Book of Business Agent

Conversational AI Agent for Relationship Managers (RMs)

Why the Book of Business Agent is Required?

Relationship Managers manage complex books spanning Accounts, Financial Plans, Goals, Holdings, SIPs, Transactions, Tasks, Cases, and Meetings. Currently, they face:

• Manual Reporting: RMs need to construct custom Salesforce reports to answer operational questions, which is time-consuming and often skipped.
• Missed Signals: High-value signals such as clients drifting from target allocations, overdue reviews, or step-ups in SIPs go unnoticed without a real-time tool.
• Fragmented Workflow: Data scattered across multiple objects and reports increases the risk of errors and slows decision-making.
• Limited Drill-Down: Managers cannot easily explore data behind aggregate numbers to understand the cause or take action.

The Book of Business Agent enables RMs to ask natural-language questions and receive decision-ready, drillable answers in Salesforce, replacing the manual report workflow entirely.

• Fragmented Book-of-Business Data: RMs manage multiple Salesforce objects (Accounts, SIPs, Holdings, Plans, Tasks, Cases). Aggregating information manually is error-prone and slow.
• Manual Reporting Burden: Constructing reports for questions like “Which clients are overdue on PMS offerings?” consumes hours per week.
• Missed Opportunities: Without real-time insights, critical client actions — like step-ups or product recommendations — may be delayed.
• Complex Drill-Down Needs: RMs cannot instantly navigate from aggregate KPIs to underlying client-level details without multiple clicks.
• Security and Governance: Role hierarchy enforcement and privacy controls must be respected in all queries — a manual process is difficult to scale.
• No fixed catalogue of supported questions, so ad-hoc queries go unanswered.

The agent functions as a conversational assistant for RMs:

• Natural Language Questions: RMs ask any question about their book of clients in plain English — like “Which clients' SIP step-ups are overdue?” — and receive instant answers without building reports.
• Semantic Data Catalog: The agent maps questions to FSC data objects and fields, enabling “query anything” without a fixed question list.
• Intelligent Response Formatting: The agent chooses one of seven response formats (Direct Answer, Ranked List, Comparison, Breakdown, Insight Narrative, Detail/Drill-Down, Clarify) based on question type.
• Decision-Ready Outputs: Answers include AUM context, priority highlights, and actionable next steps.
• Drill-Down & Action: Users can drill into any client record from the answer and launch follow-up actions directly.
• Self-Correction: If a query is ambiguous or returns no results, the system provides clarification prompts or alternative closest matches.
• Memory & Context: The agent tracks conversation state, allowing follow-ups and refinements without re-querying cold.
• Role-Hierarchy Enforcement: Managers can scope queries to their book, team, or hierarchy, ensuring data visibility respects Salesforce OWD and role permissions.

The Book of Business Copilot is a conversational pipeline built on Agentforce and Salesforce Financial Services Cloud (FSC): a chat-based Lightning Web Component on the front end, an Apex orchestrator that runs a nine-step reasoning pipeline, Einstein prompt templates for language understanding and synthesis, and the governed FSC data layer underneath. The RM asks a question, and the agent interprets it, retrieves the right data within the RM's security scope, computes meaningful metrics, and presents the answer in the most appropriate format — with the ability to drill down and act, all within a single conversation. The solution builds entirely on existing FSC data; no new business objects are required.

Each capability is delivered through Salesforce technologies as follows:

• Query Anything — Semantic Data Catalog: Exposes the entire governed FSC data model — Account, FinancialAccount, FinancialPlan, FinancialGoal, Financial_Holding__c, SIP__c, transactions, Task, Event, and Case — to the Planner AI. Because the planner understands the whole model rather than a hardcoded list of fields, almost any question maps to real data, and extending coverage means adding to the catalog rather than rewriting logic.
• Proper Format Always: The Responder AI selects the best of seven response shapes for each question — answer, list, comparison, breakdown, insight, detail, or clarify — so the answer shape is chosen per question rather than fixed.
• Meaningful Data Always: A self-correction loop repairs invalid queries, relaxes filters on zero results, or escalates to a clarify prompt, so the agent never dead-ends. Fact-binding ensures every synthesized statement cites only records actually retrieved, preventing hallucination, and each answer adds the ‘why’ plus a next-best-action.
• AUM Context on Every Answer: Every result surfaces the Total Book AUM and the Query AUM (the combined AUM of the clients the answer covers), plus the query's share of the book. This figure is computed deterministically in Apex from the retrieved facts and the book scope, never by the LLM, so it is always exact.
• Stateful Conversations: Conversation memory (recent turns, the last result's Account IDs, and the last query plan) is held client-side, so refinements such as “now only the UHNI ones” or “why is Rajesh #1?” operate on the previous result instead of re-querying cold.
• Role-Hierarchy Security & Governance: Book-of-business scope is computed in Apex from the role hierarchy and injected into every query; the LLM can never widen it. Managers can scope to ‘My Book’, ‘My Team’, the ‘Entire Hierarchy’ below them, or a specific subordinate RM — strictly within their own downward sub-tree, with peers and superiors never selectable or visible. All data access is read-only and row-capped, with sharing rules and field-level security respected, and every turn is auditable via the stored plan, compiled SOQL, and response shape.
• Architecture Principle — AI Understands, Apex Governs: The Planner and Responder AI interpret and synthesize language, while Apex deterministically enforces book-of-business security, executes governed SOQL, and computes metrics. This pairing delivers personalized, consistent, and compliant answers at scale, with every response grounded in retrieved facts and scoped to what the RM is entitled to see.

Natural Language Query

• RMs type questions in plain English.
• The system interprets intent, entities, filters, and timeframe automatically.
• No pre-defined templates are needed.

Semantic Data Catalog

• Exposes a complete, business-described map of all queryable objects and fields.
• Ensures fact-bound responses without hardcoding questions.

Intelligent Response Formatting

• Chooses the best format per question: Direct Answer, Ranked Client List, Comparison, Breakdown, Insight, Detail/Drill-Down, Clarify.
• Provides maximum clarity and actionability.

Drill-Down & Actions

• Each response allows progressive drill-down to client 360° panel.
• Enables RMs to initiate follow-ups, meetings, or product actions from within the response.

Progressive Memory

• Conversation context is preserved; follow-up questions refine previous answers.
• RM can ask incremental questions such as “Now only the UHNI clients.”

Self-Correction & Fact-Binding

• Invalid or empty queries are automatically retried or clarified.
• LLM responses are strictly bound to retrieved facts; unsupported data is suppressed.

AUM Context

• Every answer shows Total Book AUM, Query AUM, and the query's share of the book.
• RMs instantly see how much of their book is impacted.

Role-Based Scope Selector

• Managers can view “My Book,” “My Team,” the “Entire Hierarchy,” or a specific RM.
• Selection is strictly downward-only for security; peers and superiors are never visible.

Compliance & Governance

• Read-only access; all SOQL queries are scoped.
• FLS, role hierarchy, and audit logging enforced.
• Every turn is auditable with stored query plans, retrieved records, and response shapes.

• Faster Decision-Making: Answers delivered instantly, replacing manual reports.
• Proactive Client Management: At-risk clients and pending tasks are surfaced automatically.
• Consistency & Accuracy: Standardized methodology ensures all RMs receive high-quality answers.
• Actionable Insights: Embedded tasks and follow-up actions improve responsiveness.
• Scalable Operations: Handles large books with hundreds of clients efficiently.
• Regulatory Compliance: Fully compliant with SEBI, DPDP, RBI, and FSC governance.

• Time Savings: Reduces hours of manual report preparation per RM.
• Enhanced RM Productivity: Focus on client engagement rather than data aggregation.
• Accurate Insights: Fact-bound responses from the Semantic Data Catalog ensure reliable metrics.
• Proactive Engagement: Highlights overdue tasks, SIP step-ups, product gaps, and at-risk clients.
• Scalable: Supports individual RMs or managers supervising entire teams.
• Drill-Down Actions: Direct access to client 360° panel enables immediate follow-up.
• Consistent Decision Support: Standardizes insights and metrics across all RMs.