Understand Salesforce Capabilities for Reliable Data Readiness

Learning Objectives

After completing this unit, you’ll be able to:

Explain how data readiness insights guide architecture decisions for AI agents.
Identify Salesforce platform components that help improve data reliability and AI safety.
Map reliability risks to architecture capabilities using the Salesforce Customer Success Data Quality Framework

From Data Profiling to Architecture Decisions

In the previous units, Luna evaluated whether NTO’s data was ready to support an AI-powered Case Deflection Agent. Discovery interviews and data profiling revealed several risks, including:

Incomplete or inconsistent data
Disconnected or duplicate records
Unreliable fields
Metadata gaps

These findings do more than highlight problems. They guide architecture decisions that reduce hallucination risk and prevent incorrect responses.

Salesforce architects and data architects can use the Salesforce Data Quality Management framework to translate discovery insights and profiling results into architecture decisions.

Luna’s data profiling and assessments allow her to continue the steps of the data quality framework and implement important governance and security measures to ensure NTO’s data is ready and reliable for AI.

Luna partners with Charlotte to determine which platform capabilities, additional products, or AgentExchange solutions are needed.

1. Clean and Standardize Data for Consistency

Even when data is standardized within one system, formats and standards vary across sources. Without governed standardization, inconsistent values or interpretations can lead to incorrect responses.

Architectural Response	Salesforce Capabilities
Normalize values across systems. Remove invalid or placeholder data.	Data 360 Data Transforms to standardize values and to filter out “bad” values (for example, outliers identified by data profiling).

2. Enrich Data to Ensure Data Completeness

Sometimes the agent might not have access to critical information required for the use case because it exists in another system. For example:

Service Cloud has an email-marketing consent field on the Contact object, but that field is maintained in Marketing Cloud. Working with CRM records would not give access to the critical information in Marketing Cloud.
Customer address changes might be updated in one system but not others, causing inconsistencies for profile unification.

Architectural Response	Salesforce Capabilities
Ensure key information across data sources is brought together within the unified customer profile. Reconcile discrepancies. Enrich missing or outdated data when internal sources are incomplete (for example, current address determination).	MuleSoft or Data 360 to synchronize attributes across systems. Third‑party enrichment providers, such as National Change of Address (NCOA), are often discoverable through AgentExchange.

3. Unify Profiles to Ensure Agents Have the Right Customer Context

Fragmented customer records, such as duplicates or disconnected transactions, can prevent agents from understanding the full customer context. Profiling can reveal duplicate contacts, accounts, or transactions that should be consolidated into a single customer profile.

Architectural Response	Salesforce Capabilities
Deduplicate (match and merge) unintentional duplicates when needed for CRM end users. Create unified customer profiles across disconnected source records. Ensure key transactions relate to the unified customer profile.	Salesforce or AgentExchange duplicate management solutions. Data 360 Identity Resolution to create unified customer profiles. Data 360 Data Spaces, if only a subset of records is permissible or contextually significant for the unified profile.

4. Ensure Metadata Is Sufficient to Govern Access

Reliable AI responses depend not only on the data itself, but also on the metadata that explains how that data should be interpreted and who should be allowed to access it. Metadata provides essential context, including field definitions, ownership, sensitivity classifications, and usage guidance. Without this context, agents can expose restricted information, misinterpret a field’s meaning, or retrieve data that should not be used in a particular interaction.

Architectural Response	Salesforce Capabilities
Document field description, purpose, ownership, and acceptable values so agents interpret data correctly. Classify sensitive fields and ensure access controls prevent agents from exposing restricted information. Maintain consistent definitions across systems so AI agents interpret the same data concepts consistently.	CRM Object Manager Data Dictionary or Data 360 Semantic Layer to maintain descriptions within Salesforce. Informatica Data Catalog to discover, document, and govern data assets across systems. Data 360 Data Governance to assess and automatically classify sensitivity.

5. Govern Which Data Sources Agents Use for What Purpose

Ensuring data readiness within a system of reference, or system of context, is essential for agents to access fit-for-purpose data. However, unless prompt engineers ensure agents use the right data sources, the job is incomplete.

For example, case deflection agents need access to the CRM Contact object, but the complete order and case history are retrieved by using the customer profile associated with the contact and the case in Data 360.

Architectural Response	Salesforce Capabilities
Define data integration architecture to unify data within and across data sources. Determine which data source an agent should use for a given prompt.	Salesforce flows to retrieve unified context and orchestrate actions. Agentforce prompts to determine which data source(s) agents use. MuleSoft or Data 360, to bring data together across multiple data sources.

6. Limit Field-Level Access to Prevent Agentic Misinterpretation

Data profiling reveals fields that are unused, inconsistently populated, or filled with default values. If an agent reasons with these fields, it can produce incorrect answers or increase the risk of hallucinations.

Architectural Response	Salesforce Capabilities
Restrict access to unreliable fields. Limit the agent's reasoning scope to trusted data.	Permission sets and field- and row-level security to control which fields and records the agent can access. Prompt guardrails or agent configuration to restrict inputs that the agent can use.

7. Monitor Data and Metadata for Continuous Data Reliability

Data reliability is not static. New integrations, process changes, and evolving data can introduce new inconsistencies or unexpected data patterns. Ongoing monitoring detects emerging risks before they affect agent behavior.

Architectural Response	Salesforce Capabilities
Continuous monitoring of data and metadata trends. Dashboards to track data reliability trends over time. Alerts to notify stakeholders of unexpected changes.	Salesforce reports and dashboards. AgentExchange data profiling and monitoring tools. Tableau Next for analytics and visualization of data reliability ends. Flow and Slack alerts for issue notification. Agentforce to evaluate deviations and propose next best actions.

Bring It All Together

Using the insights from their discovery process and the results from their data assessments, Luna and Charlotte prepare a solution architecture proposal that they are confident will:

Provide reliable data for the Case Deflection Agent.
Reduce hallucinations and incorrect responses.
Maintain compliance and customer trust.
Support ongoing reliability through monitoring and governance.

Stay tuned for the Data Readiness Architecture and Data Assurance badges, which continue Luna’s, Charlotte’s—and your—journey to plan and implement the Case Deflection Agent.

Time Estimate

Topics

Looking for Help?

Agentforce 360 Platform Resources