Business Intelligence & Analytics Communication Exercises

A metric is any quantitative measurement — page views, revenue, session duration. A KPI (Key Performance Indicator) is a metric that has been selected because it directly measures progress toward a strategic goal, usually with a defined target and time frame. In stakeholder communication: "We track dozens of metrics, but our KPIs for this quarter are monthly recurring revenue, churn rate, and activation rate — these are the three numbers that tell us whether we're on track against our growth targets."

Lead with the business question the dashboard answers, then move to the current value, trend direction, and action required. Example: "This panel shows our conversion funnel for new signups. Conversion from trial to paid is currently at 18%, down 3 points month-over-month. The drop is concentrated in the mobile segment, where checkout completion fell sharply after last week's release." Avoid raw SQL or data model details — translate every number into a business implication.

The grain of a fact table defines exactly what one row represents — it is the level of detail captured. For example, the grain might be "one row per customer per day" or "one row per individual transaction." Establishing grain is critical when designing a warehouse because it determines which dimensions can be joined and what aggregations are valid. In a schema review: "Before we finalize the fact_sales table, we need to agree on the grain — are we capturing one row per order or one row per order line item?"

Key terms include: top of funnel (TOFU) for awareness-stage users, middle of funnel (MOFU) for consideration-stage, bottom of funnel (BOFU) for conversion-ready users, drop-off rate (the percentage of users who leave at a given step), and conversion rate (the percentage who complete the funnel). You might say: "We have a 62% drop-off between the pricing page and checkout initiation — that step is our highest-priority funnel optimization target."

A cohort is a group of users who share a common characteristic or start date. Cohort analysis tracks how each group behaves over time rather than looking at all users as a single pool. Practical explanation: "Instead of asking what our average retention rate is, a cohort analysis asks: of all the users who signed up in January, what percentage were still active after 30, 60, and 90 days? This shows whether retention is improving for more recent cohorts."

In a star schema, dimension tables connect directly to the central fact table with no further normalization — the structure looks like a star. In a snowflake schema, dimension tables are further normalized into sub-dimensions, reducing redundancy but requiring more joins. In practice: "We chose a star schema for this data mart because analyst query performance is the priority and the storage overhead of denormalized dimensions is acceptable at our data volume."

Lead with the insight, not the methodology: "The data shows...", "What this means for the business is...", "This number tells us that...", "If this trend continues, we can expect...". Avoid jargon like p-values or confidence intervals unless the audience is comfortable with statistics. Instead: "We're confident this is a real trend rather than random noise because we've seen it consistently across three different time periods and two independent data sources."

A reliable framing: "We can see that these two metrics move together — when support ticket volume rises, NPS drops. But we can't conclude from this data alone that one causes the other; there may be a third factor driving both. To establish causality, we'd need to run a controlled experiment or do deeper root-cause analysis." Using phrases like "associated with," "correlated with," or "moves in tandem with" signals appropriate statistical caution.

Window functions are frequently discussed: ROW_NUMBER(), RANK(), DENSE_RANK(), LAG(), LEAD(), and running totals with SUM() OVER(). Other important terms: CTEs (Common Table Expressions) for readability, materialized views for pre-computed aggregations, and partitioning for query performance. In a design review: "I used a CTE to stage the intermediate cohort calculation, then applied a LAG window function to compute the week-over-week change — this keeps the logic readable and avoids a self-join."

Data freshness describes how up-to-date the data in a report or dashboard is relative to the source system. You communicate it as a data freshness SLA: "This dashboard refreshes every 4 hours. The data you see reflects transactions up to 4 hours ago — it is not real-time." For operational decisions requiring real-time data, flag this explicitly: "If you need same-minute accuracy for this decision, use the live operational system rather than the data warehouse."