Minimal AI was founded a year and a half ago by Titus Ex and Niek Hogenboom. Recently, $3,6 million in growth capital was raised. Ziptone spoke with Niek Hogenboom about the company's AI solution and Minimal AI's growth plans.

 

Minimal AI focuses on the automated development of flexible AI agents for customer service in e-commerce. The solution eliminates the need for an in-house engineering team, the company says.

Niek Hogenboom (pictured, Aerospace Engineering, TU Delft) is a technologist by training, but above all an entrepreneur. During the corona pandemic, he ran a platform company that allowed businesses to arrange – and cancel – home workspaces for their employees. Strictly speaking, that was a managed service, largely automated, but as Hogenboom says: “We could automate everything except customer service. When the major language models emerged a few years ago, I thought: okay, now that issue is solved too.”

His business partner Titus Ex (Computational Science and Engineering, TUDelft) was less certain. “Titus has years of experience with machine learning and neural networks. He saw it differently: this will be a long road, which actually offers a lot of opportunities to get in with a start-up.”

Speed ​​as an advantage

Only for about a year and a half have Hogenboom and Ex engaged with Minimal AI A mixed picture among existing AI startups for customer service. The underlying idea: large providers of customer service solutions cannot develop as quickly as a startup.

At the same time, Hogenboom knows that scaling up is the biggest challenge. “Nowadays, you can develop a chatbot very quickly that provides reasonably good answers to customer questions based on a website. But only when you put it into production do you find out that the reality is much more difficult.”

Minimal AI focuses explicitly on e-commerce players. This makes it possible to thoroughly develop specific solutions. Consider, for example, the effective application of customer knowledge at the right moment in subsequent interactions with the customer.

The AI ​​Manager who develops agents

Hogenboom promises “high-performing agents you can trust.” Are we talking about agentic AI (autonomous bots that can perform tasks autonomously) or generative AI agents? The distinction proves to be blurred with Minimal AI: “Our agents can implement changes such as updating addresses, canceling an order, or modifying a subscription.” The bots do not dig through the databases themselves, but initiate 'hard-coded workflows,' or processes driven by deterministic software.

But on the other hand, there is the AI ​​Manager, who develops agents largely independently. The AI ​​Manager takes the lead by performing analyses on all contacts and deploying the most suitable tools itself. In doing so, the AI ​​Manager gathers further information by asking targeted questions to, for example, a contact center manager. The manager can simply answer via voice.

That approach ensures the rapid development of agents: an e-commerce client with 40 million euros in revenue reached 85 percent automated processing of all emails after two days of collaboration, according to Hogenboom. “If you have talked to our AI Manager for a few hours, at some point, it basically knows everything.”

Learn to work with API endpoints yourself

The AI ​​Manager also reports problems in data management – ​​the bottleneck at most companies – and offers suggestions for improvements. The solution also refers to best practices from other companies. Furthermore, the solution can determine for itself, without the intervention of developers, how best to handle API endpoints – the place where information can be retrieved for a specific task. Hard-coded limits are built in for this, such as being able to modify something only once per run, Hogenboom explains.

The AI ​​agents learn how to handle their tasks via the AI ​​Manager, including guardrails and exceptions. “When an agent needs to modify a customer's address and encounters multiple outstanding orders or multiple addresses, the agent can make the right choice and implement it in consultation with the customer,” says Hogenboom.

Currently, Minimal AI primarily works with various models from Anthropic, but it is continuously monitored which processes are best executed by which model.

What to do with 3,6 million?

To continue growing, Minimal AI has raised $3,6 million in venture capital. Among the investors are fifteen of the company's customers. Hogenboom declined to disclose their identities. Minimal AI will use the raised capital to further improve the product. “For example, we aim to enable our AI Manager to independently add new FAQs based on its own analyses. It really needs to become a central system that implements everything stemming from your service process throughout the rest of the company.”

Minimal AI currently has six people on board and a turnover of over one million euros. The goal is not so much to grow in personnel revenue. “Headcount is no longer very important. We would rather have one new employee managing AI agents than ten employees doing manual work like we did two years ago.” Hogenboom would prefer to see turnover tenfold next year: “That wouldn’t be very strange. It is very easy for customers to get started with this and be successful. We see a wow effect with our customers when they start using our solution.”

(Ziptone/Erik Bouwer)

Featured, Technology