Kairos' AI lab: a new approach for extracting value from data through artificial intelligence
In recent year, we have received many requests from companies who want to understand how they can benefit from AI in their business. Some of them have set up a costly IT infrastructure to work with AI and large-scale data analysis, but have not found applications and working methods which yield a return on investment. Others are unsure how to get started with working with AI, and what value there is.
In order to accelerate value creation with the help of AI, Kairos Future, through dialogue with a variety of actors, has developed a lab-based approach to cost effectively and time efficiently explore AI applications. The process is described, step by step, further down.
The First Round of Projects in the AI-lab
Let us first take a look at the first four companies to have used the AI Lab and their projects. Each has a unique issue and application:
1. Together with one of the world’s biggest dental clinics, we trained a deep learning model based on artificial neuronal networks to automatically identify caries in dental x-rays. The first step is about being able to focus resources on the right patients, by automatically filtering out the majority of patients who have no dental problems. In order to quickly get to work, our AI team collaborated with AI researchers at a Nordic university.
2. AI Lab participant number two is a global pharmaceutical company which recognizes the hidden value in the vast amount of information about health and medication that people share online. Through the AI Lab’s iterative approach, algorithms in natural language processing were trained to structure and classify content in social media, thus creating a searchable database. At the aggregate level, a clear picture emerged of the problems patients experienced with a given treatment; how patients were experimenting with modified medication; and medical products fit into the subjective realities of patients. The methods are now being integrated into the company’s product development processes.
3. A European tourist organization came to the AI Lab with the following question: how do tourists travel between destinations in Europe? By applying so-called unsupervised machine learning to a set of data sources with individual level information – train trips, package tours and online reviews – 20 cross-border routes were discovered in Europe. Each route could be linked to travelers’ profiles and experiences. The information will be used to create new offers for multi-destination travel.
4. A leading hygiene product company needed an AI-based methodology to explore the future for a product launch in China in 2019. The question was about which health-related concepts could be used for differentiation in 1.5 years. In the AI Lab, algorithms were developed to first discover around 80 promising concepts, and then identify the handful with the highest potential. In the experimental AI lab, Chinese e-commerce data was used to map how concepts spread between product categories and consumer tribes. The outcome was a clear picture of the concepts that can be used to evoke a premium feeling in the new product.
The AI lab’s four steps
The process in the AI-lab follows four steps:
1. Identification of use cases and data sources. Use cases are identified through a structured process where we look at local and global factors: the company’s bottlenecks and challenges, existing applications which already exist that can be transferred from other areas, and which new applications are possible based on today’s AI algorithms. In some cases, there is untapped potential in data that a company already possesses. In other cases, it is necessary to turn to external data streams or set up new ways of collecting data.
2. Exploration and experimentation of data. When questions and data sources are in place, the data exploration and experimentation starts. Initially, our team examines the data from different angles, increases the signal to noise, and lets patterns in the data emerge. After a number of iterations, interesting connections and tendencies tend to emerge from the analysis. These become a kind of map for potential value-added opportunities that exist in the data. The process is made possible by both market-leading AI solutions from suppliers like Google and Amazon, open source state-of-the-art research results, and Kairos Future’s proprietary AI platform, designed to facilitate the experimentation.
3. Training and automation. In order to take advantage of the possibilities in the data continuously, AI models need to be trained to automate the classification and analysis of new incoming data. These models are designed to improve their performance over time.
4. Evaluation, testing and scaling. When a model is trained, the solution is evaluated independently and tested on a small scale in the business. If the lives up to expectations, it is scaled up and implemented in the business as a whole. It is only at this point, when the value of AI is assured, that investments in IT infrastructure is needed.
If you are interested in participating in the next round of the AI lab, which will start in April / May, contact Mats Lindgren. You do not need to have a clear idea of what you want to do – we are here to guide you.