BI & Communication
Analysis does not end with a computed number. The last mile is turning that result into a chart people read correctly and a message that drives a decision. This is where an analyst stops being a report generator and becomes a decision partner: the value is not the table, it is the action that follows from it.
The topic splits into two skills that interviews probe separately. The first is BI and visualization: a chart is an encoding, where the data carries a relationship (trend, comparison, distribution, composition, correlation) and a visual channel (length, position, angle, color) encodes the magnitude. Pick a channel the eye reads poorly, or bend one until it lies, and a correct calculation turns into misinformation. The second is stakeholder communication: how to land a result with people who do not read SQL — leading with the answer rather than the method, showing uncertainty honestly, and defending good metrics against bad requests.
Tool and technique names (Tableau, Looker, Power BI, BLUF, PII) stay in English throughout — that is the industry standard.
Topic map
- BI and visualization — the chart for the relationship (line / bars / histogram / scatter / stacked); dashboard design (audience, hierarchy, one message per view); the encoding traps (dual axes, truncated axis, pie overuse, chartjunk, too many colors); tools (
Tableau/Looker/Power BIat a concept level); pre-aggregation versus live. - Stakeholder communication — the answer in the first sentence (BLUF), tailoring to the audience, translating statistics into business impact, answering "why did X change", pushing back on bad metrics and requests, honest uncertainty, and the analyst as a decision partner.
Common traps
| Mistake | Consequence |
|---|---|
| Passing a dual axis off as correlation | The scales are stretched until unrelated series track; the correlation is an artifact |
| Truncating the Y-axis on bars | A 2% gap looks like a cliff; length from zero is distorted |
| Putting a pie chart on nine slices | The eye ranks angle and area worse than length; use a sorted bar instead |
| Leading with the method, not the answer | The stakeholder cannot tell what to decide; the answer is lost in methodology |
| Hiding uncertainty behind "it might" | A vague phrase conceals the risk rather than resolving it |
| Handing over a raw PII export on request | Breaks data minimization and access rules, re-identifies people |
| Re-running the analysis until the number fits | That is p-hacking: the result is bent to the desired conclusion |
Interview relevance
BI and communication are asked as a test of maturity, not tooling. A candidate who says "the first sentence is the answer to the business question and the action, not the sample sizes" immediately gets ahead of one who opens a presentation with methodology. The problems are almost always about judgment: which chart fits the question, how to show 70% confidence, what to give a stakeholder instead of a raw export, how to land a finding that kills an announced project.
Typical checks:
- How to pick a chart for the relationship and when truncating the Y-axis is legitimate versus deceptive.
- How a dual axis manufactures a false correlation and what to use instead.
- How to build a dashboard around a decision: what goes above the fold, what to cut, why nobody opens it.
- How to lead with the answer (BLUF), translate statistics into business impact, and show uncertainty and a null result honestly.
- How to defend a good metric against a bad request: pushback, "look at it again", the raw-data ask.
Common wrong answer: "the main thing is that the dashboard is pretty and detailed". In reality a dashboard answers one question for one audience; extra metrics, raw tables, and a rainbow of colors add no value — they dilute the message and erode trust in the numbers.