage of large language models
  • 08Malte Kosub guesses Parloa passed $10M in revenue somewhere in 2024
  • 09Asked whether Parloa can more than double revenue in 2026, Malte Kosub calls it the clear goal
  • 10Parloa fine-tunes speech-to-text, large language models, and text-to-speech models for enterprise customers

Company Metrics at Time of Interview

MetricValueSource
Revenue (2025)$50MFounder interview, Aug 2026
Revenue (early product stage) (2020)$200K to $400K ARRFounder interview, Aug 2026
MRR (early product stage) (2020)$20K to $30KFounder interview, Aug 2026
Valuation (2026)$3BFounder interview, Aug 2026
Total Funding Raised (2026)$560MFounder interview, Aug 2026
Team Size (2026)500+Founder interview, Aug 2026
Engineering and Product Headcount (2026)~200Founder interview, Aug 2026
Average Contract Value (2026)Hundreds of thousands of dollarsFounder interview, Aug 2026
Biggest Customer Contract (2026)More than $1M per yearFounder interview, Aug 2026
Year Founded2018Founder interview, Aug 2026

Growth Breakdown

Revenue

At the 2020 split from the consulting business, Malte Kosub puts the product at roughly $20K to $30K in MRR, or maybe $200K to $400K in ARR. He guesses Parloa passed $10M in revenue somewhere in 2024, and says it had already reported $50M in revenue by the end of 2025. In August 2026 he gave no newer figure, saying only that Parloa was adding more revenue than ever before.

Customers

Early customers were signed as pilots at $40K to $100K, with the product built alongside them. The company now serves large enterprises including IKEA and Booking.com, with average contract values in the hundreds of thousands of dollars and multiple customers paying more than $1M per year.

Team

Parloa grew to more than 500 people by August 2026, with roughly 200 focused on engineering and product. The go-to-market team spans Europe, including Germany, the UK, France, Spain, Italy, and Denmark, as well as the United States.

Funding

Parloa bootstrapped through its early years before raising its first funding round in 2022. The company has raised $560M in total and was valued at $3B as of the August 2026 interview.

Growth Strategy

Enterprise Sales Motion with Pilot-to-Platform Expansion

Parloa began as a consulting company with a sales-led motion: its first customers were large companies, but on smaller pilots of $40K to $100K, and the product was built together with them. Malte Kosub calls the first million-dollar deal a big breakthrough, because it showed the platform could be proven across use cases, countries and domains, not just in one use case in one country.

Full Product Rebuild Around Large Language Models

In 2022, with roughly 20 to 30 customers live on the old product, the leadership team made the decision to throw out the existing voice chatbot product and rebuild entirely from scratch to take advantage of large language models. The company did not ship new features for roughly a year, but launched a product that Malte says significantly outperformed competitors.

Proprietary Language Factory and Fine-Tuning Infrastructure

Parloa built a language factory that fine-tunes speech-to-text models, large language models, and text-to-speech models for specific enterprise customers and languages. This is particularly valuable in Europe where smaller languages are underrepresented in foundation model training data.

Voice-First Low-Latency Architecture as a Differentiator

Because voice requires sub-one-second latency, Parloa built its multi-agent framework and guardrails for low latency. The company maintains reserved capacity across regions and load balances to guarantee P99 latency below one second, which Malte says competitors who started with chat cannot easily replicate.

Geographic and Industry Expansion

Parloa built out go-to-market teams across multiple European countries and the United States, and launched dedicated industry pages for insurance, retail, and travel and hospitality, with more industries in progress at the time of the interview.

Best Quotes

“So we help help large enterprise companies to deploy AI agents for their customer communication, right? So if someone calls a company for customer support or writes in the chat, our AI agents pick up the call and trying to help the customer with their with their problem. Companies like Booking.com, IKEA, and a lot of others are are relying on our AI agents.”
“So our ACV is hundreds of thousands of dollars. So it's like an enterprise motion. It's not an SMB or mid market motion.”
“we have customers that are paying more than a million per year, yes.”
“We started as a consulting company, right? So it was not PLG, what ac what was actually sales-led. We talked to first customers, directly pretty large customers, but initially those were like rather smaller endeavors, right? We start with forty, fifty, hundred thousand dollar pilots, we build the product together with our customers, and then we we develop the platform further, and then based on that, at at at a certain point we we hit this one million dollar deal.”
“we raised five hundred sixty million dollars. the last valuation was three billion.”
“I think we we were at like twenty, thirty thousand MRR, so maybe two, three, four hundred thousand ARR. but it was really on the side, right? That wasn't like let's build this product slowly, let's see if it works. There was no very focused go-to-market motion for the product.”
“Ten million must have been somewhere in twenty twenty four, I guess.”
“100%. That's the clear goal. We need to be aggressive. We're investing left, right, and center. We're growing our sales teams. And we have the product. So I I always say it's now just about execution, every single team.”
“everyone can show you a beautiful demo, but to scale to hundred countries, three different channels, three different versions of an agent, you're getting to 900 versions of an agent. You need to build a very enterprise scalable platform to do that.”
“that is accurate, but we actually reported fifty million in revenue already end of last year.”

What Happened Next

This interview captures Parloa in August 2026. Malte Kosub said the company had reported $50M in revenue by the end of 2025, had raised $560M in total, and was last valued at $3B. Asked whether Parloa could more than double revenue in 2026, he called it the clear goal: a target at the time of recording, not a result. Visit the Parloa company profile for current metrics and the latest funding information.

View Parloa’s current profile and metrics

Run a software or AI company? Get interviewed by Nathan Latka and tell the story behind your numbers.

Full Transcript

Introduction and Company Overview

Nathan Latka

0:01Hey folks, my guest today is Malte Kosub. He's the co-founder and CEO of Parloa, the Berlin Berlin-based agentic AI company he started in 2018 with his friend Stefan after the two met at a Berlin conference two years prior. He's taken the company from seven hard, tough, long, bootstrapping years, during which the founders put their own money in to a three billion valuation, more than 560 million raised. And he told the World Economic Forum in Davos, quote, that he intends to build a hundred billion dollar company. Out of Germany. Malte, you ready to take us the top?

Malte Kosub

0:31Let's go. Thanks for the invitation, Nathan.

Nathan Latka

0:32All right. How'd I do? Did I butcher any of that intro or are we good?

Malte Kosub

0:36We're good.

What Parloa Sells Today

Nathan Latka

0:38Great. You've been around the block a lot of times. I noticed in your bio on LinkedIn, you started off you actually had an agency focused on voice. I want to dig into your backstory. Before we do that though, let's not bury the lead. What is Parloa selling today? What's the product?

Malte Kosub

0:51Yeah. So we help help large enterprise companies to deploy AI agents for their customer communication, right? So if someone calls a company for customer support or writes in the chat, our AI agents pick up the call and trying to help the customer with their with their problem. Companies like Booking.com, IKEA, and a lot of others are are relying on our AI agents.

Nathan Latka

1:15So the old way was massive call center in the Philippines or in South America. There is none of that with you. It's all digital agents picking up, trained with a voice, you know, a voice language model.

Malte Kosub

1:26Exactly. It's like AI agents picking up the phone, twenty four seven, they are trained on all the processes and the knowledge of the company and they directly answer the questions and actually can also execute end to end use cases, for example, rescheduling a booking or cancelling a booking. yeah.

How Parloa Builds and Fine-Tunes Its AI Models

Nathan Latka

1:44And did you build and train your own voice language model, or are you sitting on top of ElevenLabs or tool calling to a bunch of different foundation models or providers?

Malte Kosub

1:52So we build the harness and we orchestrate different models. we fine-tune them, however. So language is still a huge challenge when it comes to voice agents, right? English is very well trained, but if you look at Europe, there are a lot of like rather small languages which are not on par with like large languages like English. So we are fine-tuning them. We actually build an own language factory. it's an it's an own product we have where we fine-tune models based on a country and and a specific c customer to increase the accuracy, especially on the phone where the quality of the voice is still still a subpar to like like WebRTC audio quality.

Nathan Latka

2:36All hearing about sort of fine-tuning on X and LinkedIn updates these days, but I think a lot of founders don't actually understand what that means. When we see a headline like the other day, Google buying, you know, the bankrupt data from Spear Airlines for 10 million bucks, everyone's going instantly, okay, it's a data play. That's why Mercor was the other bidder. Handshake is data labeling, Scale AI is data labeling. When you say fine-tuning your models, what data are you feeding it in that local language to fine-tune it?

Malte Kosub

3:01So i th there are three parts of where we can fine-tune and where we fine-tune the long the models. First is the speech to text. That is re actually the the the technology that translates an audio file into text, right? And there we actually helping our customers to label the data and this language factory actually is a is a infrastructure. That gets people on the platform, they just label data for a specific enterprise. That's the first part. The second part is the large language model itself itself. That that's the topic where a lot of discussion is actually happening right now, right? You use open weight models, you fine-tune them for a specific specific use case, and then you actually get to a performance or even better performance than a a frontier model with with lower latency, with lower costs. That's the second part we're doing. And the third part which we're doing is actually training text to speech models for a specific voice of a customer, because a bank or insurance company wants to have a specific voice picking up the phone and not one that also others can use. So those are the three parts where actually fine tuning is is being used.

Nathan Latka

4:10When KPMG signs up for Parloa and they say we want our voice to have a specific emotion, an empathy, a character. You know, here's ten thousand voice recordings from the past three years. Please fine tune a specific model on Parloa for us. I mean, is that how it sort of works? What is the training set your customer's giving you so you can fine tune?

Malte Kosub

4:29Yeah. Again it depends on on which of those three parts we're talking about. If we talk about the second part, the large language model fine-tuning, yes, past conversations help to do a first set, but you always need to also use your harness to fine-tune an open weight model so that can that it can work in your harness, right? So it's a bit of past conversation data and conversation data you actually generate with your own system. So it's a mix mix of both.

Nathan Latka

4:57Correct me if I'm wrong here, but I think the technical term for what you're describing when you use the word harness is that that's effectively how you're grounding your model.

Malte Kosub

5:04Exactly. So it's everything you build around the model that the agent runs reliably, d does the right tool calling, you have the guardrails in place, and and and and all the things you actually need to build around the model because the model in itself is smart but it can it can't really do end-to-end use cases in in in our in our domain.

Pricing Model and ACV

Nathan Latka

5:24And how are you pricing for this? Is it usage-based per successful phone call, some other consumption token-based model, seat-based, product-based?

Malte Kosub

5:33it's per consumption. So we believe the more our AI agents talk or chat with customers, the more value we provide. So that's how we how we price.

Nathan Latka

5:44So what is that though? Is it minutes of talk time or tokens consumed via the talk time or what's the length what's the lever?

Malte Kosub

5:51It's minutes or chats. it's not tokens because what we experience in the industry that people can't actually translate the value to tokens so far because they don't know, okay, how big will my prompt be? How many tokens do I actually need? So we need to take ambiguity out of the system for them so they can calculate their business case way easier. And we take on the ambiguity and obviously we have a lot of data to calculate the right price for the right model and so on.

Nathan Latka

6:19And I'm gonna force you into a question that'll be uncomfortable because I'm sure you have people paying all kinds of price points on average. But if I if I asked you your equivalent sort of SAS metric of what the average ACV might be across your customer base, are we talking like, you know, an SMB motion or an enterprise motion at a hundred grand average ACVs?

Malte Kosub

6:36So our ACV is hundreds of thousands of dollars. So it's like an enterprise motion. It's not an SMB or mid market motion.

Nathan Latka

6:42Okay, okay. Do you have anyone that is just so addicted to you they're paying more than a million per year?

Malte Kosub

6:48we have customers that are paying more than a million per year, yes.

Nathan Latka

6:51I love that. I love that. It's a that's a big success metric. When you study all the S one filings from these kinds of companies going public, they always talk about how many million dollar plus customers you have. What did what did it feel like to you when you close your first million dollar per year contract?

Malte Kosub

7:03So it's all from the get go, we wanted to build a a platform for large enterprises. So right, this is a big milestone having an enterprise actually pay you a million dollars for your platform, right? So it was a big breakthrough for us because we showed okay, we can't just prove our platform in one use case, in one country. We can actually prove it across use cases, across countries, across domains. and and and that was a huge proof point actually.

Nathan Latka

7:29Yeah. I know some of you guys are joining now live in studio, got fourteen or fifteen of you joining that are also founders. If you've got questions from Malte on how he did this, his first million dollar customer, drop it in the chat. I'll come back to that in a bit. but Malte, I'm curious, you when you first started this thing, did you go in trying to close your first customer at a million per year, or did you start off very small PLG and then usage just drove them via net dollar retention to a million plus per year?

Founding Story and Agency Background

Malte Kosub

7:52So we you need to we need to get deeper into our story. We started as a consulting company, right? So it was not PLG, what ac what was actually sales-led. We talked to first customers, directly pretty large customers, but initially those were like rather smaller endeavors, right? We start with forty, fifty, hundred thousand dollar pilots, we build the product together with our customers, and then we we develop the platform further, and then based on that, at at at a certain point we we hit this one million dollar deal.

Nathan Latka

8:26Tell us more about this. I've got your LinkedIn highlighted because I love this trajectory. You start off with an agency so you can really dig deep and understand the problem, build empathy for your customer base, and eventually you launch what is now Parloa. But explain to me, this is maybe a good opportunity to dive more into your backstory here. I mean, you have been around the block, you've been doing startups for a while. Where did you get the idea here?

Malte Kosub

8:45So f I was never employed, so I actually don't know how it is to work in a in a large enterprise. I always started my my own companies. and with Future Voice, after doing this e commerce store, we sold it and I wanted to do something more technical because I think e commerce is a lot about marketing and product, not too much about like deep tech. I met my co-founder and we looked into a lot of different technologies. And what happened in 2016-2017, speech-to-text technology. So the first AI algorithms actually like skyrocketed. And we believed back then, okay, voice will transform customer experience forever. It took like six years longer than we expected. so it was like at the beginning, a lot of companies were excited, but the technology was not really ready. You might remember Alexa, Google Assistant. They were okay ish, but it was not really like sophisticated and and and a great user experience. And with large language models, obviously in twenty twenty two that changed. We actually threw away our whole product, which we built initially. We built everything from scratch, was the basically the biggest decision we did in the in the whole company because we have we had a lot of customers live with the old product, right? This was like more or less old voice chatbot world.

The 2022 Product Rebuild Decision

Nathan Latka

9:55How many were live and what year was that?

Malte Kosub

9:57So that was twenty twenty two. we probably at life I don't recall it, twenty, thirty customers. so And and we had we had this one off site where like we were discussing in the leadership, should we do it yes or no? a lot of people said, Hey, we can't do that. The competition will just like accelerate and we are not able to deliver new features because we're building this whole new thing, under the hood. No one knows that we're doing that. and w we we we approached like first principle. Do we believe this is the future and do we believe there will be competitors building it from scratch the right way? And we said yes, and if we believe that's the case, you

Nathan Latka

10:35Okay.

Malte Kosub

10:37you you just need to hit hit hit it and and and take the pain. So we we we took the pain, we didn't deliver a feature for more or less a year, but then we launched this product that was just like completely outperforming all the others. and and that was like the the the big what was big for us.

Nathan Latka

10:53Let me dive a little bit deeper here on a couple quick things. So founders can obviously take much bigger risks once they are personally set, you know, financially. Did that moment come for you when you sold the e commerce business? Did you guys make enough there where you were pretty much set?

Malte Kosub

11:06No, no.

Nathan Latka

11:06Okay.

Malte Kosub

11:07No, no. So that was there was high risk. so it was all in. again, but we believed this this technology actually will eventually overtake. But back then, like when we did the decision to generate an answer with a large language model, it took like thirty seconds. and it was way too expensive.

Nathan Latka

11:26Too long.

Malte Kosub

11:27So we bet okay, this this this we'll get to like like sub one second latency and eventually it turned out to to be right.

Nathan Latka

11:35So I appreciate the honesty. Most founders will come on the show and Yeah, I made so much money at the agency, you know. But you're saying, Okay, you know, we exited, but you know, I wasn't like sitting on twenty million dollars, right? So you launched the agency. what was the best year revenue wise for the agency before you shifted focus to Parloa?

Malte Kosub

11:51It was actually the year when we sold it. So we sold the agency in 2020. That was also the best year. So we we kept growing. But with the agency, we actually built this product in parallel. And then within the agency, we had a product team, we had a consulting team. We we separated from the beginning. And then we decided, okay, we need to focus on the product part. So we sold the agency to just focus on on the product. And That was in the the initiation of of actually the product company called Parloa.

Nathan Latka

12:21Can you share like the summary economics there? Like was the agency doing five million of revenue when you sold it?

Malte Kosub

12:27it was doing millions of revenue. I can't share the the details.

Nathan Latka

12:29Okay. Okay. But can I give you a big range between zero and ten million of revenue?

Malte Kosub

12:34I I can't I I'm not allowed to talk about it. But it was in our our field one of the leading or still no now it's part of a bigger agency, but it was one of the the leading agencies in Europe when it comes to conversational AI.

Nathan Latka

12:37Okay. Now, when you see how important it is to have like you new and proprietary data sets to create synthetic data, self playing agents, et cetera, do you regret selling the agency or was that the right move?

Malte Kosub

12:57I think back then it was the right move because I I I always say money loss lies on fo focus island. You need to focus, like if you have five things in parallel. so I think back then it was exactly the right call.

Nathan Latka

13:09Okay. Let's fast forward to Parloa. Chris, I see your question in the comments about L T V to CAC. We'll talk to Malte here shortly about how he's going to market, you know, in twenty twenty six and twenty twenty seven. But Malte, I I think you have talked about revenue publicly. Before we dive deep into the product and your vision for, you know, enhancing your foundation models and your fine tuning. what is revenue today?

Revenue Milestone and Total Funding Raised

Malte Kosub

13:29we are not openly talking about revenue, but we're we're adding more revenue than ever before. so yeah we're growing we're growing like crazy.

Nathan Latka

13:38Okay, so so I mean you tell me you could tell me this reporter might be wrong, but on on June second, twenty twenty six, CFO Tech put out a story that said you guys topped fifty million of revenue after series D. Is that inaccurate?

Malte Kosub

13:51that is accurate, but we actually reported fifty million in revenue already end of last year.

Nathan Latka

13:56We love that. So you're much bigger. You're much bigger than fifty million now. That's great. Okay. So what is your t to take us into the future? I mean, you're at the bleeding edge here, right? North of fifty million bucks of revenue. You're one of the rare unicorns in AI in Europe, I believe, right? Do you want to just summarize how much you've raised in the most recent valuation?

Malte Kosub

14:14we raised five hundred sixty million dollars. the last valuation was three billion.

Nathan Latka

14:19Three b amazing. So w w where's the company going from here?

Malte Kosub

14:23So we want to win this market, right? we believe The market that is going to emerge is one of the biggest software opportunities that has ever existed, right? We're starting with customer support and customer experience, but it will go to sales, it will go to marketing. I believe every single touch point of a company, if it's a homepage, if it's an app, will be driven by an AI agent. Right now you go to a homepage, you click through a menu, it's like very clunky. We believe in the future we'll talk to the homepage or chat to the homepage. So it will be hyperpersonalized, a unique experience for every single customer. And that will be driven by agentic platforms, and that's what we bet on.

Nathan Latka

15:01Mm-hmm. And if so, if we refresh your website like a year from today and I go under industries, right now I see insurance, retail and travel and hospitality, will we see you adding more industries there or will we see you going deeper, but only in those three industries to monopolize?

Malte Kosub

15:15No, th th there will actually be more industries coming in basically one or two weeks. we are already very heavy in in in other industries. That's just like we we launched new industry pages. Those were the f first three we we launched and they're always like let's ship them. The the the next are in in working actually. So we are already way further than than those three industries.

Nathan Latka

15:39Tim, Chris, I see you in the chat. Everyone else joining. If you've got questions for Malte, you can leave them in the chat. Again, he's built Parloa into one of the rare AI unicorns in Europe, north of 50 million bucks of revenue, serving large enterprise customers who believe that his vision is the accurate one, which is things are going to be driven by voice in the future and specifically voice agents. He's built the harness, he's fine-tuning his own models. Help us understand Malte, what sort of resources that takes. What's the total team size today? And how many are engineers?

Team Size: Engineering, Product and Go-to-Market

Malte Kosub

16:07So we are about above five hundred people. we have roughly two hundred people working in engineering and product. that's more or less the team size. Then we like ops and obviously big go to market in Europe, and in in the US, in Europe, in Germany, UK, France, Spa Spain, Italy, Denmark, and so on, and then obviously the US a pretty big go to market team.

AI Stack Depth and Infrastructure Approach

Nathan Latka

16:32And how deep have you vertically integrated in sort of the traditional AI stack? A lot of people say obviously the base layer is the actual raw land, the data center. Then you've got the chips, then you've got data and inference, then you've got the application layer eventually. How deep have you gone? Like are you renting GPUs from, you know, one of these companies at 250 an hour, their B200s to fine-tune? Or are you actually, you know, co-locating on a data center yourself? How deep have you gone?

Malte Kosub

16:54Yeah. So right now we're really much focused on the application layer because there's so much to build. obviously we are we are moving a bit lower, not just using frontier models but fine tuning open weight models. but overall in the application layer there's so much to build. When we talk about customer support is the first step, but we need to go to sales, to marketing, to every single interaction point. So we believe we're we're still scratching the surface when it comes to applications. yeah.

Nathan Latka

17:27So just to see if this is right, as you sort of leading this company and you have a lot of money, you're thinking about CapEx and investments, what I'm hearing you say is you're at the application layer. You have a unique data set, it sounds like already. Do you feel like that's a good moat for you guys or are you using data everyone else has access to?

Malte Kosub

17:42it's both. We have some proprietary data sets, but we actually believe that the the biggest value add is when when large enterprises actually give the context to the platform we are providing them. So they actually train their own brain, their company brain, right? So it's it's a bit of both. It's it's data for customer support purposes, to to make customer support actually better from a speech to tech a lot. speech to text, large language model and text to speech point of view, but then it really depends on the context of a specific company to add their data to to what we call the company brain.

Nathan Latka

18:23So just so the audience can follow along, guys, the visual here, Malte, correct me if I'm wrong, you've invested a lot in building a model which is effectively a harness. And when you sign up a new customer, you're allowing them to feed their data with your help into your harness to f which grounds the the the voice model, but and now enables them to build real character for when you know their their AI agent picks up the call, it sounds like you know, SAP or sounds like one of your customers. Would that be the right analogy?

Malte Kosub

18:46Yeah. Exactly. So w we built the application so they can put the context and then and the guardrails and and the whole life cycle in into the system. we're orchestrating models, some models we're fine tuning, and then there's our data and it's their data.

Nathan Latka

19:05Okay, okay, interesting. And can you talk to me a little bit when you are fine tuning, right? Together AI, Fireworks, Baseten, Modal, these are all very popular folks in terms of renting GPUs. Are you spending money on all these right now or where are you spending money to fine tune your models? The compute tower.

Malte Kosub

19:18Yeah. Yeah. we are we're we're working with some of them. it depends a bit on which customer, which geography. but yeah, y you you you showed some of the players we're working with.

Nathan Latka

19:31How do you decide though? Because the pricing each one, let's just focus on the Blackwell NVIDIA model. But let's say they want B100s to train their model, depending on who you go with, depending on how much service they layer on to the GPU hour, it can range from as cheap as I don't know, $5 per GPU hour all the way up to like $20 per GPU hour, which significantly I imagine impacts your margins, or maybe you pass that through to your customers. How do you think about that?

Malte Kosub

19:53So we are in a very s special situation because we are talking about voice right and latency is actually a very important ingredient in delivering high quality right in a chat you can have a latency of the model to to to respond like first time to token can be three, four, five seconds. That's fine. In a chat it doesn't matter as much. In voice it matters. We need to like have a a latency of below a second. So what we need to do, we need to s have Rented capacity for the maximum load of potential voice interactions to always guarantee a P99 latency below a second. So it's a pretty sophisticated way how we have capacity across regions and we are load balancing between those capacities to always have a P99 latency for a model. If we just use an API of a of a standard frontier model, you wouldn't get to that latency. And yeah, that that's important in voice.

Nathan Latka

20:57Thanks for nerding out with me a little bit there on the technical side. I mean, this is the future though, right? Like I want to know a CEO like you, how many billions do you have to raise before you start buying down and actually co-locating, buying your own order of B200s from from from NVIDIA, right? I mean, that's not in your horizon, it sounds like right now. Yeah, interesting. Okay, we have some good questions in the chat. Again, those of you just joining, your select CEOs I invited here to ask Malte questions. Make sure I'm asking him questions that you're curious about. Malte, one of the questions we got from Lawrence, he said, How'd you guys make your first million? What was the team size, right? When did you guys officially launched the business.

Early Revenue: The First Million and the 2020 Product Split

Malte Kosub

21:30That is a very good question. So the first million revenue was with just a couple of people back then.

Nathan Latka

21:39What year?

Malte Kosub

21:40sorry?

Nathan Latka

21:41What year was that?

Malte Kosub

21:43That's a very good question. I actually can't answer that.

Nathan Latka

21:45Yeah.

Malte Kosub

21:47it was in the consulting time. So we need to split between consulting a million revenue. That was obviously earlier, product a million, but I I would need to check when that actually was.

Nathan Latka

21:58Lawrence, if you join late, Malte, just just to catch them up, like guys, he had a great agency that he sold off.

Malte Kosub

21:58but back then were like not a lot of people.

Nathan Latka

22:04He had a product team that he then spun out, which, Malte, I think that's what started Parloa officially in September of twenty eighteen, right?

Malte Kosub

22:08Exactly. Yeah. Yeah. Exactly.

Nathan Latka

22:11But do you remember when you so are you saying when you spun out the product team from that combined business that that product was already doing more than a million a year in revenue?

Malte Kosub

22:18no, back then not. So I think we we were at like twenty, thirty thousand MRR, so maybe two, three, four hundred thousand ARR. but it was really on the side, right? That wasn't like let's build this product slowly, let's see if it works. There was no very focused go-to-market motion for the product. that's why we also changed the focus. Say let's let's sell the the consulting business, let's focus on the product. and then twenty twenty was the year where we split both to to to product and and consulting.

Revenue Milestones and Growth Timeline

Nathan Latka

22:49Twenty twenty was the official split. And then do you remember the year you passed ten million of revenue?

Malte Kosub

22:58I would need to look that up. so we the first years it was not not very steep because we were like bootstrapping and then in 2022 we raised our first funding round. And from then on it actually started to to to actually increase significantly. From then on we we went pretty pretty fast.

Nathan Latka

23:17Maybe the better question, I'm trying to get just one revenue data point between where you're at today, which is north of fifty million and where you started at four hundred K of ARR. What is the revenue point in the middle of that somewhere that you remember the most?

Malte Kosub

23:29It's a good question. I would need to look up those data points. Ten million must have been somewhere in twenty twenty four, I guess.

Nathan Latka

23:45Okay. And the reason I bring it up, I mean, you took some crazy bets in 2022.

Malte Kosub

23:46Yeah.

Nathan Latka

23:48You said you had 30 customers live, you threw out and rebuilt the model. Was that new model and the power you're able to demonstrate what enabled you to raise that first funding round?

Malte Kosub

23:59it was actually the second funding round. it was the Series A that helped us with with that decision, right? So we we made that change. The the first customers were also not like comparably as big as they are right now, right? So we increased ACV like over the last four years like crazy. but obviously it was the foundation to actually be where we we wouldn't have made that decision, I'm pretty sure we we would have been where we are right now.

Growth Ambitions for 2026

Nathan Latka

24:29All right. Malte, as we wrap up here, folks have obviously heard your vision. how aggressive are you being in terms of growth? Do you think you can more than double this year in terms of revenue?

Malte Kosub

24:38100%. That's the clear goal. We need to be aggressive. We're investing left, right, and center. We're growing our sales teams. And we have the product. So I I always say it's now just about execution, every single team. I say we we we need to perform world-class in every team. And that means everyone needs to need to actually fight and think, okay, can I outperform my competitors? so yeah, it's it's all all in.

How Parloa Wins Competitive Deals

Nathan Latka

25:09Brian said, Parloa is entering a super crowded space. Quote, everyone does what they do, what actually differentiates them. how are you when you're on a sales call and people are going head to head, right? There's comparing you to somebody else, why are you winning these deals?

Malte Kosub

25:22Yeah. It ha heavily depends on the customer and To whom you are competing, but give I can give you like two, three data points. First, everyone can show you a beautiful demo, but to scale to hundred countries, three different channels, three different versions of an agent, you're getting to 900 versions of an agent. You need to build a very enterprise scalable platform to do that, to more scalable globally. Second one, we started with voice. A lot of our competitors started with chat, right? so actually enaling voice is very, very hard. So we build a lot of low latency left, right, and center, our multi-agent framework is low latency, guardrails are low latency. Third is the whole optimization and observability. We build the most powerful observability platform to actually oversee how agents are performing. and and and the next one is We actually built Navigator. Navigator is the way how you build your agents with AI, how you train and fine-tune your agents with AI. And I always say, hey, test our product for a couple of days, test the competitor, then you can make a decision on your own. So I'm a huge believer in let the product talk, not the slides. And so far we have been very successful with that.

Nathan Latka

26:37Malte, if your European friends at ElevenLabs call you up and say, Malte, come on, we just let's do a deal here and go take on the Americans. Let's do a good merger. We'll do an all stock. We'll value you guys at five billion dollars. Do you do the deal?

Malte Kosub

26:49I think I I love Marty and we're we're partnering and I think we are we are both both approaching that market and we have a lot of pride coming from Europe and I think we're doing a good way as it is right now.

Nathan Latka

27:00All right, fair enough. All right, as we wrap up here, if people want to follow your story as you continue to grow, where can they find you online?

Malte Kosub

27:06LinkedIn on X follow me write me I'm always happy to engage I'm left right and center excited about what is happening in industry and yeah if you if you want to want to engage always feel free to feel free to reach out.

Nathan Latka

27:23Guys, Malte has built one of the rare AI-focused unicorns in Europe, built it out of his agency, which he sold in 2020 for quote millions of dollars, spun out the product set though, which was doing about 400,000 bucks of revenue in 2018. The product suite's very different today. They broke 30 customers on that product, which is now Parloa in 2022. But he said at the time the space was moving so fast they threw out and rebuilt the whole model. He went out and then closed the first funding round, broke. 10 million of revenue, somewhere around 2024. Now today they've got customers that love them so much for their voice interactions. They're paying more than a million per year. The ACV average is north of 100K. They broke 50 million of ARR in December of 2025. And this goal is to more than double this year with this team of 500 people, 200 focused on engineering, which have, you know, they've built effectively a voice model harness that enabled new customers to ground their model, train their voice model, give their voice agents. character for their business, which a lot of people are arguing that's what IP is today. He said no matter how many billions we have, we're not going too far down the stack in terms of buying our own chips or data centers as they continue to scale into new industries. Malte, thank you for taking us to the top.

Malte Kosub

28:26Thanks, Nathan. Was a lot of fun.