Salesforce architect Patrice Chalamon published a four-step framework on August 6, 2026, for eliminating identity debt before companies deploy Agentforce agents on Data 360. Identity debt arises when organizations accumulate disconnected customer profiles across systems, such as a premium customer appearing in Agentforce Service with zero purchases even after buying a high-value item in Agentforce Commerce under a different email alias. The article says reliable agentic AI requires a golden record, a trusted representation of each customer, and an “architecture of truth” that every application and AI agent can use as the authoritative identity source.

The article warns that large language models rely on precise context and can give incorrect, incomplete, or conflicting answers when the data model cannot determine whether “John Doe” the lead is the same person as “J. Doe” the contact on an escalated support case. Salesforce links poor identity resolution to the Reliable and Secure principles in the Salesforce Well-Architected Framework, saying identity mismatches can create unpredictable agent behavior and risk security or privacy breaches if an agent accesses the wrong profile’s data.

Step 1 calls for standardizing identity data ingestion from legacy sources before records land in data lake objects. Salesforce says out-of-order updates can overwrite current customer data with older records when, for example, a real-time address change is followed by a weekly ERP batch without timestamps. The recommended controls include sequencing, CDC watermarks, data profiling, MuleSoft or Data 360 connectors, E.164 phone-number formatting, ISO country codes, trimmed email whitespace, and extracting addresses, alternate phone numbers, or account notes from free-text legacy fields.

Step 2 sets rules for resolving core identities and survivorship. Salesforce says native Data 360 matching may be enough unless resolved identity records are consumed outside Salesforce, stewardship needs human-in-the-loop workflows, or matching rules are governed by an enterprise team outside Salesforce; if any answer is yes, the article recommends an enterprise Master Data Management platform like Informatica. Matching should move from deterministic exact matches using a trusted enterprise identifier to fuzzy and probabilistic matching, while compound rules such as “Name + Email” help avoid over-merging spouses who share one household email. Field survivorship can assign Billing authority over physical addresses and Marketing authority over phone numbers updated in the last 30 days.

Step 3 requires harmonizing unified profiles into Customer 360 data model objects so an AI agent can traverse the data graph instead of guessing relationships across raw tables. Salesforce says cleansed data lake objects should map into Customer 360 DMOs, with Order DMOs and Case DMOs linked directly to the Individual DMO. Historical clicks, cases, and orders must also be re-parented after duplicate leads and contacts merge, because attribute-level identity resolution without re-parenting leaves a clean profile but fractured history.

Step 4 activates resolved data for agentic AI by enforcing dynamic user-context boundaries at the activation layer, including field-level permissions and sharing rules. Salesforce recommends configuring Agentforce actions and prompt templates to query the harmonized data graph and Data 360 retrievers instead of raw CRM tables. The article also says companies should precompute calculated insights such as Lifetime Value, Churn Risk, and aggregated channel sentiment on unified DMOs rather than asking the LLM to calculate them at runtime.

Salesforce says maintaining the architecture requires ongoing data governance, not a one-time cleanup. The article recommends single-source authority for critical fields, downstream systems subscribing to the new source of truth, continuous data quality monitoring, consolidation-rate measurement, Data 360 flow triggers for match-rate alerts, and audit trails explaining why records merged or why one data point won over another. Companies are told to audit where identity resolution happens now and resolve identity once upstream before agents reconstruct customer identity independently.

Salesforce architect Patrice Chalamon published a four-step framework on August 6, 2026, for eliminating identity debt before companies deploy Agentforce agents on Data 360. Identity debt arises when organizations accumulate disconnected customer profiles across systems, such as a premium customer appearing in Agentforce Service with zero purchases even after buying a high-value item in Agentforce Commerce under a different email alias. The article says reliable agentic AI requires a golden record, a trusted representation of each customer, and an “architecture of truth” that every application and AI agent can use as the authoritative identity source.

The article warns that large language models rely on precise context and can give incorrect, incomplete, or conflicting answers when the data model cannot determine whether “John Doe” the lead is the same person as “J. Doe” the contact on an escalated support case. Salesforce links poor identity resolution to the Reliable and Secure principles in the Salesforce Well-Architected Framework, saying identity mismatches can create unpredictable agent behavior and risk security or privacy breaches if an agent accesses the wrong profile’s data.

Step 1 calls for standardizing identity data ingestion from legacy sources before records land in data lake objects. Salesforce says out-of-order updates can overwrite current customer data with older records when, for example, a real-time address change is followed by a weekly ERP batch without timestamps. The recommended controls include sequencing, CDC watermarks, data profiling, MuleSoft or Data 360 connectors, E.164 phone-number formatting, ISO country codes, trimmed email whitespace, and extracting addresses, alternate phone numbers, or account notes from free-text legacy fields.

Step 2 sets rules for resolving core identities and survivorship. Salesforce says native Data 360 matching may be enough unless resolved identity records are consumed outside Salesforce, stewardship needs human-in-the-loop workflows, or matching rules are governed by an enterprise team outside Salesforce; if any answer is yes, the article recommends an enterprise Master Data Management platform like Informatica. Matching should move from deterministic exact matches using a trusted enterprise identifier to fuzzy and probabilistic matching, while compound rules such as “Name + Email” help avoid over-merging spouses who share one household email. Field survivorship can assign Billing authority over physical addresses and Marketing authority over phone numbers updated in the last 30 days.

Step 3 requires harmonizing unified profiles into Customer 360 data model objects so an AI agent can traverse the data graph instead of guessing relationships across raw tables. Salesforce says cleansed data lake objects should map into Customer 360 DMOs, with Order DMOs and Case DMOs linked directly to the Individual DMO. Historical clicks, cases, and orders must also be re-parented after duplicate leads and contacts merge, because attribute-level identity resolution without re-parenting leaves a clean profile but fractured history.

Step 4 activates resolved data for agentic AI by enforcing dynamic user-context boundaries at the activation layer, including field-level permissions and sharing rules. Salesforce recommends configuring Agentforce actions and prompt templates to query the harmonized data graph and Data 360 retrievers instead of raw CRM tables. The article also says companies should precompute calculated insights such as Lifetime Value, Churn Risk, and aggregated channel sentiment on unified DMOs rather than asking the LLM to calculate them at runtime.

Salesforce says maintaining the architecture requires ongoing data governance, not a one-time cleanup. The article recommends single-source authority for critical fields, downstream systems subscribing to the new source of truth, continuous data quality monitoring, consolidation-rate measurement, Data 360 flow triggers for match-rate alerts, and audit trails explaining why records merged or why one data point won over another. Companies are told to audit where identity resolution happens now and resolve identity once upstream before agents reconstruct customer identity independently.