Calculated Insights are SQL-defined aggregation queries that compute business metrics like LTV, churn score and purchase frequency against DMO data and store the results directly on Unified Customer Profiles. They use pre-computation rather than real-time querying because the SQL aggregations are expensive — computing LTV across 500 million order records for 10 million customers would take hours if run on demand. Pre-computing on a nightly schedule means the computation runs once, the results are stored on profiles and every subsequent segment evaluation, API call and Agentforce query retrieves the pre-computed answer in milliseconds. This follow the compute once, retrieve many principle that makes Data Cloud viable at enterprise scale. Real-time on-demand recomputation of these metrics would make every segment refresh impossibly slow and consume enormous Data Credits repeatedly.

A Dimension is a grouping field that appears in the GROUP BY clause — it defines who or what the metric belongs to. The Individual ID is always a dimension because it tells Data Cloud which customer each row of computed data belongs to. A product category can also be a dimension if you are computing category-level insights. A Measure is an aggregated computed value — the actual number that represents the metric. SUM of order amounts is a measure. COUNT of orders is a measure. AVG order value is a measure. This distinction matters because Data Cloud uses it to correctly link computed metrics to Unified Profiles — dimensions are the keys that establish the link and measures are the values stored on the profile. Incorrectly classifying a dimension as a measure or vice versa breaks the profile linkage and makes the insight unusable in segment filters.

The query selects from the Sales Order DMO using its Data Cloud API name. The Individual ID field is selected as the dimension — this links each computed row to the right Unified Customer Profile. SUM of the total amount field gives lifetime value. COUNT of order IDs gives total order count. AVG of total amount gives average order value. MAX of order date gives the last purchase date using DATEDIFF to compute days since last purchase. A CASE WHEN statement classifies each customer into Premium, Standard or Value tiers based on their computed LTV. WHERE clause excludes cancelled orders to avoid inflating LTV. GROUP BY Individual ID ensures one output row per customer. The dimensions are Individual ID. The measures are lifetime value, total orders, average order value, days since last purchase and value tier. The insight would be scheduled to run nightly after the Sales Order DMO refresh completes so LTV always reflects orders up to the previous day.

A sudden performance degradation from DMO growth almost always means the query is doing a full table scan with no date partitioning. My optimization approach follows a priority sequence. First I add a WHERE clause date filter — for email engagement, the last 90 days is almost always sufficient for business purposes and eliminates 80 to 95 percent of historical rows from the computation. Second I review the SELECT list to ensure I am not using SELECT star — only selecting the columns actually needed. Third I check for any COUNT DISTINCT operations which are computationally expensive — replacing with COUNT where uniqueness is not critical. Fourth I check if the insight is joining multiple large DMOs and consider pre-aggregating the higher volume DMO in a subquery before the join. Fifth I reschedule to run at 2 AM to avoid contention with other workloads. Finally I check whether this insight can be split into two simpler chained insights — a base LTV insight and a tier classification insight — which is easier to optimize than one complex monolithic query.

Calculated Insights power Agentforce by providing pre-computed metrics on the Unified Customer Profile that the AI agent accesses via the Data Graph before generating any response. Without Calculated Insights the agent would need to run complex aggregation queries in real-time during a conversation — which is too slow. With pre-computed insights the agent retrieves the results in milliseconds. A specific example is a customer service Agentforce agent for an insurance company. When a customer contacts support the agent queries the Data Graph which includes a Customer Value Score Calculated Insight that combines years as customer, annual premium paid and claims history. If the customer's value score is in the top 20 percent the agent's guardrails activate a priority response mode — offering to waive a processing fee or expedite a claim review. The agent says “As a valued long-term customer I can expedite your claim review to 48 hours instead of the standard 5 business days.” This personalization is only possible because the value score was pre-computed by a Calculated Insight and is instantly available on the profile — not computed during the live conversation.