By Richard Saldanha1
1Founder and Managing Director, Oxquant
We are still decades away from a true generalized machine or artificial intelligence but quantitative methods in general can be used far more extensively than most people suppose. Whilst such methods have typically been employed in the search for investment performance, the asset management industry spends far less time considering how quantitative methods can be used to improve profitability through cost savings or better task efficiency. Data-based automation is here now and those firms that fail to understand and embrace this fact will struggle.
One of the biggest hurdles to applying good models with the aim of automatic decision-making is poor data management. The failure to design systems that can share data easily means that much time and effort is wasted wrestling data into the correct format. This assumes that relevant data are available in the first place. Many of the simplest questions posed by boards or risk committees can be difficult to answer because of incomplete data or the failure to collect the right data. Boards should not blame their IT departments for these failings. Instead, they should think in advance about the important questions that might need answering.
Planning carefully, collecting the right data, storing it in an appropriate manner (whether that is in a conventional relational, NoSQL or semantic graph database) and making sure that database can talk effectively to other systems are key elements in implementing any form of automation system. Think of a strong data backbone for the firm. Databases1 abound and any number of programming languages2 can be used in conjunction with them.
With the right data infrastructure in place, some form of rules-based automation should be a genuinely achievable goal for all firms.
There has been a tendency to focus AI/ML efforts in the front office. Because of this, the potential for cost savings through better technological efficiency may be much greater in the middle and/or back offices.
Some of the typical tasks undertaken in the middle and back offices and potential automation possibilities are outlined below. The suggestions are in no way the stuff of science fiction. Much can be accomplished with existing computing methods. Quants and IT professionals may find much glamour outside of the front office in the near future.
There are no hard and fast rules as to what functions are to be found where. In general, risk and performance management is found in the middle office. Sales teams dominate client interaction at present but that has already changed for retail-focused firms with a strong outline presence. Thus, elements of client service are increasingly finding their way into the middle office.
Table 38.1: A selection of middle office tasks likely to benefit from automation efficiency
| Task | Description and automation possibilities |
| Client risk appetite, mandate requirements and other guidelines | Encoding of risk appetite, mandate requirements and other guidelines in machine readable form for downstream monitoring and reporting [C, R]. |
| Performance reporting and attribution | Provision of (near) real-time on-demand performance reporting and attribution [C, R]. |
| Risk reporting | Delivery of risk analytics, stress and scenario analyses, capacity and liquidity estimation in a near real-time context [C, R]. Remedial actions or merely suggestions for what action might be taken [ML]. |
| Portfolio modelling | Theoretical portfolio modelling, what-if analyses, associated risk characteristics and provision of relevant commentary [C, R, ML]. |
| Manager reporting | Seamless manager reporting based on relevant holdings [C, R, ML]. |
| Economic insights | Provision of relevant economic commentaries [C, R, ML]. |
| Compliance | Trade costs analysis, mandate adherence [C, R]. Automatic mandate breach signalling and possibly automatic correction [C, R, ML]. |
Note: The letters inside the square brackets refer to the delivery mechanism, i.e. conventional computing [C] such as server batch processing or on-demand request; event-triggered robotic automation [R] replaces some or all of the actions of a human worker in the same role; machine learning [ML] mechanisms use logic to decide on what action to take based on problem analysis and might consist of anything from a simple regression model used in a predictive manner through to text analysis and natural language processing.
No asset manager can function properly without a good back office. In contrast with the middle office, the back office will tend to have plenty of staff and computing resources. It will also receive lots of attention, particularly when errors occur. The key questions centre on back office efficiency.
Table 38.2: A selection of back office tasks likely to benefit from automation efficiency
| Task | Description and automation possibilities |
| Account setup | Automatic account creation for large clients as well as individuals [C, R]. (Note what FinTechs such as Stripe, N26 and Starling Bank, to name but three, already do now.) |
| Account compliance | The ability to obtain information automatically from reputable government websites and other trusted sources should allow seamless confirmation for most clients [C, R]. |
| Pre-trade checking | Most investment management firms implement automated pre-trade mandate and other checks. It is merely only the degree of automation, efficiency and intelligence around the rules that typically deserve scrutiny [C, R]. |
| Trade processing | Trade exception reporting should require no human intervention [C, R]. Automated correction should be a key focus for all [C, R, ML]. |
| Trade reconciliation | Reconciliation between broker, custodian and internal books and records should be fully automated [C, R]. Subsequent actions around breaks/exceptions is where intelligent automation can play a big part [C, R, ML]. |
| Post-trade compliance | Automated post-trade and daily close-price monitoring should be achievable for all firms [C, R]. |
| Fund administration | Automated rules-based fund administration is perfectly achievable with the right infrastructure [C, R, ML]. |
| Transfer agency | Much scope exists for transfer agency activity to be fully automated [C, R]. |
| Client reporting | Clients already expect a degree of automated and/or on-demand reporting [C, R, ML]. There is scope for much greater sophistication in this area. |
Note: The abbreviations given inside the square brackets are explained in Table 38.1.
Firms interested in intelligent automation should review their current data-handling procedures as a matter of priority. Typical questions to ask might include:
Considering data collection, storage and handling seriously is a necessary first step in any form of intelligent automation. Whilst it may be time-consuming to do this, implementing the right data policy may produce immediate benefits in terms of efficiency. Firms should only then think in terms of
AI/ML is changing the way investment management services are delivered.3 Those institutions which understand the power of their data and have learnt to harness AI/ML methods to their advantage can expect to be rewarded.