As the analytics landscape has matured over the last two decades, the topic of how to build an efficient AI team in the financial services sector has become a priority from the C-suite on down. Today, organizations are embracing AI, such as the promise of generative AI and large language models (LLMs), and thinking about the right size investing in the mix of people, tools, roles, and collaborations.
Generative AI is set to touch every aspect of financial industry operations, from sales to customer service. The need for timely feedback and iterative progress in implementing the novel creative process of generative AI means that the speed, breadth, and magnitude of team collaborations have to be greatly expanded with efficient teams. New AI capabilities comprise prompt engineering to output novel ideas and designs, incorporating custom data into private LLMs for conditional generation, and choosing a balanced architecture with privacy and ethical considerations. Having an efficient AI team will determine the value of the content generative AI can create in the financial services sector.
Back to the Future: AI Teams Than & Now
A decade ago, the trend for many financial firms was to have Analytics Center of Excellence (CoE) representing significant dollar investments. New CoEs would spring up, sometimes comprising hundreds of talented data scientists. Fast forward, and CoEs are mostly gone, replaced by Amazon’s “2 pizza” teams, Spotify’s squads, Airbnb’s ‘autonomous’ teams, and Uber’s ML teams, among others. These new team structures are small, highly dynamic, creative organisms that often have less than ten people in diverse roles while taking innovation to new heights.
What forced AI/ML teams to change?
Looking at AI team changes begs the question of why the metamorphosis, given that organizational goals have remained unchanged. The simple answer is technology. The CoE focus was on tools due to the large upfront investments in foundational technologies, which required ample support from the engineering team along with the dedicated large analytics team. This choreographed cast of talent was built on delivering key results on the same scale as the financial outlays – which often it did not. This paradigm broke down once the Xaas cloud revolution started with open source and web services. This new reality is fueled by the machine learning, AI, and data (MAD) landscape, with hundreds of domains served by thousands of firms vying to offer pay-as-you-go services.
AI Team: Towards a Lovable AI Product
Centralized large teams that could not efficiently devolve knowledge to business units because of focus on centralized technical resources gave way to a more enlightened focus on core outcomes of producing a lovable AI product. This was possible by abstracting or sublimating technology into an enabling role, which allowed the AI team's focus to be set on tangible outcomes, measuring success through user satisfaction and impact. Shifting the spotlight away from technological intricacies and towards user experience, the team empowers itself to craft solutions that truly resonate with audiences. Balancing swift iteration and thoughtful refinement is imperative, ensuring timely execution. Moreover, nurturing a continuous state of ‘flow’ within the team cultivates a creative and efficient atmosphere, fostering innovation and collaboration.
The ‘Secret’ to AI Team Structure & Function
An emerging open secret of successful AI teams is that data scientists work in sync with products to support data-driven decision-making for customers. This is enabled by seasoned data scientists embracing the entire process continuum from understanding the business problem to deploying a model pipeline. Github repo to API. This is currently possible because of the many available end-to-end Machine Learning platforms, such as Databricks, Azure, AWS, and others.
“Generative AI is set to touch every aspect of financial industry operations, from sales to customer service.”
The new reality emphasizes the empowerment of data scientists to take care of much of the CI/CD pipeline with plan, building, testing, releasing, monitoring, and focusing on AI learning patterns implicitly, while data engineers, who implement rules explicitly, have a minimized role of operating the deployed models.
Resiliency: Outliving the Hype Cycle of an AI Team
Since AI teams take advantage of novel technologies, it follows their rise and fall with the technology hype cycle. Embracing generative AI, we have to think of team resiliency in planning and executing with the goal of making it past the ‘Peak of Inflated Expectations’ cycle of 2-3 years. What the hype cycle d oes not show is the productivity yield, and that productivity must align with initial expectations. Unsurprisingly, the ultimate goal is delivering productivity throughout the ‘honeymoon’ period. Reaching the singularity point, where productivity equals expectations, is the pivotal moment of team achievement when AI products resonate with stakeholders, offering tangible and quantifiable business value. Outliving the hype cycle and surpassing the transient phase of technological hype means delivering on AI team productivity from day 1.
Promoting Team Flow
Managing a thriving AI team lies in the seamless progression of talent, resources, ideas, and results. Cultivating an environment that promotes the concept of team ‘flow,’ the team experiences evolution. To endure over time, the team must embrace systems thinking such as, for example, the design principles of Constructal Law. This involves the intricate orchestration of teamwork processes - a dynamic in which the pursuit of greater access to resources is pivotal for the team's longevity. As the configuration of the team's flow changes, it drives evolution. This continuous adaptation ensures the AI team's vibrancy and progress, setting the stage for enduring success.
Avoiding Pitfalls
Constantin Brancusi, the founder of modern sculpture, wrote, "The challenge lies not in the act of creation itself, but in cultivating the right mindset for it." Similarly, navigating the path of AI team success requires vigilance against common pitfalls. Striking a balance between team learning and execution is crucial. If the AI team spends most of its time training, then it doesn’t have the required talent and needs new hires with requisite capabilities – instead of the ultimate learning team, remain open to augmenting talent. A blend of direct and contingent hires adds versatility. To stave off complacency, team diversity is paramount. Also, avoid blind adherence to Agile dogma that might lead to the deceptive ‘Busyness Trap.’ Agile methodologies should empower rather than mask genuine progress. There's no grace period; productivity commences from Day 1. Finally, early showcases of successful results should take precedence over delays. These principles serve as guiding beacons to a resilient and accomplished AI team journey toward business value.
