AI starts to take director seat in the boardroom

Algorithm appointed board director” was the title of an article on the BBC News website on May 16, 2014. “Artificial intelligence gets a seat in the boardroom” was a similar headline three years later on 17th May 2017 on the Nikkei Asian Review news website.

Both articles were referring to a computer algorithm called Vital that had been “appointed” to the board of directors of a Hong Kong venture capital firm known as Deep Knowledge Ventures. Citing the Nikkei Asian Review article, “Dmitry Kaminskiy, managing partner of Deep Knowledge Ventures (DKV), believes that the fund would have gone under without Vital because it would have invested in "overhyped projects." Vital helped the board to make more logical decisions, he said.”

By using an algorithm that could sift through masses of data on past investments, the company was able to narrow down on what the least risky investments were in the biotech space that they were playing in.

The article continues, “DKV started as a traditional biotechnology fund, with a team of advisers and analysts using traditional methods for trend analysis and due diligence. But the biotech sector has a very high failure rate, with around 96% of drugs not successfully completing clinical trials. DKV then acquired a team of specialists in the analysis of big data – large data sets that can be analysed by computers to reveal patterns. The team created Vital, the first artificial intelligence system for biotech investment analysis, enabling the fund to identify more than 50 parameters that were critical for assessing risk factors. Kaminsky said: ‘ As we analyzed more and more companies, we were failing to identify those patterns and factors that made a company likely to achieve success. But surprisingly, as we began to analyze thousands of companies, we discovered certain parameters that were good at predicting the risk of failure.’ ”

The primary role of a director is twofold: a monitoring and oversight role of past decisions made by management and a forward looking role to oversee formation and execution of strategy. In the DKV example cited above, the role of the algorithm was to help the venture capital board make the right investment decisions. Using big data, the algorithm was able to narrow down which specific drug research areas were yielding better outcomes and provided support to the board on which drug companies to invest in.

How could this translate to other non-investing type of companies? It is easy to draw a parallel to the banking industry for example where bank boards have to review and approve lending decisions based on analysis that has been done by a credit manager.

While smaller loans have already moved to algorithm-based decision making (M-shwari is a good example), the bigger and more complex loans still require human analysis largely due to a poor use of big data within the banking industry. Not sharing historical lending data, which can be easily done on a no-name basis to protect client confidentiality, prevents the banking industry from building a critical database that can be used to provide granular risk patterns for different market and industry segments.

While it can be argued that the information is being shared at a credit reference bureau level, what remains to be seen is how this information can be consolidated, analysed and churned back to the banks to use for determination of probability of repayment. But credit risk analysis which is largely technical, is mainly a management undertaking, and brought to the board for approval.

Having AI sort out that decision at management level would significantly reduce the work of the credit committee of the board. One can further argue that AI can also review the entire lending book of the bank, assess the current and potential portfolio at risk, and determine what amount of provisioning is required, as is currently demanded by the new international accounting standards. Which would then eliminate the need for numerous risk analysts within bank management.

AI could also potentially review the financial reports produced by management (if not produce the reports themselves) for accuracy. We could go very far with this argument, which is that if machines are able to do a lot more of the monitoring role that management undertakes and reports to the bank’s board, then technically, a lot of the work of the bank board can be reduced to oversight on the formulation and execution of strategy and the more human role of oversight of key stakeholder engagement such as employees, customers and regulators.

The DKV example is really a hyped version of a management decision making tool that is being elevated to board use. But it does spur some thinking for both directors and management on how daily operating decisions can be moved to more accurate algorithm driven processes.