The AI Gold Rush: Should Companies Build or Buy AI Solutions?

AI is no longer a futuristic concept; it’s here, and companies are racing to integrate it. But with rapid advancements, many businesses are facing a tough choice: Should they build AI solutions in-house or buy them from external providers?

In this 2-episode podcast of the MVPF InnoSanity Podcast, host Baby Jessi Parker speaks with Jasper Wilmes, Venture Architect at MVP Factory, about the AI boom, the risks of adopting off-the-shelf AI, and how companies can strategically integrate AI without losing control over their data. If you’re trying to figure out how to unlock AI’s value while staying competitive, this episode is a must-listen.

Here's what they talked about:

Baby Jessi Parker: Welcome to InnoSanity, the podcast where we dive deep into the accelerating world of corporate innovation and venture building.

Today's episode is the first in our two-part series on monetizing data with Generative AI. We’re exploring one of the most exciting and transformative topics in tech—how AI, particularly Generative AI, is reshaping industries, fueling new business models, and presenting both challenges and opportunities for startups and enterprises alike.

Joining me today is Jasper Wilmes who will talk about the AI boom and venture building and the opportunities it brings, and how somebody, maybe yourself, a colleague, a friend, or even a family member, can leverage it. We’ll talk about some of the tools and concepts that can help achieve strategic business goals.

But I’ll let the expert talk about that.

Baby Jessi Parker: Jasper, welcome to the show. Before we dive into the topic, let’s have a short introduction about yourself. Who you are, what you do, and, I don’t know, maybe a hobby if you wanna throw something in there. But tell the people who you are.

Jasper Wilmes: Hi everyone. Jasper here. So I'm a venture builder at MVP Factory. Essentially, what I do is build ventures from scratch to investment and then hand them over to external founders. My role is to take a company’s strategic investment scope, turn that into a specific idea, validate whether that idea makes business sense, and then help founders secure their first round of investment. Once that’s done, I move on to the next venture.

Right now, I live as a bit of a nomad. I’m currently in Portugal, but I spend most of my time between Portugal, the Netherlands, and Germany, depending on the season and the weather.

Baby Jessi Parker: You’re getting all the sun right now. There’s none here in Germany at the moment. There was some yesterday. But anyways - so, yeah, AI. It’s nothing new, but at the same time, it’s new because you see it everywhere. Maybe it’s lost a bit of that “newness.” AI has been around for a long time. But then, in November 2022, ChatGPT hits the market and takes the world by storm.

With it came a whole revolution of things. So the question is, what is your take on the AI hype that's currently going on, specifically in the startup world?

Baby Jessi Parker: So, let’s jump into the topic. AI is everywhere - it’s new, but it’s also not new. AI has existed for a long time, but then in November 2022, ChatGPT launched, and everything changed. Suddenly, AI wasn’t just for tech companies, it was everywhere. What’s your take on the AI hype that’s happening right now, especially in the startup world?

Jasper Wilmes: It’s fascinating. Like you said, AI isn’t new, and neither is Generative AI. The first chatbot existed in the 1960s. But the key difference is accessibility.

Before ChatGPT, AI models existed, but they weren’t accessible to the public. Only big corporations could afford them. I remember giving a workshop on AI at EY a few years ago, and most people weren’t interested, it felt too far off.

Then, ChatGPT launched, and for the first time, anyone could use AI for free. That was the game-changer. Suddenly, AI wasn’t just a theoretical concept; it was a tool people could interact with. And because of that, the demand for AI skyrocketed overnight.

Baby Jessi Parker: That’s an important shift. And now we’re seeing companies scrambling to adopt AI. But with that comes a challenge - how do they do it in a way that makes sense for their business?

Jasper Wilmes: Exactly. This is where we see the buy vs. build dilemma.

Baby Jessi Parker: Let’s break this down. Companies who want to integrate AI have two options: buy an AI solution from a startup or build it in-house.

Jasper Wilmes: Yes, and both options have their pros and cons.

If you buy AI from a startup, it’s faster and doesn’t require hiring AI engineers, which are expensive and scarce. Startups also move faster and can iterate on technology quicker than corporates can.

But there’s a huge risk: when corporates buy AI solutions from startups, they’re essentially feeding data into the startup’s models. Over time, the startup learns from the corporate’s data and improves its AI.

What happens next? The startup can take that AI and sell an even better version to your competitors. That’s what I call the Trojan horse problem.

Baby Jessi Parker: Wow, so companies could end up training a startup’s AI, which later gets used against them?

Jasper Wilmes: Exactly. That’s why some companies choose to build AI in-house. But that also comes with challenges - AI talent is expensive, development is slow, and by the time an internal AI tool is ready, the technology may already be outdated.

Baby Jessi Parker: So, Klarna did it, and they did it very well. I don’t think this was their first try. I think they’ve tried things beforehand, but this is definitely a successful one.

You know, kind of shifting it then towards a company that is starting from zero, how could they tap into AI and leverage it for their business? Where do they start?

Jasper Wilmes: Yeah, so there's this very classical dilemma that corporates face when adopting AI solutions. But before we get into that, let me give you some actual data on how Klarna’s AI assistant has performed.

So Klarna rolled out their AI-powered customer service assistant in January 2024. After just one month, they had 2.3 million conversations, which accounted for two-thirds of all customer service chats. It performed the equivalent work of 700 full-time agents. The satisfaction rate was as high as if a human handled the conversation.

And here’s the most impressive part: the AI assistant was able to handle customer service requests in two minutes, compared to 11 minutes when handled by a human agent.

They estimate that, in just that single month, it improved Klarna’s profits by $40 million.

Baby Jessi Parker: That’s crazy. $40 million just from optimizing customer service efficiency?

Jasper Wilmes: Exactly. And this is just one example of how companies can leverage AI to increase efficiency. But that’s only part of the story. The real opportunity lies in how AI can be used to monetize corporate data and create entirely new revenue streams.

Baby Jessi Parker: Jasper, you mentioned offline that this is where a lot of companies are looking for answers — and it’s easy to see why. It’s everywhere. But how do they go from intention to action? Where do they start?

Jasper Wilmes: It’s a good one. And as many in innovation or venture building already know, you always start with a problem.

The reason is simple — if you start with a solution, the odds of building something no one wants is close to 100%. Most innovation frameworks start with empathy for that exact reason: who has the problem, how big is it, and what’s the current experience?

And here’s where GenAI throws a wrench into things — because it’s a solution. If you say “We want to use GenAI,” you’re starting with the solution. That’s where things fall apart.

So the method I follow is to first deeply understand the solution — its capabilities. That means understanding what models are available and what tasks they can perform better than a human.

You also need to know what the model requires in order to work. For instance, ChatGPT is good at writing essays but poor at basic math. It’s not general intelligence — it's narrow AI.

Baby Jessi Parker: Yeah, and I think that’s one thing people overlook — training the model. It’s a huge lift.

Jasper Wilmes: Exactly. And every model has different requirements. You need to understand what it can do, and what data it needs to do that.

Once you have a shortlist of models and tasks they can perform, then you bridge back to the problem. You look across your company — or even the market — and ask, where do these tasks show up? Can they be automated? Do we have the data to support that automation?

That’s where corporates have an edge. They sit on mountains of data. Maybe it's messy or unstructured, but it’s there — and with some work, it can be cleaned and used.

Baby Jessi Parker: Right. So once you’ve identified a process that fits a model, and you know you’ve got the data — what next?

Jasper Wilmes: Then you validate it the same way you would any other venture: desirability, feasibility, and viability.

Desirability — is it something the user or customer wants?
Feasibility — can we actually build it with the current tech and data?
Viability — does it make commercial sense?

If you get green lights on all three, you start building. But before jumping into an MVP, you usually begin with a proof of concept to test if the model can deliver accurate outputs.

Baby Jessi Parker: So it’s still the classic build-measure-learn loop, right?

Jasper Wilmes: Absolutely. And to your point, you want to start small. Go for use cases that are low-risk, repeatable, and highly standardized. Things like answering “What’s my balance?” in customer service are perfect.

You don’t want your GenAI giving safety instructions for repairing appliances right out of the gate. Start where a mistake won’t be catastrophic.

Baby Jessi Parker: Makes sense. So build the first small process, test, iterate, and expand. But how do you stay ahead of the curve once you’ve built something?

Jasper Wilmes: That part is the same as for any venture. Keep learning from users, adapt based on feedback, and improve the product with every iteration. Stay close to the problem. Stay close to your users.

What’s different with AI is the speed of advancement. It’s exponential. So the faster your feedback loops, the better your odds of staying relevant.

Baby Jessi Parker: And now to one of my final questions — what are the biggest pitfalls companies should avoid when building GenAI ventures?

Jasper Wilmes: First, don’t fall in love with the solution. Fall in love with the problem. Start there and stay there.

Second, understand that data is your only true source of competitive advantage. The models are democratized — everyone has access to them. What sets you apart is the data you can feed into those models.

And third, move fast. Work like a startup. The biggest blocker for corporates is speed. You need to build quickly, test constantly, and make decisions based on evidence.

Baby Jessi Parker: So basically — act like a curious kid. Play, learn, test, and don’t be afraid to fall down a few times.

Jasper Wilmes: Exactly. Be a bit more childish. It’s the best way to learn.

Baby Jessi Parker: One thing we didn’t cover — your whitepaper! It’ll be live by the time this episode airs. Want to give people a quick teaser?

Jasper Wilmes: Definitely. The whitepaper is called Monetizing Data with GenAI: Building Scalable AI Ventures for Industry Leadership. It goes deeper into everything we’ve discussed today — how to identify AI opportunities, evaluate feasibility, and monetize corporate data through scalable ventures.

Baby Jessi Parker: Amazing. Also, if you haven’t heard Part 1, make sure to go back and listen. Jasper — thank you for your time. Always a pleasure.

Jasper Wilmes: Thanks a lot, Baby Jessi. Always great to be here.

‍

Interested in a deeper dive into AI Venture Building? Have a look at our whitepaper - get your free copy here.