Using Artificial Intelligence to your advantage
New technology should improve existing processes, yield better results, and save time and costs. Artificial intelligence has the potential to tick all boxes when it comes to analysing large amounts of data and the ability to draw the right conclusions based on analytics. Identifying investors whose investment criteria are aligned with your company’s investment story and segmenting target groups to personalise investor engagement, in short, investor targeting, is a prime example of a problem where machine learning and artificial intelligence will play to your advantage. Huge amounts of company and investor data are readily available, while at the same time filtering, analysing, and drawing conclusions from the data manually is a time-consuming exercise and takes substantial domain knowledge to avoid mistakes.
This article undertakes to address the common concern that the machine returns black box suggestions in a world which still is a people’s business. It aims to provide an understanding of the principles of the algorithms and their advantages and shortcomings to conclude how AI can be deployed to your benefit.
Where traditional approaches fall short
Taking a traditional approach, you will devise a list of comparable listed companies based on various features to see if there are investors who are invested in various peers but not in your company. There are three obvious challenges to that exercise. First, it takes a very detailed analysis of the relevant investment criteria your company offers. Second, it requires excellent knowledge of the investment universe to identify the right list of peers. Third, the result will contain an excessive number of lines entailing a significant effort to rank the fit. Leaving this problem to the sell-side means all the above but adds a bias to the ranking based on the inherent conflict of interest the sell-side faces when servicing both you and their investor client.
How recommender systems approach the task
You will know recommender systems as a consumer where the machine makes suggestions of which other products or services you may be interested in and most likely your marketing/sales teams are applying them as well to enlarge the customer base. Target investor identification is a similar problem.
There are two basic types of recommender systems which can be used as a starting point.
Content-based filtering
Content-based filtering uses investment characteristics to identify potential targets based on their existing holdings. This is like the first step you take when using a merged shareholder analysis where you define your group of peer companies. To reduce complexity, you will manually limit the number of investment characteristics whereas the algorithm potentially has the whole investment and investor universe to choose from.
Advantages of content-based filtering are:
- The model looks at each investment portfolio individually and is easy to scale
- The model can capture specific features of an investment portfolio, and can recommend investors based on niche features
Disadvantages of content-based filtering are:
- To train the machine one will have to supply investment characteristics to the model. This requires a lot of domain knowledge since the model will only be as good as the investment characteristics it was provided with.
- The model is constraint to existing investments, it does not have the ability to “think” laterally and expand beyond those investments.
Collaborative Filtering
Collaborative filtering addresses some of the limitations of content-based filtering. It uses similarities between investment portfolios and companies simultaneously to produce results. By calculating those similarities, so called embeddings, the model can identify investor A based on investments of a similar investor B. Calculating the embeddings means they are learnt automatically, eliminating the drawback of having to provide the key characteristics to the algorithm. The ESG strategies could be an example of such an embedding.
Advantages of collaborative filtering are:
- Embeddings are automatically learnt.
- The model takes a broader approach and can identify targets beyond peer group holders.
- The system does not need to know anything about the investors other than their holdings to produce results.
Disadvantages of collaborative filtering are:
- The model needs to have data points about your company to calculate the embeddings. If you don’t have enough known investors the algorithm has a problem.
- It is difficult to introduce side features. In our case, side features might include country or portfolio size.
When handling large amounts of data, the machine is at an advantage
Artificial intelligence follows a different approach compared to conventional investor targeting. It turns away from analysing data in a lopsided way. Similar peer group investments are no longer the sole criterion to identify potential investors. With artificial intelligence it’s possible to combine different models that analyse gigabytes of data.
Five key take-aways from applying AI to investor targeting
- Machine learning is well accepted for successfully solving similar problems. AI is more efficient and will produce superior results both in terms of input/output time as well as precision.
- Artificial intelligence will help you handle complexity. While taking more features into account, the machine’s output requires less human interpretation to prioritise your efforts.
- Results do not have to be a black box.
- Recommendations based on artificial intelligence are free of bias, analyse a broader universe and present their output on portfolio level.
- Target investor identification is only the starting point of investor targeting. Segmentation, personalisation, and the right engagement strategy turn targets into contacts and contacts into long-term relationships.
Artificial intelligence will work to your advantage in identifying potential increment investors. Making the machine learning tool an integral part of your workflows will save time and costs while yielding better results. To find out more about how the different algorithms work in target investor identification, visit our webcast. The 15-minute webcast goes into more detail how embedding works and provides an example of the differences in results using a merged shareholder analysis versus artificial intelligence.
