Data 360, Calculated Insights & Metric Governance
A Data 360 calculated insight is a data product—not a formula
The metric becomes operational truth
A team creates a customer-value calculation for a dashboard. Six months later, the same value qualifies loyalty tiers, prioritises service queues and personalises offers. Finance changes the definition of net revenue, but nobody knows which consumers depend on the original logic. The SQL still runs successfully while different business processes silently make decisions from different meanings of value.
The JSBC Labs position is that a Data 360 calculated insight is a production data product—not a dashboard formula. Salesforce makes calculated measures available to segmentation, activation and analysis. That reuse is powerful, but it turns a calculation into a shared contract. The owner must govern meaning, grain, freshness, lineage, quality and change impact before downstream teams treat the result as fact.
Define grain before writing SQL
Start with the business question and the entity represented by one output row. Is lifetime value calculated per individual, unified profile, household, account, product or channel? Salesforce defines dimensions as fields that group measures and measures as aggregated quantitative values. Selecting dimensions therefore sets the grain; adding one can multiply rows and change the interpretation of every measure.
Write a metric contract that names the grain, source data model objects, population, filters, time window, currency treatment, null policy and aggregation rules. Include examples for edge cases such as refunds, merged profiles, reopened orders and missing identifiers. A syntactically valid query can still produce a semantically invalid metric. Data Lens previews shape and sampled results, but the team must decide whether that shape matches the business meaning.
Choose latency as part of the contract
Salesforce distinguishes calculated, streaming and real-time insights. Batch calculated insights support complex, high-volume metrics over historical data; streaming insights aggregate event data in time windows; real-time insights operate through a real-time data graph. These are different consistency and cost models, not faster and slower settings for the same design.
Choose the slowest mode that still supports the business decision. A monthly value tier rarely needs millisecond recomputation, while an in-session engagement signal may. State the promised freshness in business terms: for example, orders posted by midnight appear in the value score by 07:00. Avoid vague labels such as real time. Consumers need to know when a result is complete enough to act upon and when it is still catching up.
Scheduling is part of metric semantics
Data 360 calculated insights can run on schedules such as every 1, 6, 12 or 24 hours. Salesforce also notes that if one run has not completed before the next scheduled run, the next run is skipped. A six-hour schedule therefore does not guarantee a value no more than six hours old. Runtime, source arrival and failed or skipped executions all affect observed freshness.
Align the schedule with upstream ingestion and downstream activation. Running before source data is complete creates a precisely calculated incomplete answer; running after a segment publishes can leave yesterday's population active for another cycle. Record the dependency chain, expected completion time and maximum tolerated age. Monitor the last successful run and data watermark, not merely whether a schedule is enabled.
Validate business outcomes, not only syntax
SQL Insights Builder can check syntax and preview sampled results, while Salesforce instructs teams to validate an insight in Data Explorer before using it in a segment. Treat those as starting points. Build reconciliation datasets with known customers and hand-calculated expected measures. Test zero activity, duplicated events, late-arriving data, negative values, profile merges, currency differences and records outside the intended window.
Validate at three levels: row-level examples prove local logic, population-level totals expose missing or duplicated groups, and consumer tests confirm segments and activations interpret the output correctly. Compare results with an independent source where possible. A metric that reconciles overall revenue can still assign the wrong value to individual profiles, and a correct insight can still be misused by an activation filter.
Design dependency and performance boundaries
Calculated insights can be defined from data model objects and from other calculated insight objects, which enables reuse but also creates dependency chains. Salesforce's architecture guidance says computed aggregates are written as calculated insight objects and orchestrated as jobs. Deep nesting can make lineage, runtime and failure impact difficult to understand even when every individual query appears reasonable.
Keep the dependency graph shallow and intentional. Reuse a shared metric only when its meaning and grain are genuinely stable across consumers. Review data volume, join cardinality, filters and time windows before increasing frequency. Salesforce publishes processing limits and notes that long-running insights may be terminated. Optimise the data shape and logic rather than turning repeated manual runs into an operating procedure.
Version the meaning, not just the query
Editing logic in place can change every segment, activation and API consumer that reads the calculated insight. The Calculated Insights API exposes metadata, dimensions, measures, filters and time aggregation to applications, so field names and types form an interface as surely as an Apex method signature does. A renamed measure or altered dimension can be a breaking change even when the insight activates successfully.
Store the SQL or builder specification, metric contract and test cases in version control. For a breaking definition, publish a new insight with an explicit version, migrate consumers and compare both results through an agreed observation period. Record the effective date and decision owner. Retire the old version only after usage is known and downstream teams have confirmed their transition.
Operate calculated insights as owned products
Assign a business owner for meaning and a technical owner for execution. Maintain a catalogue containing grain, description, source lineage, schedule, consumers, validation evidence, service expectation and support route. Alert on failed, skipped or unusually long runs, stale watermarks and material shifts in row counts or distributions. The list-view status is useful operational evidence, but it needs an accountable response process.
Review important metrics on a cadence and whenever source mappings, identity rules or business definitions change. Measure freshness compliance, reconciliation variance, consumer adoption and incident recovery—not the number of insights created. A calculated insight earns trust when decision-makers can explain what it means, developers can trace how it was produced and operators can prove when it last became safe to use.
Official references
- Salesforce Help — Create a Calculated Insight with SQL
- Salesforce Help — Use SQL to Create Insights
- Salesforce Help — Managing Insights
- Salesforce Help — Schedule a Calculated Insight in Data 360
- Salesforce Help — Enhance Data with Insights
- Salesforce Help — Data 360 Limits and Guidelines
- Salesforce Developers — Query Calculated Insights
- Salesforce Architects — Data 360 Architecture