Understanding and Roadmapping AI / ML Program Implementation

By George Marootian, Executive Vice President and Head of Technology for US Distribution, Natixis Investment Managers

In this role, Mr. Marootian spearheads initiatives that leverage technology to help drive the firm’s revenue growth, digitalize capabilities and manage cost containment.

How do you distinguish between machine learning and artificial intelligence? Where does natural language processing fit in this discussion?

George Marootian, Natixis

Based on my experience, artificial intelligence is the first iteration of an Artificial Intelligence (AI) / Machine Learning (ML) program, then transforms into quantitative “inputs” of a learned program. The initially developed model is the beginning stage, and ML is the continuous improvement and continuous delivery of refinement of the initial model (algorithm). The continuous delivery ML program is to identify data sets and inputs, and evolve an algorithm by carefully applying inputs that improve its quality and accuracy. Parallel to this program, there can be continued development of models and data preparation capabilities. The key elements of the lifecycle are:

A. Data Gathering

B. Data Preparation

C. Data Analysis

D. Model Execution

E. Model Evaluation

F. Model Test

Steps A through C cover the AI segment of the lifecycle, and Steps D through F cover the ML piece, which complete a full iterative cycle. These initial steps are imperative in order to extract the highest amount of value from the program.

Capabilities like Natural Language Processing (NLP) are blended into the program, as it provides a continuous stream of qualitative inputs that have been transformed into quantitative inputs. As with any modern model-based platforms, having the ability to “blend” in qualitative inputs is important to identifying metrics to improve the quality of the result derived.

Qualitative inputs have an immediate effect on Steps A through C, in that they increase the volume of data being sourced and analyzed. The greatest value of these inputs are demonstrated with Steps D through F, where business owners can see the solutions being delivered. The ability to quantify these qualitative factors, and direct actions against them, is the very competitive advantage that most organizations will seek once the qualitative parts are commoditized in the market.

All of these components will need to be defined to work in concert to continuously drive the evolution of a successful AI / ML program.

How specifically is machine learning being applied in the buy-side front office?  

There are a multitude of ways to apply machine learning to the buy-side front office. Identifying where AI / ML may create new opportunities depends on the areas a particular organization chooses to prioritize.

Following are a few use case scenarios to identify where applying AI / ML may create new opportunities:

Equity research – creating a list of top ten opportunities based on market research, sector analysis and qualitative factors (e.g., news feeds, political influence, social media)

The equity research use case could optimize the process of security selection and analysis by taking the rote operation of base research and data compilation and leveraging the power of commodity computing to deliver highly accurate, highly consistent and more timely results to decision makers.

Account level – analysis of trade to determine tax benefit (e.g., loss harvest), optimum position keeping and benchmark alignment

The account-level use case could pre-vet a trade decision by weighing all of the key factors, and generating the required compliance details to support the decision of execution. This change would deliver higher accuracy and improved consistency and would support a movement to apply the human factor to the final decision versus the assembly of facts and details.

ESG alignment – on-going analysis of ESG benefit of portfolios based on trade activity

The ESG alignment scenario could pre-vet a trade decision by weighing all of the key factors around ESG and generating a “score” that could be used both internally and as a selling point to clients.

The above outlines a few potential scenarios based on my experiences and conversations with peers, but each firm has a unique starting point that begins with the organization determining where the application of AI / ML will provide the highest level of value. Regardless of the chosen approach, as with any software project, AI / ML strategies should be considered and executed in an organized and prioritized manner versus implementing wide spread change.

What are the primary challenges in implementing machine learning?

As previously noted, I believe a successful machine-learning program requires a significant investment in People, Process and Technology. Proper investment in each pillar can ensure that the program creates a transparent and supportive operational model for the target internal client (e.g., trading, research, sales teams). Even with a significant investment in these key pillars, challenges can remain:

People: Exemplifying not only the correct skill set (i.e., development and data analysis), but also an understanding of the space is critical to successful staffing. I have heard of many instances where teams were created by assembling current staff that may have performed reporting and visualization tasks. Those skills are important to both the front and back office solutions; however, they are not forward-facing capacities, which are key to implementing machine learning.

Process: As organizations continue to define and leverage algorithms to drive decision-making, automate operations and engage compliance, these processes will need to be verifiable, complete and accurate. The ability to define clear processes will not only aid in development of solutions, but it will also make the transition from idea to operations much easier. With that in mind, it is important to approach an AI / ML multi-year program with a process-oriented mindset at the very start.

Technology: Retaining the technology to implement these new capabilities is not only important but is paramount to gaining timely analysis to support decision-making. The capabilities around High Performance Computing, Robotic Process Automation (RPA) and Machine Learning (ML) have become commoditized by the major cloud platforms (e.g., Amazon, Microsoft, Google). This commoditization has allowed all organizations to increase their computing capabilities at minimal marginal cost, as opposed to the large capital investments required less than a decade ago. The key with the technology piece is to engineer a platform – but not restrict the future growth of ML initiatives – that can address both current needs and supporting areas such as:

  • Data aggregation and organization
  • Data analysis
  • Processing
  • Solution generation
  • Learning / Regeneration

Artificial intelligence and machine learning initiatives are long-term programs. Programs of this magnitude and reach should be invested with a forward-thinking approach that includes a roadmap outlining both the short- and mid-term deliverables that will drive value. Assessing the program on an annual basis against measurable success points is key to maintaining strategy alignment. •

This material is intended for informational purposes only, does not constitute investment advice and should not be construed as a recommendation for investment advice.

There can be no assurance that developments will transpire as forecasted. Actual results may vary. The views and opinions expressed may change based on market and other conditions.  

Natixis Distribution, L.P. is a limited purpose broker-dealer and the distributor of various registered investment companies for which advisory services are provided by affiliates of Natixis Investment Managers. 

Understanding and Roadmapping AI / ML Program Implementation first published in the Q1 2021 issue of GlobalTrading.