Emad ElSheshtawy – AI Manager Services Engineering at Finaira.
Introduction
AI discussions often start with job titles. Data Scientist. AI Engineer. Machine Learning Engineer. The assumption is that choosing the right title will lead to successful AI outcomes.
In practice, AI initiatives succeed or fail based on something far more fundamental: the type of problem being solved. Financial AI problems are rarely one dimensional. They sit at the intersection of data uncertainty, decision making, and software systems. When teams are designed around roles instead of problems, organizations end up with stalled initiatives, fragile solutions, or systems that never reach production.
The real question is not whether a company needs a Data Scientist or an AI Engineer. The real question is what kind of problem the organization is trying to solve.
The Data to Software Spectrum
AI problems exist on a spectrum. On one end are data dominant problems, characterized by uncertainty, noisy signals, and the need for exploration. On the other end are software dominant problems, where reliability, determinism, and system behavior matter most.
On this spectrum, Data Scientists naturally lean toward the data heavy side. Their strength lies in understanding data, discovering patterns, and reducing uncertainty. AI Engineers naturally lean toward the software heavy side, where building dependable systems and integrating intelligence into real world operations is critical.
This is not a hard boundary. Strong practitioners can move across the spectrum. Data Scientists can write production code, and AI Engineers can train and evaluate models. The distinction is not about capability. It is about where each role is optimized to deliver value efficiently.
The title does not define the ceiling. It defines the center of gravity.
Problem Types, Not Job Titles
Reframing AI work around problem types rather than roles clarifies where each capability is most effective.
Data Dominant Problems
Data dominant problems are characterized by ambiguity and exploration. The primary challenge is understanding whether useful signals exist at all.
Examples include demand forecasting exploration, risk signal discovery, and customer behavior analysis. In these problems, feature discovery, statistical rigor, and uncertainty estimation matter more than system architecture.
Data Scientists excel in this space because they reduce ambiguity and help organizations avoid investing in problems that cannot be solved reliably with data.
Software Dominant Problems
Software dominant problems are characterized by operational constraints. The intelligence may already exist, but the challenge lies in making it run reliably at scale.
Examples include real time decision engines, automated workflows, and AI embedded into operational systems. Here, latency, stability, observability, and cost control are critical.
AI Engineers excel in this space because they reduce operational risk and ensure that intelligent behavior is dependable in production environments.
Hybrid Problems: The Reality of Financial AI
Most meaningful financial AI problems are hybrid. They combine uncertainty from data with strict operational constraints.
Examples include cash demand forecasting combined with replenishment decisions, fraud detection paired with automated response, or risk scoring linked to policy driven actions.
These problems cannot be solved by predictive models alone or by software systems alone. They require layered solutions.
Predictive AI and Agentic AI Working Together
Consider a financial operational problem where future demand or risk must be estimated and actions must be taken under constraints.
The predictive layer focuses on estimation. Models forecast outcomes, quantify uncertainty, and continuously learn from new data. This layer is naturally aligned with Data Science capabilities.
The agentic layer focuses on action. Agents apply business rules, enforce constraints, coordinate steps across systems, and decide when and how to act. This layer is naturally aligned with AI Engineering capabilities.
Prediction informs decisions. Agentic systems operationalize them. Together, they form a complete decision making system rather than isolated components.
Why Specialization Still Matters
In financial systems, reliability is not optional. Context switching across deep statistical work and deep system engineering introduces risk.
While generalists are valuable, specialization allows teams to move faster and with greater confidence. It is not about what someone can do in theory. It is about what they can do consistently in production environments.
Implications for Building AI Teams
Effective AI teams are built by starting with the problem. Leaders should identify whether a problem is data dominant, software dominant, or hybrid, then assemble capabilities accordingly.
Predictive AI and agentic AI should be treated as complementary layers of a single system. Designing teams around these layers rather than job titles leads to clearer ownership, reduced delivery risk, and more sustainable outcomes, especially in regulated and high impact environments.
Conclusion
The most effective AI organizations do not choose between Data Scientists and AI Engineers. They understand the problem space first, then align capabilities to solve it end to end.
By shifting focus from roles to problems, organizations move closer to building AI systems that deliver lasting business value rather than isolated technical successes.