POC to production in eight phases.
The POC has already validated feasibility. The full product is a larger platform build because it connects voice AI with campaign strategy, dealer intelligence, scheme rules, order capture, Salesforce workflows, MCC/Oracle data, and auditable governance.
Build Phases
Architecture & Documentation Pack
Client-approved architecture brief, module definitions, integration contracts, governance requirements, and project plan.
Campaign Manager Data Model & UI
Database schema for campaigns, dealer audiences, call queue, retry rules, and suppression. Admin UI for campaign creation and monitoring.
Scheme & Product Catalog Ingestion
Scheme metadata model, eligibility rule engine, balance lookup, and product catalog with SKU/pack-size/order-unit structure.
Sales-Specific Agent Mode & Prompts
New SALES conversation mode, campaign-objective loading, scheme pitch logic, dynamic tonality controller, and objection-handling RAG integration.
Order Capture & Validation
Structured order extraction from speech, dealer confirmation loop, order validation engine against catalog and scheme constraints, and draft management.
Salesforce & MCC/Oracle Adapters
Integration adapters for order submission to Salesforce and master data sync from MCC/Oracle. Internal staging tables for replay and failure recovery.
Dashboards, Audit & Feedback Loop
Sales outcome dashboards, full audit store, campaign conversion analytics, order funnel reporting, and structured feedback workflow.
Future Enhancements
Salesforce case management, colloquial voice tuning, live agent transfer, and automated scheme-balance follow-up engine.
What the POC proves vs. what the product needs.
The POC demonstrates that voice AI can be connected to Berger dealer data and can use backend tools during conversations. The full product requires dedicated sales, campaign, scheme, order, and integration capabilities.
| Area | POC State | Product Need |
|---|---|---|
| Business Objective | Payment follow-up and dealer account queries. | Sales engagement, scheme pitching, and order capture. |
| Conversation Modes | FAQ and COLLECTION modes. | SALES mode with campaign objective, scheme pitch, objection handling, order taking. |
| Campaigns | No full campaign manager. | Audience selection, scheduling, prioritization, call attempts, outcome tracking. |
| Schemes | No dedicated scheme engine. | Scheme metadata, eligibility rules, balances, qualification criteria, recommended pitch. |
| Orders | No order capture workflow. | Structured draft, confirmation loop, validation, Salesforce submission. |
| Product Catalog | Not modeled for sales order taking. | Product/SKU/pack size/unit lookup and constraint validation. |
| RAG | FAQ-like support exists, sales knowledge not complete. | Scheme explanation, objection handling, policy-approved answers. |
| Tonality | Basic voice prompt behavior. | Standard, assertive, and persuasive tonality based on dealer behavior. |
| Languages | English, Hindi, Gujarati, Bengali covered in POC. | MVP needs 9 languages: + Marathi, Tamil, Telugu, Kannada, Malayalam. |
| Integrations | Internal DB and call providers. | Salesforce and MCC/Oracle integration contracts with full audit. |
| Governance | Message and call logging exists. | Full audit: transcripts, tool calls, order extraction, campaign decisions, integration events. |
Key Message for Client
The POC has already validated the feasibility of AI-assisted dealer conversations over voice. The full Berger Sales Agent is a larger platform build because it must connect voice AI with campaign strategy, dealer intelligence, scheme rules, order capture, Salesforce workflows, MCC/Oracle data, and auditable governance.