When Good Analysis Is Not Enough

In applied analytics education, student teams can produce technically sound analysis and still leave project sponsors unsure what to do next. Suppose a project sponsor asks the team to “understand retention.” The students clean the data, compare several methods, and identify patterns associated with attrition. Yet during the final presentation, the sponsor asks questions that catch the team off guard: Which groups should we prioritize? What action should we take? When should we intervene? What trade-offs should we expect?

The team has answered an analytical question,but the project sponsor still lacks enough guidance to make the underlying decision.

AI makes this gap more visible. It can reduce the time needed to generate code, compare methods, summarize information, and build prototypes. But the practical value of analytics increasingly depends on the ability to structure an end-to-end analytics workflow: clarifying the decision, developing an analytical agenda, evaluating evidence and uncertainty, and translating findings into actionable recommendations. Our experience in the Georgia Tech Business Analytics Practicum suggests that applied analytics courses are more effective when organized around the decisions students are trying to improve, and not only the models they are able to build.

The Georgia Tech Practicum Context

The practicum provides a real-world setting in which to test these ideas. Expected to be renamed the AI & Business Analytics Practicum beginning in fall 2026, courses will be offered at both the undergraduate and graduate level.

Student teams work with industry and community partners to address real business problems and develop practical analytics and AI-enabled decision-support solutions. Recent partners have included Comfama, The Home Depot, Cherry Street Energy, Sparck Technologies, and Augusta National.

These engagements not only create authentic learning opportunities, but also expose practical challenges in applied analytics. The client’s decision may initially be unclear. Available data may not fully support the original question. Teams may produce analytically sound findings that are disconnected from implementation realities. Final presentations may explain what the analysis found without clarifying what the client should do next.

These challenges make the practicum a natural environment for combining faculty expertise with practitioner perspectives on how analytics is designed and applied within organizations.

The Role of Industry Engagement

Practitioners introduce dimensions of analytics work beyond technical instruction: ambiguous problem statements, imperfect data, competing priorities, and constraints that may not be cleanly captured in a model. Industry projects also require teams to structure analytical recommendations that earn stakeholder trust and reflect organizational realities.

These considerations are especially relevant to Georgia Tech’s Business Analytics Practicum, in which practitioner engagement is key to helping students navigate applied project challenges. Guest lectures from experienced analytics practitioners play an important role in the practicum. Through case studies and firsthand examples, practitioners show students how analytics changes when it moves from the classroom into an organizational setting. They also broaden students’ understanding of how analytics is applied across industries and business functions, from supply chain management to marketing.

The value becomes even greater when practitioner experience is distilled into a structured approach that students can apply beyond the classroom. During the past two semesters, we have introduced decision-centric principles in the practicum to help students connect analytics design with organizational decision-making. These principles can guide students in project framing, interim reviews, final recommendations, and client presentations – areas that may receive less explicit focus in technically focused coursework. Faculty contribute expertise in course design and student learning, while practitioners contribute their current experience with how analytics is designed and implemented inside organizations. Together, they can turn industry exposure into a repeatable part of the learning experience.

A Decision-Centric Approach

In the practicum, student teams often begin with a sponsor’s broad business challenge and must translate it into a well-defined analytics project. Students’ initial instinct is often to take a method-first approach. Once they receive the data, they begin asking which model to use, which variables to include, or how to improve predictive accuracy. Although these are important questions, they are not always the best place to start.

A decision-centric approach starts with problem framing: What decision is the organization trying to make? Who owns it? What actions are available? When must it be made? What resources and constraints limit the options? What evidence would cause the decision-maker to choose one action over another?

There is rarely one correct answer. Teams may first need to explore alternative problem definitions, stakeholder perspectives, data sources, and potential actions. They must then converge on a tractable decision, a defensible analytical approach, and an implementable recommendation. AI can support both exploration and execution, but it can also encourage premature convergence by producing a fluent first answer that appears complete but lacks substance.

Students must therefore consider the broader context surrounding the analysis. A single key performance indicator may be useful, but it rarely captures the full decision context. To develop recommendations that project sponsors can act on, student teams need to account for resource limitations, stakeholder incentives, operational dependencies, timing, equity considerations, and second-order effects.

To move students away from method-first thinking, we organize projects around four interdependent elements: data, analytics, decision, and narrative. Beginning with the decision changes the questions students ask about every other element:

• Data: Does the available information provide the scope, granularity, quality, and timeliness required by the decision?
• Analytics: Are the methods reliable and relevant for the intended use?
• Decision: What objectives, available options, constraints, and trade-offs shape the decision?
• Narrative: How should the findings be framed so that stakeholders can understand and act on the recommendation?

The four elements are iterative. As teams learn more about the sponsor’s business context, objectives, and data, they may need to revisit the problem statement and refine the analytical agenda. This shifts the focus from “What analysis can we perform with this data?” to “What decision is at hand, and what analysis would improve it?”

In one engagement with Comfama, a Colombian social impact organization, the initial challenge centered broadly on understanding changes in its affiliated population. A method-first team might have begun by testing forecasting models. A decision-centric team, on the other hand, starts by asking what organizational outcomes the forecast is intended to improve, such as workforce planning, service allocation, employee retention strategy, or a combination of the three. 

Clarifying the decision objective first influences downstream analytics design, including the required time horizon, level of geographic detail, outcome measures, and acceptable uncertainty. Under this decision-centric approach, the model is no longer the project’s starting point; instead, it becomes one component of a broader decision-making workflow.

Incorporating Decision-Centric Principles Into Student Projects

Decision-centric principles are most valuable when students apply them consistently throughout a project. Early in the process, teams should keep the problem open long enough to understand the decision the sponsor is trying to make. They should explore alternative problem definitions, stakeholder perspectives, and data sources. As the project develops, teams must commit to an analytical approach, execute, and refine based on feedback. The challenge is knowing when to broaden the inquiry, when to commit to a direction, and when new evidence warrants reopening the question.

The decision-centric principles help guide this iterative process across data, analytics, decision, and narrative. New data may change the problem definition. Analytical results may reveal previously hidden trade-offs. Sponsor feedback may expose implementation challenges or change how the recommendation should be framed. Interim reviews can help teams reassess whether the analysis still supports the right decision and refine the analytical agenda accordingly. 

In an AI-enabled environment, student ownership is essential. While AI can accelerate exploration and analysis, students should not delegate judgment to it. They remain responsible for defining the decision, evaluating the data, verifying the analysis, and shaping the narrative. Before using AI, teams should be encouraged to complete a brief ownership check: What do we believe is happening? What evidence supports that view? What remains uncertain? What would change our conclusion? Which parts of the recommendation must we personally verify and defend?

Course milestones and evaluation can reinforce this ownership. Instead of rewarding the fastest plausible answer, the course can reward teams for identifying data limitations, framing decisions clearly, calibrating their claims, and developing actionable recommendations. This creates productive friction and encourages students to develop and defend decision-ready analytics.

Lessons for Industry-Academic Collaboration

The practicum experience suggests that industry engagement is most valuable when practitioner insight is converted into something students can repeatedly apply. Paired with a guest lecture, a structured framework can both broaden students’ perspectives and shape how they approach the project itself.

Effective collaboration depends on clearly defined roles. Faculty design the learning sequence, assessment, and feedback process to ensure that practitioner insights are closely aligned with overall course objectives and flow. Practitioners help curate representative case studies or project work to surface the ambiguity, constraints, and stakeholder dynamics of real analytics work. Project sponsors can add another critical perspective by clarifying the decision context, feasible options, and stakeholder expectations.

Applied projects should not shy away from AI, specially as it becomes widely used across industry. Instead, programs should be intentional about guiding students on how to use it effectively. AI can support analysis, but students still need to verify outputs, exercise judgment, and engage with stakeholders. Practitioner involvement can help connect classroom guidance with the realities of how AI tools are applied inside organizations. Practitioner-informed frameworks can also help ensure that AI expands inquiry rather than encourages premature convergence.

The goal is not to add disconnected assignments or replace technical instruction. Instead, industry-informed principles can be embedded into existing project milestones, feedback, and evaluation. This makes the collaboration easier to sustain across cohorts and helps students carry these learnings into their own work.

From Practitioner Exposure to Applied Learning

Applied analytics education should prepare students to improve decisions instead of only producing analyses. AI raises the stakes by lowering the friction in technical production, while problem definition, integration, trust, and adoption remain challenging.

Decision-centric principles give students a durable way of working in this environment: explore before converging, retain ownership of the question, and view the problem as a broader system rather than a single metric. AI should expand the field of exploration, not close it prematurely. The goal is to develop recommendations that are evidence-based, open to challenge, and practical enough to act on.

References 

Zhu, J., 2025, “How One Google Team Built Storytelling Into Analytics,” MIT Sloan Management Review, https://sloanreview.mit.edu/article/how-one-google-team-built-storytelling-into-analytics/. 

Zhu, J., 2025, “Beyond Use Cases: A Strategic Framework for Scaling AI in B2B Sales,” Analytics, https://pubsonline.informs.org/do/10.1287/LYTX.2025.04.18/full/. 

Fan, Z., Von Behren, S., Narasimhan, S., García Rivera, S., 2025, “Forecasting Social Impact: How Georgia Tech Students Are Helping Comfama Predict the Future of the Middle Class in Colombia,” OR/MS Today, Vol. 52, No. 3, https://pubsonline.informs.org/do/10.1287/orms.2025.03.11/full/. 

Kapur, M., Bielaczyc, K., “Designing for Productive Failure,” Journal of the Learning Sciences, Vol. 21, No. 1, pp. 45-83, https://www.tandfonline.com/doi/full/10.1080/10508406.2011.591717. 

March, J. G, 1991, “Exploration and Exploitation in Organizational Learning,” Organization Science, Vol. 2, No. 1, pp. 71-87, https://strategy.sjsu.edu/www.stable/pdf/March,%20J.%20G.%20%281991%29.%20Organization%20Science%202%281%29%2071-87.pdf.