Customize a dashboard
Duplicate a default view or start from a blank board, then build tiles that group your data quality results the way your organization is actually structured.
The three default views answer questions most organizations share. A custom view answers the ones only you have: how quality looks per data domain, whether the contract rollout is landing in the schemas that matter, which class of check keeps failing, how much of your coverage comes from data standards rather than hand-written checks.
Custom views are built from the same tiles as the defaults. Nothing here is a separate feature or a different query language. You pick a measure, choose how to break it down, and pick a shape to draw it in.
During public beta, dashboards are not customizable via Soda AI.
Create and customize your own dashboard
You can create a full dashboard from scratch from the Dashboards page, or duplicate and edit an existing one.
Select + beside the view selector in the top right
Choose one of two starting points:
Duplicate Current View copies the board you are looking at, tiles and filters included. This is the faster route in most cases. Duplicating Executive overview and regrouping two of its tiles by your own attributes gets you a domain-level board in a couple of minutes.
Start Blank View gives you an empty board that you can fully customize and tailor to your needs.
Either way you land in edit mode:
Select Edit to change a view, and Done to leave. Nothing saves while you are in view mode, and nothing is hidden while you are in edit mode.
Edit mode adds four things:
+ Add tile in the header
+ Add filter on the filter row
A configure, duplicate, and delete control on each tile
A drag handle in the bottom right corner of each tile for resizing
Name your dashboard
Add filters to tailor the whole board to your needs
The filter row applies to every tile on the board at once. Data source and dataset filters are there by default, and + Add filter in edit mode adds more (e.g. Dataset attributes).
This is what makes one well-built view serve several use cases. A board grouped by check type, filtered to a single data domain, is that domain's board. Change the filter and it is another domain's board, with no tile rebuilt.
Click Add tile to start adding plots
Pick a visualization
Four shapes sit across the top of the tile editor: number, line, bar, and donut. The choice is not only cosmetic, because it changes which options the rest of the panel offers, and it changes what the tile measures.
Learn more about visualization types.
Pick a measure
Each measure carries a one-line description of what it counts, and counts are marked apart from percentages, so it is worth reading before you commit. Coverage and DQ Health Score are ratios and behave differently from the counts around them.
Learn more about metrics.
Choose how to group it
Group by is the breakdown that gives the tile its meaning. The options depend on the measure you selected, because not every breakdown makes sense for every measure. For example, datasets can be grouped by onboarding status, by whether a contract exists, and by whether monitors are attached. Scans can be grouped by scan status. Checks can be grouped by check type and by where the check came from.
Three categories appear for most measures:
Structure groups by data source, schema, or dataset
Attributes groups by the dataset and check attributes your organization maintains, such as data domain, data product, owner, or data quality dimension
Time plots the measure across the selected period
The attribute list is scoped to the measure. A dataset measure offers dataset attributes, a check measure offers check attributes, so the same board can carry one tile grouped by data domain and another grouped by check severity.
Select Show as % of total to convert the tile from absolute counts to shares. Useful when the absolute numbers are large enough to obscure the proportion, which is most of the time on a real estate.
Add a second breakdown
Then by splits each group again.
For example, grouping checks by schema and then by check outcome gives you passed, warned, and failed within every schema, rather than one bar per schema and no idea what is inside it.
The second breakdown draws from the same categories as the first, so you can cross structure with attributes: checks by data domain, then by data source, shows whether a domain's problems are concentrated in one system, for example.
Choose stacked or grouped
When a second breakdown applies, a Bars toggle appears with Stacked and Grouped.
Stacked puts the segments end to end in one bar, which keeps the total readable and shows composition. Grouped puts them side by side, which makes the individual values easier to compare and the total harder to see. Pick stacked when the question is "what is this made of" and grouped when it is "which of these is bigger".
Name the tile and add it
Write a title and a description. The description sits under the title on the board and is the right place for the question the tile answers, which is the pattern the default views follow: "Which sources need attention first?" rather than a restatement of the title.
Select Add tile. The preview on the left has been showing real data throughout, so what you saw is what lands on the board.
Configure, duplicate, or delete each tile from its top right corner
Drag or resize tiles from the bottom right corner
Tile components
Metrics
Every tile includes a plot that can follow one of ten metrics:
DQ Health Score
Share of checks passing
Leadership: is quality improving?
Coverage
Datasets under a check or a monitor
Leadership: is more data being monitored?
Checks
How many ran, and how they landed
Data teams: what ran?
Rows tested
How many rows those checks evaluated
Data teams: what ran?
Datasets
How many datasets Soda is watching
Data teams
Data sources
How many sources are connected
Data teams
Contracts
How many datasets have one
Leadership: how much data is behind an agreement?
Monitors
Metric monitors in place
Data teams: what is watching?
Anomalies
What those monitors flagged
Data teams: what flagged?
Scans
How many runs happened
Data teams
You can group any of them by data source, schema, dataset, check type, check source, outcome, monitor type, onboarding status, or by whether a dataset has a contract or a monitor. Or by any of your own dataset and check attributes: domain, data product, owner, data quality dimension, etc.
How each score is calculated, and why the dashboard score can differ from the one on a dataset's Checks tab: Soda Cloud dashboards
Visualization types
Once you've picked a metric, you can pick how it looks depending on what you need to visualize.
What to build
Some combinations that answer common data teams' questions:
Quality per domain or per data product. Group DQ Health Score by the attribute your organization uses to mark ownership. This is usually the first custom tile anyone builds, because it turns an estate-wide number into a set of numbers with names attached.
The onboarding backlog. Group datasets by onboarding status. Soda discovers far more than it checks, and the gap can be turned into work queue rather than a fault.
Contract adoption. Group datasets by contract and break down by schema, or plot contracts over time. This shows whether a rollout is reaching the schemas that matter or only the easy ones.
Which class of check fails. Group checks by check type, then by outcome, stacked. Freshness failing constantly is a different problem from duplicates failing constantly, and an estate-wide health score hides which one you have.
Where coverage comes from. Group checks by check source to separate data standards, contracts, metric monitoring, and the check library. This can tell you how much of your coverage is generated by policy and how much is maintained by hand.
Scan reliability. Group scans by scan status. A scan that fails to complete is an infrastructure problem wearing the costume of a data quality problem, and it can drag a health score down without failed checks in place.
Last updated
Was this helpful?