The interdisciplinary field combining statistics, programming, and domain expertise to extract insights and build predictive models from data. For agencies, data science capability is what separates teams that can advise clients on AI strategy from teams that can only execute on it.
Also known as applied data science, data analytics, statistical data analysis
Data science draws on statistics for analysis and inference, computer science for working with large datasets and building models at scale, and domain expertise for knowing which questions are worth asking and which answers are actually useful. The combination is what makes data science different from pure statistics (which may not scale computationally) and pure engineering (which may not know what to build).
Practically, data science work spans exploration (understanding data structure, distributions, and anomalies), modeling (building predictive or classification models from labeled data), evaluation (assessing model quality on held-out data), and deployment (putting models into production systems where they generate predictions). Each phase requires different skills and has different failure modes.
The field has evolved significantly with the rise of foundation models. Tasks that previously required custom model development, such as text classification, entity extraction, or image description, can now be handled by prompting a large pre-trained model. This shifts data science work toward problem framing, evaluation design, and integration rather than raw model building.
Agency work has always required analytical judgment: understanding what the data says, what it does not say, and what decisions it should drive. Data science formalizes that judgment into reproducible methods that can be applied to larger datasets and more complex questions than manual analysis permits.
It is increasingly a client expectation. Clients with in-house data science teams expect their agency partners to be able to have peer-level conversations about model selection, evaluation methodology, and data infrastructure. Agencies that cannot participate in those conversations are positioned as execution vendors rather than strategic partners.
It changes the strategic contribution agencies make. An agency with data science capability can recommend that a client test a new channel based on a predictive model of expected incremental reach. An agency without it recommends the channel based on intuition and industry reports. The first recommendation is harder to dismiss and easier to optimize.
LLMs have democratized entry-level data science work. Code generation tools allow practitioners without deep programming backgrounds to perform data exploration, run standard models, and generate visualizations. This shifts the value from writing the code to knowing what questions to ask and whether the answers make sense, which is firmly in strategy territory.
An agency strategy team is developing a channel investment recommendation for a consumer brand client. Rather than relying on industry benchmarks and category-level data, they pull two years of the client’s own campaign and conversion data, run a regression analysis on the relationship between channel mix and revenue outcomes, and build a simple predictive model for expected return at different spend levels. The analysis surfaces that paid social has been underperforming relative to its historical contribution at current spend levels, and that incremental investment in connected TV shows a higher marginal return. The recommendation is grounded in the client’s own data rather than generic channel heuristics, and the client funds the analysis as a quarterly deliverable.
The generative AI foundations module of the workshop covers how today’s models work, what they can and can’t do, and how to use them to deliver analysis that clients can’t get from a dashboard.