For exploratory root-cause analysis, an interactive dashboard is essential. It allows the user to ask ad-hoc questions, filter by region and time, drill down from quarterly totals into monthly or weekly data, and see how different widgets (like a map and a bar chart) interact. This self-service model empowers the user to investigate the ‘why’ behind the numbers, which is the core goal described in the scenario.

For a classification model (like churn prediction) where the cost of a false positive (incorrectly labeling a valuable client as a churn risk) is a primary concern, Precision and Recall are critical. Precision tells you what percentage of accounts predicted to churn actually will, ensuring actions are targeted correctly. Recall tells you what percentage of all actual churners the model catches. Balancing these two metrics assesses the model’s reliability for the specific business action of targeted outreach.

An inner join returns only records where there is a match in *both* tables. If you inner-join Opportunities to Customer Feedback, the resulting dataset will contain only opportunities that have a corresponding feedback record. This excludes any opportunities without feedback, skewing metrics like Average Deal Size. The correct approach would often be a left outer join from Opportunities to Feedback to preserve all opportunities, with feedback data populated where it exists.

Tableau CRM supports user-specific, private dashboard filters. If a previous user (perhaps from national HQ) applied a filter and saved the dashboard with it as their private view, the next user (the regional VP) would inherit that filtered view. The VP’s data access is intact, but her view is constrained by the saved private filter. Instructing her to reset the filter or checking/removing any saved private filters would resolve this.

To analyze what *drove* changes in a metric like pipeline, you need to compare the state of records (opportunities) at two different points in time. A historical tracking (snapshot) dataset, scheduled to run daily or weekly, captures the state of key fields (Stage, Amount) for each record at each run. By comparing snapshots, you can categorize changes as new, amount changed, stage changed (to closed-lost), etc., enabling true pipeline movement analysis.

Einstein Discovery surfaces correlations, not causations. A counter-intuitive finding like this is a red flag to dig deeper. It’s crucial to investigate for confounding variables (e.g., perhaps senior managers are assigned the most complex, high-risk projects) or non-linear relationships (e.g., medium tenure might be lowest risk, with both low *and* high tenure being higher risk). Using lenses to explore the raw relationship between tenure and risk is the necessary next step to validate or explain the model’s discovery.

The most impactful performance optimization is to reduce the volume of data processed. Applying a filter at the dataset level (in the recipe) means the lens queries a much smaller, pre-filtered dataset (e.g., 500k records instead of 5 million). This improves performance for all lenses built on that dataset. Dashboard or lens-level filters are applied *after* the full dataset is queried, so they don’t reduce the initial data processing load as effectively.

For two widgets to interact via selection (clicking on one filters the other), they must be built from the same dataset. This shared data foundation allows the dashboard’s interaction engine to pass the selected dimension value (e.g., ‘Equipment’ from the bar chart) and automatically filter the data context for the other widget (the table). If the widgets are from different datasets, this automatic cross-filtering is not possible without complex workarounds.

Einstein Discovery provides local, record-level explanations. The ‘Prediction Details’ (or similar named tab) within the embedded component breaks down the factors that contributed to *this specific opportunity’s* predicted score. It will list factors that pushed the score down (e.g., ‘Deal Size is very high’) and factors that pushed it up (e.g., ‘Opportunity is in the Negotiation stage’). This is the actionable insight the rep needs to understand and potentially counteract the model’s reasoning.

For exploratory root-cause analysis, an interactive dashboard is essential. It allows the user to ask ad-hoc questions, filter by region and time, drill down from quarterly totals into monthly or weekly data, and see how different widgets (like a map and a bar chart) interact. This self-service model empowers the user to investigate the ‘why’ behind the numbers, which is the core goal described in the scenario.0