Matthieu Bleuse, principal at the asset management consultancy Alpha FMC, looks at common challenges for investment firms developing AI and machine learning capabilities.
Funds which are not developing AI and machine learning capabilities to support their investment processes may soon find themselves at a competitive disadvantage. Beyond improving investment decisions, these programmes help to predict client behaviour, efficiently process contractual documents, and explore new ways to interact with client-bases. However, asset managers still face a number of significant operational challenges when it comes to integrating AI into their value chain.
The integration of extra-financial data is key to developing more efficient prediction algorithms, successfully complementing traditional models based exclusively on financial data. One common example is the analysis of news feeds, online and on social networks, and its predictive power over financial variables. Another example is the flow of traffic on corporate websites of companies in the investment universe, or on the textual analysis of language during corporate earnings announcements.
However, to make full use of such information, investment firms will have to build and deploy more data-centric strategies. Firms will then face the challenge of structuring raw data into meaningful information. Acquiring and processing this type of data in-house is still far from the core business of traditional asset managers, which generally opt for an outsourcing model, strategic partnerships, or acquisitions. Multiple fintech companies have already developed offerings dedicated to capital market players to help with structured data.
Incubate or outsource
Investment firms are increasingly developing AI programmes in-house in dedicated innovation labs, at least for design and incubation phases. Labs allow programmes to be tested away from the firm’s operations until it is ready to be deployed and ensures that solutions are truly disruptive.
However, incubating AI development requires, in addition to high capex, access to very specific data science skillsets, which is sometimes difficult for an investment firm to acquire, and potentially leads to long go-to-market phase. If this barrier to entry is too high, fund managers can opt for business models based on decentralised intelligence, in which the construction, training and production phases of the machines are outsourced to more specialised players.
We anticipate that this type of decentralised model could be implemented more systematically between investment firms and specialised fintechs. The role of fund managers will then include designing the general architecture of such a program, consolidating all the decision signals and then constructing the portfolio.
Lack of transparency has long been a criticism levied at the investment community. Transparency around management processes is now a pre-requisite for ensuring investor trust in such algorithmic programmes. Investment managers must ensure they implement the appropriate control frameworks, notably in the pre-trade sphere, and provide investors with a sufficient level of transparency around investment processes and risks, which is complex to integrate in such an investment approach.
Beyond the role of portfolio constructor, fund managers should not lose sight of their ability to constantly explain and control the way in which AI is applied, to comply with an investment discipline for which they will remain the ultimate guarantor.
©2018 funds europe