One Year of ReplyFabric and What the First Year Taught Me

A year ago, ReplyFabric officially became a company.

It feels much longer.

Not because the year went slowly. Quite the opposite. It has been a roller coaster of building, throwing things away, rebuilding, talking to customers, changing assumptions, getting audited, exhibiting internationally and continuously discovering that something I thought was finished was actually only the beginning.

The interesting part is that our core idea hasn't really changed. But the execution has changed completely.

The first version ended up in the bin

We had an early MVP. A proof of concept. It worked. And then we threw it away. That wasn't an easy decision. As a founder, you naturally want to move forward. Starting again feels like moving backwards. But the more I thought about the architecture and what ReplyFabric could become, the more convinced I became that the original thinking wasn't good enough.

So we started again.

Today, I'm extremely happy we did. The fundamental thinking behind that second version is still the foundation of ReplyFabric today. It gave us something much more important than a working demo. It gave us an architecture that could evolve. And our customers made sure it had to evolve.

Customers have changed our product

You can have a great idea. You can understand the technology. You can study a market and talk to experts.

But real customers have real processes. And real processes are messy. They have exceptions. Priorities. Different teams. Different working hours. Customer-specific rules. Attachments. Databases. Historical conversations. Internal responsibilities. Information hidden somewhere in an ERP or CRM.

That's where things become interesting.

A year ago, many of the capabilities we now consider fundamental didn't exist. Today, ReplyFabric can perform advanced categorization using far more than the words in an email. We extract features, understand business context, apply rules, look up information in databases and use that information to determine what should happen next.

Reply generation evolved in the same way. A while ago, everyone talked about prompt engineering. We increasingly think in terms of context engineering. A good reply isn't created because you wrote a clever prompt. It is created because the AI has the right context. That can include the complete email thread, mailbox memory, company knowledge through RAG, information about the sender and data retrieved from other business systems.

Then there is workload management. Understanding an email is one thing. Understanding who should handle it is another. Teams have different responsibilities, workloads and availability. Assignment has to take those realities into account and keep track of the workload over time.

None of this appeared on a whiteboard on day one. It emerged because we listened.

Really listening is harder than it sounds

"Customer first" is one of those phrases almost every company uses. But there is a difference between collecting customer feedback and really listening.

Sometimes a customer asks for something that exposes a much bigger problem. Sometimes they describe their current way of working and you realize there is a better way to solve it. Sometimes they ask a question and you think, "That's a very good point. Why don't we handle that?"

Those conversations have shaped ReplyFabric.

Our customers aren't designing the product for us. That's still our responsibility. But they constantly confront our technology with reality. And reality is an excellent product manager.

Becoming a serious company early

Another decision I'm very happy we made was to structure the company properly from the beginning. Policies. Procedures. Security controls. Responsibilities. Documentation. And then we asked external auditors to challenge it.

Going through ISO 27001 and SOC 2 Type 2 wasn't easy. It required an enormous amount of work for a company our size. But it was worth it.

An auditor's job isn't to tell you how wonderful everything is. They ask questions. Sometimes uncomfortable ones. Even when you believe you have followed best practices, they can notice something you overlooked. Then you investigate, adjust a policy, improve a procedure or make a control more precise.

That process made us better.

And becoming ISO 27001 certified and completing SOC 2 Type 2 in our first year says something about how we want to build ReplyFabric. AI can move extremely fast. Trust can't be an afterthought.

Getting out of the building

Then there was another kind of learning. We had to show ReplyFabric to the world. Viva Technology in Paris was our first real public appearance. Then came GITEX Europe in Berlin. And now we're heading to LEAP in Saudi Arabia.

For a startup, these events are more than booths and badges. You explain what you do hundreds of times. People ask questions you didn't expect. Some immediately understand the problem. Others don't. And every conversation teaches you something about your own company.

We exist for a reason. But having a reason to exist and making the world understand that reason are two different challenges. We're still working on the second one.

The founder train

Building a startup sometimes feels like driving a train while you're still laying the tracks. You want to go faster. Then faster again. There is always another feature, customer conversation, event, integration, idea or opportunity. When something starts working, my natural reaction isn't to slow down. It's to push harder. But during this first year, I've also learned that speed isn't the only thing that matters.

Sometimes you have to listen to the people around you.

Sometimes you have to listen to your own body.

Sometimes you have to stop for a moment and ask whether you're still driving in the right direction.

That doesn't mean losing ambition. It means making sure you can keep going.

From emails to business processes

When I look at ReplyFabric today, I see something much more mature than what we started with one year ago. We don't really see email as text anymore. We see an operational email as the start of a business process.

An incoming email contains information. That information can be understood, extracted and transformed into structured data. It can trigger a workflow, retrieve information, assign work, prepare a response and create knowledge that the organization can use again.

That's why I believe the opportunity is much bigger than simply helping people write emails faster. If a business process starts with an email in a shared mailbox, we want to understand that process.

Tell us how you work. Then we'll figure out how to automate it.

After one year, it is still about people

There is a lot of AI inside ReplyFabric. There are models, embeddings, feature extraction, RAG, memory, classification, routing, context engineering and integrations. But the most important lesson from our first year has surprisingly little to do with AI.

It's about people.

Listen to the people using your product. Understand their problem before deciding on the solution. Build what genuinely makes their working day better. And keep improving it.

One year in, that's probably the clearest principle I have for ReplyFabric.

I want our customers to be happy.

It sounds almost too simple. But I think that's exactly how I want us to build this company for the years ahead.