Making Sense of Ambiguity in Applied Analytics

We often assume that applied analytics begins with clear questions, reliable metrics and well-defined data. In practice, however, many valuable analytical projects start in ambiguity. I see teams operate in situations in which the problem is loosely defined, the data reflects past system behavior rather than objective truth and success metrics are unclear or openly debated. Not all hard problems are ambiguous. Many are complex but clear: We know the goal, even if reaching it is difficult. Ambiguity exists when that clarity is missing. We may not know what should be measured, which signals to trust or what a good outcome looks like. Applied analytics often begins here: turning vague questions into clearer ones before any serious analysis occurs.

To illustrate, consider a hypothetical social media platform trying to decide whether it has “enough” influencers for a particular domain, such as home fitness. At first glance, the problem looks simple – count the number of creators in the domain, compare supply against audience demand and decide whether to recruit more influencers. In practice, ambiguity appears immediately. What qualifies as an influencer? How do we define the domain? How do we infer demand? What does “enough” actually mean? The challenge here is not computational but conceptual. Before running any analysis, we should first decide what question truly needs answering.

Ambiguity in the Problem-Definition Space

In ambiguous settings, the first step is to state the problem as it is currently understood, with the acknowledgment that this framing may be incomplete or incorrect. Stakeholders typically describe symptoms rather than root causes. In some cases, no underlying problem exists. What appears to be an issue may instead require clearer definition, validation or ongoing monitoring rather than immediate intervention.

In the influencer example, the observed symptom might be that users repeatedly see the same influencer or that engagement growth slows in certain interest domains. From this single observation, several plausible problem statements can emerge. One interpretation is that the platform lacks enough influencers in the domain. Another possibility is a discovery or ranking issue, where the recommendation system exhibits popularity bias, overamplifying a small group of established creators and reinforcing a rich-get-richer dynamic due to exploit-heavy optimization and limited exploration. A third possibility is that the influencer pool itself is
skewed. If most creators share similar demographics, formats or production styles, the platform may appear saturated while still underserving meaningful segments of user interest. These explanations are not mutually exclusive. The apparent imbalance may reflect a mix of influencer availability, algorithmic bias, creator incentives, user-interest inference and measurement artifacts, rather than a single root cause.

When the problem itself is unclear, applied analytics should resist locking into a single explanation too early. Instead of treating the first problem statement as correct, it helps to pause and examine it. This means separating what is being observed from what might be causing it, being clear about assumptions and considering more than one possible explanation. At this stage, the goal is not to optimize or fix anything but to build understanding. Applied analytics under ambiguity becomes an iterative process of refining the problem before deciding how to solve it.

Ambiguity in the Solution Space

As analysts, we rarely arrive at a solution in one step. Each stage involves choices about methods, assumptions and how different pieces fit together. Even after a solution is built, it often needs testing, comparison and refinement, along with discussion with business partners. In ambiguous settings, solving a problem is not a single decision but a series of connected choices that gradually shape the final outcome.

In the influencer example, the solution may look simple at first: Define the topic, estimate audience interest and count how many relevant influencers exist. In practice, each step involves judgment. Changing how the domain is defined changes which influencers are included. Interest estimates often reflect what users were shown in the past, not what they truly want. Measures of supply depend on whether influence is defined by activity, reach or engagement. Additional factors such as recommendation bias, creator incentives and uneven visibility also matter. Over time, seasonality, language, geography and behavior changes further affect the results. For these reasons, the solution develops through iteration rather than appearing fully formed at once.

In practice, this works best when the solution is broken into smaller hypotheses, and each part is tested step by step. Iterative data exploration helps build intuition along the way. Regular check-ins and proactive updates with stakeholders help keep the work aligned with the real objective. Even then, the final solution may have limitations. Good applied analytics acknowledges these limits, explains the trade-offs and shows why the chosen approach makes sense given the constraints. Conclusions should be grounded in evidence and communicated clearly.

Recommendations based on my experience working through ambiguity in applied analytics:

1. Listing hypotheses and building intuition: When a problem is unclear,  start by writing down the different hypotheses, which helps separate what we observe from what we assume. Early analysis stays simple; basic summaries, rough comparisons with other business reports and quick visual checks are often enough to build intuition. At this stage, focus less on accuracy and more on understanding how the system behaves and where signals may be misleading.
2. Designing for iteration and guarding against blind spots: Ambiguous problems rarely have answers in a single pass. As understanding improves, definitions, metrics and methods often need to be refined. At the same time, repeated iteration can create blind spots if we keep looking at the data in the same way. To avoid this, we should deliberately explore alternative views, test different definitions
   and ask what might be missing.
3. Engaging in peer review, collective reasoning and trust building: Ambiguity makes results easier to question. Sharing work early with other analysts, scientists or domain experts helps surface hidden assumptions and strengthen reasoning. This process also helps build trust. When stakeholders understand how conclusions were reached and what limitations remain, they are more likely to place trust in the results – even when the answers are not definitive.
4. Avoiding analysis paralysis through deliberate stopping rules: Ambiguity can lead to endless exploration. Although some uncertainty is unavoidable, applied analytics must still support decisions. Ask whether additional analysis would meaningfully change the decision. If not, it may be time to stop.
5. Writing executive summaries with actionable recommendations: Ambiguous analysis only creates value if decision-makers know what to do with it. A short executive summary should be provided that clearly states what was learned, what remains uncertain and what actions are recommended. The recommendations should be concrete and feasible, not just analytical observations. Even with limited confidence, proposing thoughtful next steps is better than leaving stakeholders unsure how to proceed.
6. Treating execution as part of analytics: After decisions are made, track outcomes and feed them back into future analysis. Over time, this feedback loop reduces ambiguity and improves judgment. In practice, this works best when analysts partner closely with product teams and other stakeholders to execute recommendations, observe real-world
   effects and refine understanding together.

Ambiguity is a normal part of real-world problems. The most useful applied analytics work often starts before any modeling, by being clear about what problem we are trying to solve and what we do not yet know. Many challenges are not about complex computation but about understanding the situation and helping people make better decisions. In this sense, applied analytics is the art of solving business problems using science.