Talkwalker

Christophe Folschette and I started with the idea of analysing social media. After two and a half years, we joined with Robert Glaesener, who is now the CEO of Talkwalker, and we decided to work together. That's when we launched the Talkwalker platform, which is our only product. It's a social media monitoring platform to analyse consumer insights. Companies can enter in their brands and their topics or whatever they want to monitor, and then they will see what people write about them in the public internet. So you can analyse what the audience is saying on Twitter, on Facebook, and also on the traditional media like blogs, news, and message boards. The goal is to have a 360 degree view about everything which is being written online. So you can use it just for brand monitoring, but it can also be used for research to see what is being written about your brand in different markets. Opinions can differ completely from one market to the next. Or you might see, for instance, that your brand is actually often mentioned in a different and unexpected context in another market. Then you can adapt your advertisements to better fit that context, for example.

There are always lots of challenges in IT. Recently it has been all the regulations which have been changing, like the GDPR. Then it's processing and analysing all the data. We index over 300 million posts a day and there's more than 3 or 4 billion updates to these articles, like when the share count changes, etc. And everything needs to work smoothly 24 hours a day. Another challenge is to develop the company so that we are prepared for the future and so we can remain innovative. So from the start when we created the company, we have been customers at Hetzner. We started first with 8 servers, and now we have over 4,000 servers in total, which is quite huge. And Hetzner has always been a partner who supported us with our needs so we could really grow the company.

Before starting Talkwalker, I ran a small personal project, a Luxembourgish search engine. I was already using Hetzner at that time. Afterwards when we started the company, we were looking for servers. And it was natural to choose Hetzner because I knew them from before, and they were always a very reliable partner. And we stayed with Hetzner and we’re pretty happy. The main reason for that is that they don’t make any promises they cannot keep and they are very reliable. If we discuss something, they really make it happen without any long discussions and negotiations. And also they listen to us if we need something like the GPU servers. They brought them into production and made them available for us. Then the support, as well. There's the 24 hour support in the data centers. If something happens, there was always somebody there, and the issue was resolved very quickly, even at night. So they listen to our feedback and help us to grow the company. With Hetzner, we are able to focus on our company.

The first one is for training. On the machine learning side, we do, for instance, automated text translation, sentiment analysis (which today actually works in over 90 languages), and image recognition. If your brand is in an image or even a video, we can actually detect that the brand is being mentioned there. So that actually takes a lot of resources to process, and we need the GPU machines for that. So we use them both for training and preparing our models, also because creating those AI models takes a lot of time to compute. And afterwards, the second reason is for the evaluation/inference of those models in order to annotate the data.

It's very important for us to automate things since we have so many servers. We want to be able to monitor them automatically, to handle incidents automatically and also not to be awoken by small hardware issues which might occur. We want to make sure that services run all the time and that there is no interruption. So Hetzner provides us with an interface where we can automatically trigger, for example, hardware tests or the replacement of machines. This helps us enormously in our efficiency.

For Talkwalker, AI/machine learning has always been a very important topic. We already had a team of dedicated AI specialists which was created four years ago to really develop Talkwalker in that direction. We have many things on the AI side, like sentiment analysis, which are helping us and giving us an advantage compared to our competitors. So it's very important to use these technologies. For the future, we have many things lined up there, and it is also something we have already heavily invested in, and there's more coming. We will certainly invest even more into machine learning (and thus have more servers with GPUs) because of all the data we actually work on and because of the even greater amount of data that we process and transform. So the main mission for us is to always find the needle in the hay stack – to find the important information hidden around all the other millions of pieces of information – to help the customer to reduce the time which they spend on the platform and to give them valuable insights. And I think AI is the main tool in helping us to really achieve that goal. So I’m really looking forward to what will come and how we’ll continue to develop.