Calculated insights and data graphs

Calculated insights: revenue metrics on golden records

A calculated insight (CI) is an ANSI SQL aggregate saved as a __cio object. Its dimensions are what you group by: always the golden record's ID, plus any slices. Its measures are the aggregates.

Lifetime revenue per golden individual, following the join Salesforce's Trailhead query uses (sales order → link DMO → unified individual):

SELECT
  u.ssot__Id__c                                     AS unified_individual_id__c,
  SUM(o.ssot__GrandTotalAmount__c)                  AS lifetime_revenue__c,
  COUNT(o.ssot__Id__c)                              AS order_count__c,
  MAX(o.ssot__PurchaseOrderDate__c)                 AS last_order_date__c,
  SUM(o.ssot__GrandTotalAmount__c) / COUNT(o.ssot__Id__c) AS average_order_value__c
FROM ssot__SalesOrder__dlm o
JOIN IndividualIdentityLink__dlm l
  ON o.ssot__SoldToCustomerId__c = l.SourceRecordId__c
 AND o.KQ_SoldToCustomerId__c = l.KQ_SourceRecordId__c
LEFT OUTER JOIN UnifiedIndividual__dlm u ON l.UnifiedRecordId__c = u.ssot__Id__c
GROUP BY u.ssot__Id__c

• Key qualifiers. Salesforce's query also matches the KQ_ key-qualifier columns, which keep IDs from different sources apart. The exact link-DMO column names above are the hub's reading. Confirm them with GET /ssot/data-model-objects/{name}.
• Account revenue. Group ssot__BillToAccountId__c through the account link DMO to the unified account instead.
• Output columns end in __c. Name them as you want them to appear in prompts and in the search-index narrative.

API (under /services/data/v66.0/ssot):

• definitionType is CalculatedMetric (batch), StreamingMetric or ExternalMetric. Revenue rollups are CalculatedMetric.
• Create the insight before the search index and data graph that read it, and run it after each identity resolution run, so metrics follow the latest golden records.

Data graphs: a golden record with its orders, as one document

A data graph precomputes a root DMO (for example UnifiedIndividual__dlm) with related objects (link records, sales orders and products, insights) into a JSON record per root ID. That's the shape a prompt wants: this customer with their orders, in one lookup.

• Use a data graph for "tell me about this account": grounding by a known ID, in real time.
• Use a search index for "find accounts like this" or "who is ready to buy": semantic or hybrid search across many golden records (reference/data360/search-indexes-and-retrievers.md).
• The revenue pipeline uses both (patterns/golden-record-rag-pipeline-for-revenue.md).
• In an agent, a flow that queries Data 360 is the action Salesforce's workshop uses ("Get Merchant Order Insights"). Prompt Builder templates can ground on Data 360 too (family 6).

Failures