How Centers of Excellence Can Help Financial Services Firms Achieve AI Success

Centers of Excellence, CoEs, financial services, artificial intelligence, AI, AI initiatives, AI CoEs, Infosys, Microsoft, NTT Data, business processes, business goals, data scientists, data analysis, product management, project management, machine learning, ML, DevOps, annotated data, ROI, Centers of Excellence, CoEs, financial services, artificial intelligence, AI, business processes, ROI
How Centers of Excellence Can Help Financial Services Firms Achieve AI Success blog
Wilson Pang
Wilson Pang, Chief Technology Officer at Appen

Centers of Excellence (CoEs) are increasingly common in enterprises as they enable teams to develop successful projects consistent with their business needs and industry requirements. Today, financial services firms are looking to integrate complex artificial intelligence (AI) initiatives into their core business processes to provide business innovation, operational efficiency, customer experience improvements, and competitive advantage. An AI CoE can help these firms create a clear vision for the use of AI, identify business-driven use cases, keep up with the latest best practices, and share success stories to promote additional AI projects. Infosys, Microsoft, and NTT Data, for example, have all discussed the benefits of their AI CoEs in detail.

IT Leaders Driving Inside-Out Digital Transformation for Businesses

While many firms have hired teams of data scientists for AI projects, this isn’t enough. As a think tank, a CoE accumulates knowledge over multiple projects and connects the dots between the potential of AI and current business goals. An AI CoE can also help financial services firms develop a resilient way to collect and annotate data to ensure projects will work as expected and scale as required.

Here are the key considerations when creating an AI CoE.

Build the right team

A successful AI CoE requires pulling together expertise from multiple areas, including:

  • Machine Learning – ML engineers understand the architecture they need to build and train a model or series of models. They understand the different data types, how to test and tune models, and how to put models into production.
  • Data Analysis – Data analysts use business intelligence (BI) tools to understand the impact of the data. They may also work to identify issues with the data and devise strategies for fixing them.
  • DevOps/AIOps – DevOps (or AIOps – an AI-infused evolution of DevOps) launches ML models and manages the continuous delivery pipeline. DevOps is also responsible for managing cloud infrastructure.
  • Engineering – Engineers ensure proper connectivity between applications and the ML models.
  • Product Management – Product managers focus on the success of the business case. They should be experts in the area of the business developing the AI project, such as credit risk, fraud detection, or payment systems.
  • Compliance – A member of the compliance organization should ensure AI projects have strong security and comply with industry and region-specific data regulations, such as SOC2 Type II, HIPAA, GDPR, and CCPA.
  • Project Management – A project manager coordinates activities among the other groups and evaluates and curates the tools they use. As such, they must understand each group’s requirements as well as budget constraints.

AI models and training data

The success of every AI project depends on a high-quality ML model and abundant annotated data to train that model. The AI CoE ensures project teams take the right approach to developing ML models and collecting and annotating the training data.

To create machine learning models, firms can:

  • Buy –Generally the cheapest and fastest way to create ML models, but these models tend to be generic, which can limit the effectiveness and ROI of the AI project.
  • Build–Control over the model ensures alignment with business goals over the long term and gives the organization owner of the IP, but this is likely the most expensive option and requires having the people and resources to create, train, and test the model in the time required.
  • Outsource – Provide the same potential for alignment with business goals as building the model in-house but without the significant overhead. However, this approach can end up being more expensive if the outsourcer doesn’t have the experience and expertise to produce the required model.

Digitalization Led IT Revolution Transforming the Global Labor Market

A training data practice within the AI CoE focuses on the critical best practices for data collection and annotation. If the huge amounts of annotated data used for model training are biased based on sex, race, or other criteria, the resulting ML model will likely produce skewed results or results that are unfair to a particular group.  Avoiding bias typically means collecting and annotating data from all around the world and from all relevant groups. To ensure this, firms will likely want to rely on a proven third-party vendor to globally crowdsource the right number and types of annotators to fit each project.

Getting started

The goal of the AI CoE is to make AI a core part of a company’s products and services, with the power of AI continuing to increase over time. To set the right foundation for this and build momentum, be sure to build a strong business case for AI and the CoE, and then get executive buy-in. Next, work with project teams to complete smaller projects, and make sure you have a clear measurement in place so you can demonstrate and communicate success and encourage additional projects. Finally, expand the AI initiatives to a broader scale and drive more AI successes.