7 The CEO strikes back: Supercharging our dashboard

This chapter focuses on turning a launched Streamlit metrics dashboard into a more polished, practical, and executive-friendly product by responding to user feedback. The dashboard for Note n’ Nib already helps leaders explore sales metrics, but real-world use exposes gaps: charts become hard to read over long time spans, filters allow irrelevant combinations, metrics lack historical context, advanced users need deeper exploration, and dashboard states are difficult to share. The chapter frames these issues as a natural part of software iteration: launch is only the beginning, and quality improves through targeted refinements.

The first set of improvements makes the dashboard more flexible and informative. The time series chart is enhanced with selectable time grains—day, week, month, and year—so users can view short periods in detail or long periods more clearly. Filters are made interdependent by applying earlier selections before generating later filter options, preventing nonsensical choices such as selecting a product segment outside the chosen category. The dashboard also gains date range comparisons, including month-over-month, quarter-over-quarter, year-over-year, and previous-period comparisons. These comparison ranges are used to calculate percentage deltas in the metric bar, giving users context about whether current performance is better or worse than a prior period.

The chapter then adds more advanced capabilities for power users and collaboration. A drilldown view is implemented with a modal dialog, allowing users to inspect aggregated metrics across any combination of dimensions in a pivot-table-like format, complete with comparison tabs, total rows, formatted values, and styling. Deeplinks are added using query parameters and session state so users can share a URL that recreates the same dashboard selections, including dates, filters, chart settings, and other widget values. Finally, the app is made more production-ready by replacing the static CSV data source with Google BigQuery, using service account credentials stored as Streamlit secrets and updating the data loading layer while leaving the rest of the modular dashboard largely unchanged.

The day granularity in the existing time chart works for small time frames like a month (left), but not for longer time frames (right).

The line chart now has a time grain selector (see chapter_07/in_progress_01 in the GitHub repo for the full code at this point).

The filter bar options can have mismatching combinations.

Old filtering approach: Get unique values for all filters first, then filter based on selections.

New filtering approach: Get unique values for each filter, apply filter based on selection, then repeat for other filters.

The filter bar only shows valid option combinations (see chapter_07/in_progress_02 in the GitHub repo for the full code).

Sales for RoyalQuill were $1.32M, but there's no indication of whether that's good or what the sales were in a previous comparable period.

Sidebar showing a start/end date selector for the main date range as well as a "Compare to" input (see chapter_07/in_progress_03 in the GitHub repo for the full code).

Metric bar showing how each metric has changed from the comparison date range (see chapter_07/in_progress_04 in the GitHub repo for the full code).

A basic raw dataframe view rendered in a dialog using st.dialog (see chapter_07/in_progress_05 in the GitHub repo for the full code).

A mock UI for the drilldown view.

Drilldown view with a dimension selector and aggregated for selected dimensions (see chapter_07/in_progress_06 in the GitHub repo for the full code).

Drilldown dataframe view with a total row (see chapter_07/in_progress_07 in the GitHub repo for the full code).

Completed drilldown view with shaded total row and formatted values (see chapter_07/in_progress_08 in the GitHub repo for the full code).

Approach for implementing deeplinking.

We get an error when we inadvertently pass a string to an st.date_input (see chapter_07/in_progress_09 in the GitHub repo for the full code).

The widgets in the dashboard are populated based on the URL values (see chapter_07/in_progress_10 in the GitHub repo for the full code).

The most reliable way to find something on GCP is to use the search bar.

The Explorer pane in BigQuery.

The table creation screen in BigQuery.

• Launching an app is only the first step to making it successful. You also have to land it, ensuring that it meets your users' needs. For this, it is critical to hear from users directly.
• st.select_slider is a cross between st.selectbox and st.slider, used when you want to impose a logical order between options.
• st.metric can show the delta associated with a metric, i.e. how a value has changed over time.
• st.dialog is a decorator that lets you create a modal dialog, an overlay that blocks interaction with the rest of the app.
• You can use st.container to create placeholders in your app, only rendering the content you want to show when it's available.
• Pandas dataframes have a style property that can be used to set conditional rules that modify how they are displayed on the screen.
• st.query_params is a dictionary-like object that lets you read and update URL query parameters—this can be used to enable deeplinks in an app.
• A data warehouse is a specialized system designed to store and retrieve large amounts of data.
• Google BigQuery, part of GCP, is an example of a data warehouse. To enable an app to connect to it, you need to create a service account with a key, and record the credentials in st.secrets.

FAQ