Author: By Kirill Odintsov
Both data science and analytics play key roles in any modern financial institution. But what are the similarities and differences between these two teams? What skills and personality types are best suited for each role? And how can we develop talent in both teams to ensure they drive real business impact?
In this article, I’ll attempt to answer these questions based on my experience working with and managing both types of teams across different geographies and companies. Please note that my conclusions are mostly driven from working in financial companies.
Starting with the Similarities
In simple terms, both teams share the common goal of using data to help the company achieve its objectives. The key difference lies in how they do it.
The Role of Data Science
Data Scientists support the business by automating and improving decisions through machine learning models. In financial institutions, the most typical example is credit risk modeling. A data scientist might build a machine learning model—most commonly using LGBM, XGBoost, or even logistic regression—to predict the probability of default for each client (default meaning the client fails to repay their loan). Based on this prediction, the institution can automate decisions like approval or rejection, pricing, and other underwriting parameters. Please note that while data scientists build the model how to practically use it is usually job of data analyst
This model-driven approach is far more efficient than using manual rules or evaluating clients one by one. Manual rules can’t handle many variables at once and are usually not statistically grounded. Manual underwriting, on the other hand, is time-consuming and lacks objectivity—and humans often struggle to account for multiple dimensions of data at once.
Another common use case is propensity-to-buy modeling, which estimates the likelihood of a customer taking a new financial product. These models help optimize call center resources and improve customer satisfaction by avoiding irrelevant marketing calls.
The Role of Analytics
Analytics, on the other hand, is less about building models and more about interpreting data to support and drive strategic decisions—typically at the portfolio level rather than the client level. For example, if credit risk starts increasing, a risk analyst’s job is to figure out why. What’s driving the change? What action should be taken to mitigate the risk increase? Will the new risk level be within the risk appetite of the company?
Reacting to the increased risk without understanding the root cause would not be wise. For instance, if risk increases due to a spike in identity theft, tightening risk policy for all clients indiscriminately may even worsen outcomes as often stolen identities might by prediction of risk model have even good risk so the tightening could decrease non-fraudster to fraudster ratio. The right move would be to tighten rules specifically related to identity verification or adding new rules that would help reduce fraud.
Skills and Mindset
Both roles require curiosity, analytical thinking, and the mental flexibility to revise one’s views when new data contradicts previous assumptions.
For data science, a strong mathematical foundation is essential—these individuals must understand the models, not just run scripts. If you’re only hiring people to run scripts, you might be better off using an AutoML solution, especially for building models where the cost of error is low and feedback loop for a model is short.
Another key skill is communication: a data scientist must be able to simplify and explain their models to the end user. In my view, the best model is the simplest one that works. Simple models are easier to monitor, implement, and explain.
Analytics, by contrast, is more like detective work. Unlike data science projects, which follow a relatively fixed structure (data collection → explanatory data analysis → feature engineering → modeling → model and feature selection → final model testing → implementation), every analytics project is unique without clear structure. Analysts need to form multiple hypotheses, design data-driven tests to validate or reject them, and iterate based on findings. The key is being able to cut down improbable branches of potential cause very quickly but surely. Sometimes the result of the work of an analyst is an action to be taken; sometimes it’s a realization that more data is needed—perhaps via set up of A/B testing to gather more data and test the hypothesis for which there is no answer in currently available data.
A crucial skill for analysts is knowing when to stop. You’ll rarely have full certainty, and endless analysis delays action. The ability to apply the 80/20 rule—getting 80% of the insight in 20% of the time—is vital. Good analysts also know how to break down their work into smaller steps and deliver early insights quickly, even before the full picture is available.
Common Pitfalls and Development Tactics
In data science teams, a common pitfall is being too far removed from the business context, resulting in models that work on paper but not in reality. Understanding the data and how they are generated is crucial. Most of the data sets in finance have biases (e.g. we don’t observe behaviour of clients who we rejected) and the job of the data scientist is to understand them and adjust the final models to them. Without a deep understanding of the business processes this is not possible. Another challenge is to build models that the business team needs, not which they describe. Often the business teams are not sure themselves and have unrealistic expectations. In analytics teams, the risk is stopping at describing graphs and findings without translating them into recommended actions. I’ve seen many analysts present beautiful graphs and describe them but not showing the connection between the graphs and what to do next. With modern LLMs increasingly able to describe and interpret graphs, this kind of output is losing its value.
To avoid these traps, the development approach must differ for each team.
For data scientists, giving them the occasional analytics project helps build business understanding. It’s also important to keep their project mix diverse—repeatedly building the same kind of model reduces creativity and limits learning.
For analytics teams, it’s essential to challenge their thinking: ask how they reached their conclusions, whether the logic holds, does the story make sense as a whole, is the story consistent with previous findings and if the result is actionable. Encourage storytelling, not just reporting.
How to further insure development
Personally, I believe in rotating people between analytics and data science roles when possible. Data scientists gain a deeper understanding of the data and their genesis, while analysts learn what tools and techniques are available to support more advanced use cases. This cross-pollination accelerates growth on both sides.
Read more at 10 Lessons for Business Leadership
