The Agentic Intake Engine: Multi-Channel AI That Stops Legal Leads From Leaking

In today's legal landscape, client intake is the critical first touchpoint that determines case quality, client satisfaction, and conversion. But traditional intake processes are fragmented: clients call, email, text, and submit forms, while each channel often operates in isolation. That fragmentation creates intake leaks: slow responses, lost context, incomplete records, and missed opportunities.

We've built something different: an agentic intake engine powered by MayaGPT that operates across web portal, email, and SMS, while also connecting back to voice intake. The system is designed to keep legal leads moving from first response to qualified handoff without dropping context.

Three Platforms, One Intelligent Engine

Our multi-channel intake system is built on a simple but powerful principle: meet clients where they are, then preserve the context everywhere else. Instead of forcing clients into a single channel, we created independent platforms that share the same intelligent AI brain: MayaGPT.

1. Web Platform (client.saleda.ai)

The web portal provides a structured, guided intake experience perfect for clients who prefer a traditional form-like interface. Clients receive secure access links via email or SMS, then complete intake through an interactive chat interface powered by MayaGPT.

Flexible Link Delivery: Business owners can send intake links via email, SMS, or both—ensuring maximum reach
Secure Access Codes: Time-limited access codes (24-hour expiration) protect client privacy
Real-Time Chat: MayaGPT guides clients through intake questions conversationally
Document Upload: Clients can upload required documents directly through the portal

2. Email Platform

For clients who prefer asynchronous communication, the email platform enables complete intake via email threads. Clients simply send an email to intake@saleda.ai, and MayaGPT responds intelligently, maintaining conversation context across the entire thread.

Thread Management: Automatic email threading using In-Reply-To and References headers maintains conversation continuity
Client Identification: Automatic client identification from email address—no pre-existing session required
Rich Formatting: Email responses can include HTML formatting, links, and structured information
Asynchronous Flow: Clients can respond at their own pace, perfect for busy schedules

3. SMS Platform

Text messaging is the most accessible channel for many clients. Our SMS platform enables complete intake via SMS conversations, with MayaGPT adapting responses to SMS character limits while maintaining full functionality.

Conversation Threading: Automatic SMS conversation management via Twilio Conversation SID
Concise Responses: MayaGPT automatically adapts to SMS character limits (1600 chars) while maintaining clarity
Phone-Based Identification: Automatic client identification from phone number
Real-Time Interaction: Instant responses create a chat-like experience via text

The Power of Channel Independence

What makes our system unique is that each platform operates independently. A client doesn't need a pre-existing session to start intake via email or SMS—they can simply send a message, and the system:

Identifies the client from their email address or phone number
Determines the business owner from the client relationship
Creates a new intake session automatically
Begins the intelligent conversation with MayaGPT

While each channel can start independently, seamless channel switching (e.g., starting via email and continuing on web portal) is a planned enhancement. Currently, each channel maintains its own session, but all channels share the same unified data model and conversation history structure, making cross-channel continuity a natural next step in our roadmap.

MayaGPT: The Intelligent Brain Behind It All

All three platforms are powered by the same AI engine: MayaGPT. This agentic AI system provides:

Contextual Understanding: Maintains full conversation history across all channels
Intelligent Question Routing: Asks intake questions in the optimal order based on client responses
Answer Extraction: Automatically extracts structured answers from natural language responses
Channel Adaptation: Adjusts response style and length based on channel (email can be detailed, SMS is concise)
Completion Detection: Recognizes when intake is complete and triggers next steps automatically

Automated Follow-Up: Convert Failed Contact Attempts Into Intake Opportunities

Published follow-up workflows ensure failed calls do not become lost opportunities while keeping voice and messaging responsibilities separate.

How Automated Follow-Up Works

When a MayaCall (our AI voice agent) fails to reach a human—whether due to voicemail, no answer, or call busy—the system can automatically:

Detect Failure: Identify when a call didn't reach a human based on answered_by status
Evaluate the Published Workflow: Apply the current post-call conditions and channel policy
Start or Resume Intake: Use the same canonical intake definition and conversation context
Continue in Chat or Messaging: Route the client to the follow-up chat agent through an available channel
Track Engagement: Monitor when clients access and complete intake

Published Workflow Policy

Follow-up behavior is defined by the published workflow:

Platform Selection: Choose which platforms to enable (web, email, SMS, or all three)
Delivery Channels: For web platform, select how to deliver links (email, SMS, or both)
Outcome Conditions: Select which call outcomes enter the follow-up journey
Workflow Publication: Activate changes through the versioned workflow lifecycle

AutoRecall: Intelligent Call Retry System

Alongside automated messaging follow-up, AutoRecall schedules another call when an initial attempt does not reach a human.

Smart Retry Logic

AutoRecall uses configurable rules to determine when and how to retry calls:

Configurable Triggers: Retry on specific answered_by statuses (voicemail, no-answer, etc.)
Call Windows: Respect business hours and timezone preferences
No-Call Windows: Avoid calling during lunch hours or other specified times
Retry Limits: Maximum retry attempts per phone number (default: 3)
Intelligent Delays: Wait 24 hours (configurable) between retry attempts

Integration with Bland.ai

AutoRecall leverages Bland.ai's built-in scheduling infrastructure, ensuring reliable execution without requiring additional queue management or Lambda functions. This proven approach handles thousands of scheduled calls across multiple days with high reliability.

Security & Resilience: Production-Grade Architecture

Building a multi-channel system that handles sensitive client data requires enterprise-grade security and resilience. Our architecture includes:

Webhook Security

AWS SNS Signature Verification: Full certificate validation for email webhooks
Twilio Signature Verification: HMAC-SHA1 signature validation for SMS webhooks
Channel-Agnostic Security Service: Unified WebhookSecurityService handles all verification

Rate Limiting

Per-Channel Limits: Different limits for email (10/hr) vs SMS (20/hr) vs web (100/hr)
Per-Identifier Tracking: Rate limit by email address or phone number
Distributed System: Uses DynamoDB for distributed rate limiting across instances

Idempotency

Message Deduplication: Prevents duplicate processing of webhook messages
External ID Tracking: Uses email message_id and SMS message_sid for idempotency
Safe Retries: Webhook retries are completely safe—no duplicate processing

Error Handling & Resilience

Exponential Backoff Retry: Automatic retry with exponential backoff for transient failures
Dead Letter Queue (DLQ): Failed messages after max retries go to DLQ for manual review
Async Processing: Fast webhook responses (< 200ms) with async message queue processing
Circuit Breakers: Protects against cascading failures from external services

Unified State Management

Despite operating across three independent platforms, our system maintains unified state. All channels share:

Same Data Models: Reuses ClientIntakeSession and ClientIntakeChatMessage with channel metadata
Unified Conversation History: Full conversation context available across all channels
Shared Question Tracking: Knows which questions have been asked and answered, regardless of channel
Cross-Channel Continuity: Clients can switch channels mid-intake without losing progress

Real-World Impact

This multi-channel intake system delivers measurable value by reducing the most common intake leaks:

Higher Completion Rates: Clients complete intake through their preferred channel, reducing abandonment
Faster Response Times: Published follow-up workflows convert failed calls into intake opportunities
Cleaner Handoffs: Transcript, summary, qualification status, and next step stay attached to the lead record
Better Client Experience: Clients choose their preferred communication method
Reduced Manual Work: Automation handles routine intake tasks, freeing staff for high-value work
Improved Data Quality: Structured, consistent data collection across all channels

Technical Architecture Highlights

For the technically curious, here's how we built it:

Channel-Agnostic Services: Core services like ClientIntakeChatService work across all channels
Typed Channel Metadata: Pydantic models for type-safe channel-specific data (email thread IDs, SMS conversation SIDs, etc.)
Single Table Design: DynamoDB single-table design stores all channel data efficiently
Discriminated Unions: Type-safe platform configuration using Pydantic discriminated unions
Enum-Based Platform Selection: Type-safe IntakePlatform enum prevents invalid configurations

What's Next

This multi-channel intake engine is just the beginning. We're continuously improving:

Rich Media Support: Image and document uploads via email
Voice Messages: MMS support for voice messages via SMS
Multi-Language: Intake in multiple languages with automatic translation
Proactive Reminders: Intelligent reminders if intake stalls
Channel Preferences: Let clients set preferred communication channels

Conclusion

The future of client intake is multi-channel, intelligent, and measurable. By meeting clients where they are, whether that is voice, email, SMS, or web, and powering interactions with agentic AI, we are creating a system that reduces intake leaks and gives teams cleaner conversion data.

With the follow-up chat agent continuing intake after failed calls and AutoRecall intelligently retrying calls, the entire client acquisition funnel becomes more efficient, reliable, and measurable.

Ready to stop leads from leaking between channels? Contact us to see how the multi-channel intake engine can work for your firm.

The Agentic Intake Engine: Multi-Channel AI That Stops Legal Leads From Leaking