Industry analyst Vernon Keenan cites Overlay AI¹ to Salesforce as something distinctively different to Agentforce, which he defines as an Embedded AI. In this post I will introduce areal world example of how at Being Guided we are applying an Overlay Ai Agent to Salesforce.

As explained below, we have created a Salesforce Data Analysis and Optisation Plan from our real world success in delivering Salesforce to the UK National Health Service (NHS), where we manage over 75,000 patient appointments each year in a private hospital.

Our Salesforce implementation for the NHS private hospital included integration with Electronic Health Record (EHR) via Fast Healthcare Interoperability Resources (FHIR). So, Data Analysis and Optimisation is key in such mission-critical environments.

The advantages with an Overlay AI versus Agentforce are: (a) prebuilt Tools with a wide variety of apps that work with Salesforce; (b) choice of Large Language Models (LLMs); and, (c) predictable, affordable business value pricing.

Today, Overlay AI Agents work with a growing list of popular, world-class technologies, beyond Salesforce and HubSpot CRM, including: Amazon Web Services (AWS); Atlassian Jira; Google Workspace; Intuit Quickbooks; LinkedIn; Microsoft Office 365; Postgres; Sage Intacct; ServiceNow; Snowflake, and many more.

Salesforce Overlay AI

As an Overlay AI technology for Salesforce partners and customers, this becomes a pragmatic solution: an AI Assistant for analysising and optimising Salesforce Production Orgs.

AI Assistants are built on AI Agents, which are semi-autonomous programs that follow simple English instructions. Each Agent is powered by a Large Language Model (LLM), which gives it the ability to understand language, make plans and execute tasks.
The AI Agent Platform consists of three elements: Large Language Model (LLM); Agent Instructions; and, Tools. Here's brief explanation of how they come together to deliver AI Assistants that extend a Salesforce Org environment:

Large Language Model (LLM)
The LLM is powering the Agent. You can choose amongst many different LLMs, including GPT4o-mini and Anthropic Claude.

Agent Instructions
The Agent Instructions describe the work that you would like the Agent to perform. If you want to devote your Agent to a specific task, then you will want to explain that task in the Instructions.

Tools
Agents perform work by giving them Tools. Templated Tools currently available are: Amazon S3; Atlassian Jira; Discord; DuckDB; GitHub; Google Calendar; Google Docs; Google Gmail; HubSpot; Hunter.io; Intuit Quickbooks; REST API; SMS; Salesforce; ServiceNow; Slack; and growing.

Salesforce Data Analysis and Optimisation Plan

The Plan set out below provides an example of step-by-step instructions for analysing and optimising a Salesforce Org using an AI Agent. As an integral part of our Licence and Services offer, we work with Salesforce partners and customers to identify and resolve data quality issues, remove redundancies, and reduce technical debt.

Prerequisites
Salesforce Org access.
AI Agent connection to Salesforce Org.
AI Agent connection to .csv Reporting Tool.
Salesforce Data Import Wizard access.
Appropriate Salesforce User Permissions.

Phase 1: Initial Assessment and Planning

Generate Current State Report
Field Usage Analysis Across All Objects.
Duplicate Record Count by Object.
Last Modified Date Analysis.
Empty Field Analysis.
Custom Field Inventory.

Review Data Quality Metrics
Duplicate percentage by Object.
Incomplete Record percentage.
Data Accuracy Score.
Field Utilisation Rates.

Create Data Optimising Plan
Business Impact.
Data Volume.
Complexity of Fix.
Dependencies.

Phase 2: Duplicate Management

Identify Duplicates
Generate Duplicate Analysis.
CopyAction: Run Duplicate Analysis.
Objects: Account, Contact, Lead, Opportunity, Custom Objects (Specify).
Matching Rules: Email, Phone, Account Name, Other (Specify).
Output: Report of Potential Duplicates.
Primary Record Identification.
Related Record Count.
Last Activity Dates.
Record Completeness.

Prepare Merge Strategy
Identify Master Records (usually Most Complete/Recent).
List related Records to Merge.
Note any Special Handling Requirements.

Execute Merge Process
Create Backup of Affected Records.
Use Data Import Wizard to process Merges.
Upload prepared .CSV file.
Map Fields correctly.
Run in batches in say, 5,000 Records.
Validate Results.

Phase 3: Field Cleanup

Field Usage Analysis
Generate Field Usage report via Supercog.
Last Used Date.
Population Percentage.
Reference in Processes/Flows.
Report Usage.

Field Categorisation
Tag Fields.
Business Critical.
Regulatory Required.
Nice-to-Have.
Deprecated.
Unknown.

Field Consolidation
Define Deprecated Fields.
Document Current Usage.
Plan Data Migration.
Schedule Deletion.
Update any Dependent Processes.

Phase 4: Data Standardisation

Identify Standardisation Needs
Phone Number Formats.
Address Structures.
Company Name Variations.
Industry Classifications.

Create Standardisation Rules
Define Standard Formats.
Phone Numbers.
Addresses.
Company (Account or Organisation) Names.
Job Titles.
Industry Values.

Apply Standards
Generate Standardisation.
Current Values.
Future Values.
Standardised Values.
Record IDs.
Use Data Import Wizard to Update Records.
Validate Changes.

Phase 5: Technical Debt Reduction

Process Automation Review
Create AI Agent Inventory.
Workflow Rules.
Process Builders (If not Retired).
Flow Definitions.
Validation Rules
Ongoing AI Agent Instructions and Prompts.

Consolidation Planning
Identify Key Elements.
Redundant Processes.
Unused Automations.
Performance Impacts.
Consolidation Initiatives.

Implementation
Create Process Documentation.
Build Consolidated Solutions.
Test Thoroughly.
Deploy Changes.
Monitor for Impacts.

Phase 6: Maintenance and Monitoring

Establish Ongoing Monitoring
Configure Alerts.
New Duplicates.
Data Quality Scores.
Field Usage Changes.
Process Performance.

Regular Maintenance Tasks
Weekly and Monthly Reports.
Review Data Quality Scores.
Process New Duplicates.
Validate Standardisation Rules.
Field Usage Analysis.
Process Performance Review.
Technical Debt Assessment.

Applying Best Practices and Tips

For Business Users
Always export Data before making mass changes.
Use consistent Naming Conventions.
Document business reasons for changes.
Test changes with small batches first.
Communicate changes to affected teams.

For Administrators
Maintain backup of Deleted Fields/Processes.
Use Salesforce Sandbox for Testing.
Document Technical Dependencies.
Monitor System Performance during updates.
Build-in Validation Rules for new Standards.

Troubleshooting Common Issues

Data Import Errors
Check Field Permissions.
Verify Data Types match.
Confirm Record Ownership.
Review Required Fields.
Validate External IDs.

Merge Conflicts
Review related Records.
Check Sharing Settings.
Verify Field-level Security.
Validate Workflow Triggers.
Check Validation Rules.

Change Management
Document all changes.
Communicate with all stakeholders.
Provide training as needed.
Monitor User adoption.
Gather feedback.
Adjust processes based on feedback

Summary

Salesforce Data Analysis and Optimisation is an ongoing process, not a one-time project. Regular maintenance and monitoring are essential for long-term success.

As illustrated at the top of this post, future AI Agents include Tools to bring together Salesforce and other apps in automating everyday processes and tasks. This includes popular technologies, such as Google Workspace (G-Drive, Gmail, Docs, Sheets); Microsoft Office 365 (Outlook); Slack; and many more.

Reference

1. Keenan, V. (2024) SalesforceDevOps. https://salesforcedevops.net/