At the beginning of March, Microsoft held a three-day hackathon* on the topic of AI & Analytics. Together with Helsana, ipt was at the start with its own team.
As consultants on site at the client's premises, we regularly exchange information on innovative topics and often drive these forward together with the client. At Helsana, we had already been working for some time on the question of how to offer policyholders the products that best suit their individual needs. This led us to the topic of "recommendation systems".
I heard about this hackathon through a colleague and saw it as an opportunity to move this topic forward. I simply encouraged participation while having coffee with the customer. After the positive feedback we finally started the hackathon with the idea to develop a recommendation system for insurance products.
The development steps on the way to the prototype included the preparation of the data, the training of the model and the evaluation of the results. At the end of the hackathon we were able to present a working prototype in the form of a machine learning model. The model is already able to select the best fitting products from a given number of products and to reproduce them as suggestions.
The next step is now to use the model to generate real added value. To do this, we must further optimise the model, test it again in detail and finally integrate it into Helsana's IT infrastructure. These steps will now be carried out as part of a continuing collaboration with Helsana.
In addition, we have been able to achieve a significant gain in knowledge. While at the beginning of the hackathon it was still relatively unclear how to approach the topic, we now know much better than before what works and what doesn't. We have achieved this by trying and testing different things. Other Artificial Intelligence (AI) projects at Helsana will also benefit from these findings.
A hackathon is not a self-runner and presents the teams with various challenges. The following tips helped us to master them:
Particularly suitable are questions or problems that leave room for creative solutions and whose complexity is at the same time manageable. For analytical topics, the following model can be used in the search:
The starting point for the AI and Analytics Hackathon was the problem that insurance customers sometimes choose products that are not optimally suited for them.
A hackathon has four major advantages from the participants' point of view:
A hackathon is an event in which several teams work under time pressure on creative ideas and solutions. The following characteristics are typical:
The hackathon at Microsoft lasted three days and focused specifically on AI & Analytics topics. In our team of ipt and Helsana employees we brought together both technical and professional expertise. In addition to us, three other teams took part. Each team was also supported by a Microsoft architect.