RAIN OS: Setting Up Agents, Outreach Lists & Comms
- Jun 8
- 22 min read
Jump to a section
Overview
What Is the RAIN OS?
RAIN OS is the AI agent platform built natively into 8020CRM. It is a Salesforce-native solution - not a third-party integration that bolts on via Zapier or webhooks. This means agents have direct, real-time access to every standard and custom field, every related lookup, every activity record, and every contact in your org without any external authentication or data bridge.
RAIN OS supports three types of agents:
Voice Agents: AI agents that make and receive phone calls, handle objections, qualify leads, book appointments, and update Salesforce records during or after the call.
Chat Agents: AI agents that handle inbound SMS and web chat conversations with the same logic as voice agents.
Assistant Agents: Background AI that monitors records, fires next best actions, runs proactive outreach campaigns, and keeps the CRM in self-drive mode without making calls.
Each agent type can be configured separately or combined. A single agent can handle inbound/outbound voice calls and chat using the same conversation flow and prompt logic.
Before You Begin
What You Need
System Admin access to the Salesforce org where RAIN OS is installed.
The RAIN OS package must be installed. If it is not in your org, contact your 8020CRM account team.
For voice agents: at least one phone number purchased or ported through the platform (Twilio or Telnyx). Phone numbers cost $5/month each.
For RAIN Intelligence (AI data queries): at least one API key configured - either org-provided RAIN credits or your own API keys (BYOK) from OpenAI, Anthropic, Google, or other supported providers.
Voice agents run on a separate, static LLM model. Your BYOK keys for RAIN Intelligence are not used for voice agent calls - these are billed separately per minute.
Why It Matters
No other solution in the REI CRM space has native Salesforce integration at this level. Competing voice agent tools - third-party solutions that connect via webhook or Zapier - require you to pass a limited, pre-defined set of fields to an external system. They cannot access your custom fields, your related records, or your real-time Salesforce data. They break when APIs change and credentials expire.
RAIN OS agents pull data directly from your org's tables in real time. Every custom field, every related lookup, every activity log is available to the agent without any setup beyond selecting which fields to pass in. The closest comparable native option is Salesforce Agentforce.
KEY DIFFERENTIATOR: RAIN OS voice agents can call your own Salesforce Flows and Apex actions during a live call. This means an agent can run your underwriting calculator, send a DocuSign contract, post a Slack notification to your team, and update the Opportunity record - all in one phone conversation, without any third-party automation.
Part 1: Initial Setup
When RAIN OS is installed, new users land in the setup wizard. The wizard hides most tabs until setup is complete to prevent confusion. Work through the steps in order.
Step 1: Add the RAIN OS Permission Set
Before any user can access RAIN OS features, the RAIN OS permission set must be assigned to them in Salesforce Setup.
Go to Setup > Permission Sets.
Locate the RAIN OS permission set (look for 'RAIN' or the package name).
Click Manage Assignments > Add Assignment.
Select the users who need access and save.
ADMIN ONLY: Only System Admins should build and manage agent configurations. Regular users (Lead Managers, Acquisition/Dispo Managers, TCs) will interact with the agents and the RAIN Hub, but do not need access to the agent builder.
Step 2: Terms, Conditions & Business Profile (TCPA Compliance)
The first screen in the setup wizard is terms and conditions acknowledgment. This is not optional - it is required for TCPA compliance and for vendor registration.
Open the RAIN OS app from the App Launcher.
The setup wizard will appear. Read and acknowledge the terms. By accepting, you confirm that you understand this is an AI voice agent (bot) and that you will adhere to quiet hours and ethical standards in its use.
Enter your business title and business type. This creates your first Business Profile, which is required for future ATP (10DLC) registration for SMS.
Click I Agree - this functions as an electronic signature.
What happens next | Details |
Workspace creation | Your RAIN OS workspace and API keys are created automatically in the back end when you sign. |
One manual step currently | A manual toggle must be activated by the 8020CRM team after your compliance is signed. This is the one step that is not yet self-service. Expect a short activation window. |
ATP registration | Your business profile is used later for 10DLC/ATP campaign registration for SMS sending. ATP approval typically takes 2 to 3 weeks - same timeline as all other providers. |
Step 3: API Webhook Setup
This step connects your Salesforce org to the RAIN OS infrastructure. Select the Salesforce site you want the webhook to run through. The site is built for you automatically - you do not need to configure it. Simply select it from the dropdown and save.
UNIQUE TO RAIN OS: The automatic site creation is a differentiator. Other solutions require you to manually configure webhook endpoints. Here it is built for you.
Step 4: Features & Settings
After webhook setup, configure the core feature settings. These control org-wide behavior for all agents.
Setting | What It Does |
Co-Pilot (RAIN Intelligence) | Enable or disable the RAIN Intelligence AI chat interface for all users. Can also be disabled for individual users. |
Insights & Suggestions | When enabled, every inbound interaction (email, SMS, call, web form, record update) is analyzed for field suggestions and next best actions. These are queued in the RAIN Hub inbox. |
Field Suggestion Confidence Threshold | Default is 90%. Any field update suggested at or above this confidence level is auto-applied to the record. Below this threshold, it queues for manual approval. Admins can adjust per action type. |
Execution Wait Times | Controls how long RAIN waits before executing a suggested action. For email responses, you may want a 5-minute wait so a live agent can respond first. For SMS replies, you may want zero delay. Configurable per action type. |
Inbox Notifications | Aggregates all new inbound activity (email, SMS, call, web form) into the RAIN Hub inbox with an audible alert. Eliminates the need to refresh list views to see new leads. |
SMS Assist | When enabled, templates used in drip campaigns are automatically personalized by AI before sending. The template content is preserved but tuned for the individual recipient based on their record and conversation history. |
Max SMS per Hour | Rate limiting for high-volume SMS sends. Set this to match your carrier's limits. RAIN OS is carrier-agnostic - it works with smrtPhone, Aircall, JustCall, Five9, DialPad, RingCentral, and others. |
Object Availability & Routing | Define which custom objects agents can act on (beyond the standard set). Also configure the waterfall hierarchy for inbound call routing - the order in which objects are searched when a phone number is matched (e.g., Opportunity first, then Contact, then Lead). |
Org-Wide Email Address | Set the email address the system uses for sending appointment reminders and notifications. |
Inbox Field Mapping | Customize which fields appear in the quick-view of the inbox record (score, status, speed-to-lead, lead source, etc.). |
Step 5: Credits, Models & BYOK
Configure how RAIN Intelligence accesses AI models. This is separate from voice agent billing.
Option | How It Works |
Org-wide credits | All users share from a single org credit pool. Admin adds credits. Best for teams where usage is managed centrally. |
Per-user credits | Each user has their own credit bank. Users can add credits individually. Best for large teams with varied usage patterns. |
Bring Your Own Key (BYOK) | Users plug in their own API keys from Anthropic (Claude), OpenAI, Google, or other supported providers. Best for power users who already pay for API access elsewhere. If your Claude subscription runs out, configure a fallback model so the system switches automatically rather than stopping. |
Model selection | Choose a default model for RAIN Intelligence. You can also turn specific models on or off. For example, if you have security concerns about certain Chinese-origin models (such as GLM or Minimax), you can disable them org-wide. |
IMPORTANT - VOICE AGENTS USE SEPARATE BILLING: BYOK keys for RAIN Intelligence do NOT apply to voice agent calls. Voice agents run on a static, pre-configured LLM model to ensure consistency - calls cannot switch models mid-conversation. Voice agent costs vary per models per minute all-in (LLM model + voice model + carrier fees).
Part 2: Building Your First Agent
Agent Types: Assistant, Voice and Chat
Agent Type | What It Does |
Assistant Agent | Runs in the background. Monitors records, fires next best actions when triggered, and drives proactive outreach through Next Best Action Outreach Lists. Does not make calls or send messages directly - it decides what to do and routes through configured action types. |
Voice Agent | AI voice calls - both inbound and outbound. Answers the phone, qualifies leads, handles objections, books appointments, updates Salesforce records in real time. Can also be used for outbound batch dialing via Outreach Lists. |
Chat Agent | Same conversation flow and logic as a voice agent, but for SMS and web chat. Can be deployed as a web chat widget on your website or used for SMS conversations. |
Combined Voice + Chat | One agent configured to handle both voice calls and SMS/chat using the same prompt and conversation logic. Useful when the lead intake flow is identical regardless of channel. |
Agent Builder Walkthrough
To create a new agent: go to the RAIN Agents tab in the RAIN OS app and click New. The setup wizard will walk you through the following:
1. Name, Model & Voice
Give the agent a name (e.g., 'Lead Intake Agent'). Select the LLM model. For voice agents, Kris recommends starting with GPT-4 - not GPT-5 for voice at this time. Then select a voice: male, female, various accents. There are 40+ voices available from Cartesia, OpenAI, 11 Labs, and Minimax. You can also clone a custom voice using 11 Labs community voices.
2. Template
Select a template to start from or create from scratch. The most common template is 'Real Estate Lead Manager.' Templates pre-load a suggested model, voice, and base prompt.
3. Agent Settings
Configure the welcome message - what the agent says when the call connects. Optionally add a disclaimer (e.g., 'This is a Smart Agent for [company].'). Configure pausing logic (how long the agent pauses between speaking). The agent is automatically aware of whether the call is inbound or outbound and adjusts behavior accordingly.
4. Model Settings
Standard Tier vs. Fast Tier. Standard is sufficient 80% of the time. Fast Tier provides dedicated LLM resources for lower latency and higher accuracy - recommended for high-value inbound calls where every second of response quality matters. Fast Tier is roughly double the per-minute cost.
VOICE SETTINGS: Speed, temperature (how calm vs. emotional), and volume are adjustable. Recommended range for temperature: 0.4–0.7 for natural, flexible conversation. For speed and volume, we recommend keeping settings near the middle initially, then adjusting based on the specific voice model selected - behavior varies significantly between voice providers.
The Prompt: Single Mode vs. Structured Mode
The prompt is the most important part of your agent. It defines everything: how the agent introduces itself, how it handles objections, what questions it asks, how it handles silence, how it qualifies leads, and what it does at the end of a call.
Mode | When to Use |
Single Mode | One text block. All instructions in one prompt. Best for simple agents or when getting started. |
Structured Mode | The prompt is broken into sections using markdown (# for headers, ## for subsections). The builder displays each section separately for easier editing. Recommended for production agents - it is easier to navigate and update specific sections without disrupting others. |
Best Practice - ONE GOOD PROMPT OVER AN IVR TREE: Do not try to build an IVR-style multi-prompt decision tree. Give the agent a single, clear prompt with your business logic, conversation flow, and objectives - then trust the model's intelligence to handle variations. Overly rigid, deeply nested prompts limit what the agent can do and miss scenarios you did not anticipate. A focused, well-written single prompt outperforms a complex node-and-edge tree in almost every case.
Prompt Structure (Recommended Sections)
Use the structured prompt mode and organize your prompt with these markdown sections:
# Role & Objective - who the agent is and what it is trying to accomplish
# Personality & Tone - how it should sound, its pace, its emotional range
# Instructions & Rules - communication rules, what it should and should not do
## Conversation Flow - the step-by-step flow of the conversation and qualification questions
## Lead Qualification - the specific Pillar fields it should capture
## Objection Handling - how to handle common objections
## Small Talk & Topic Guard - guardrails on how long to engage off-topic
TOKEN BUDGET: Keep prompts under 2,000–3,000 tokens for optimal performance. Prompts can extend up to 8,000 tokens before performance degrades. To stay within budget, move supporting content (scripts, objection handling guides, property types) into the Knowledge Base rather than the prompt directly. The prompt should provide structure and goals; the knowledge base provides additional context.
Prompt Assistant
If you want to build a new prompt from scratch, use the Prompt Assistant. Click the Prompt Assistant icon and type 'Create New Prompt'. The assistant asks you questions about your objective, business, and conversation flow, then generates a starter prompt in the correct structured markdown format. You can also select Analyze, Fix or Rewrite.
Knowledge Base
The Knowledge Base is a separate content store the agent references during conversations. It is distinct from the Salesforce Knowledge base and is purpose-built for RAIN OS agents. Use it to store:
Call scripts and conversation guides
Objection handling playbooks
Property type definitions or buy box criteria
Company information, FAQs, or policy guides
You can add knowledge base content in three formats: a web page URL (RAIN OS can scrape and index it, with auto-refresh every 12 hours), an uploaded file (PDF, Word, etc.), or plain text entered directly. Attach knowledge bases to agents from the agent configuration screen.
BEST PRACTICE: Put context in the Knowledge Base, not in the prompt. If your prompt is over 3,500 tokens, the most effective way to reduce token count is to move supporting content to the knowledge base. Keep the prompt focused on structure and goals.
Context Manager: Objects, Fields & Data Capture
The Context Manager controls what Salesforce data the agent has access to during a call and what data it writes back after.
Setting | What It Does |
Objects | Select which Salesforce objects the agent acts on - typically Lead and Opportunity for REI use cases. For each object, specify whether to pass in activity history, record details, and related contact information. |
Activity History | Include recent call transcripts, emails, SMS threads, and campaign responses so the agent has full conversation context when it speaks to a returning lead. |
Record Info Fields | Which fields to pass in to the call. These are what the agent can reference during the conversation - existing data about the lead, property, or opportunity. |
Data Capture Fields | Which fields the agent should specifically look for and capture during the call. After the call, these are the first fields checked in post-call analysis. If the agent captured a value for that field, it writes it to the record. |
Create New Lead (Inbound) | For inbound calls where no matching record exists, the agent creates a new Lead automatically from the caller's phone number. At the end of the call, all captured data is written to the record. |
Record Type | Specify which Salesforce Record Type new inbound leads should be created as (REI, Retail, Buyer, etc.). |
Default Field Values | Set default values for specific fields. For example, defaulting Country to 'United States' so it is always pre-populated without needing to ask. |
WATERFALL ROUTING FOR INBOUND CALLS: When a call comes in, RAIN OS searches for a matching phone number across your configured objects in priority order. For 8020CRM orgs, we recommend putting Opportunity first, then Contact, then Lead. If the lead has been converted and has an Opportunity, you want that record's financial context - not just the original Lead. If no Opportunity exists, it falls to Contact, then Lead.
Tool Manager
Tools are actions the agent can take during a call. Configure them in the Tool Manager within the agent builder.
Tool | What It Does | Admin Setup Required |
Book Appointment | Checks agent availability in Salesforce calendar and books an event during the call. Confirms with the caller: 'I've scheduled you for Tuesday at 2 PM.' | Specify which agents to book appointments for, appointment duration, buffer times, and advance notice required. |
Check Calendar Availability | Reads Salesforce calendar to find open slots without booking. Useful if the agent presents options and a human confirms. | Same as Book Appointment. |
Transfer Call | Transfers the live call to a live agent or a different number. Supports cold transfer (just connect) and warm transfer (whisper message to the receiving agent first). | Set a static transfer number or configure dynamic routing (agent decides based on call context, e.g., 'transfer to manager if caller asks'). |
Send SMS | Sends a text message during or after the call. | Configure which phone system to send through. |
Extract Variables | Pulls specific values out of the conversation (e.g., asking price, timeline) and stores them as variables for use in other tools. | Specify which variables to extract and when. |
End Call | Politely ends the call. Include a closing script in the tool configuration. | Optional closing message. |
Salesforce Action | Calls any Salesforce Flow or Apex action during the call. This is RAIN OS's most powerful tool - it enables completely custom automation triggered mid-call. | Must have a Salesforce Flow or Apex action already built and exposed. Example use cases: run underwriting calculator, trigger DocuSign, post Slack notification, update custom fields. |
SALESFORCE ACTION - THE POWER TOOL: The Salesforce Action tool is what separates RAIN OS from every third-party voice agent solution. A live call can run your deal calculator, generate an offer amount, send a DocuSign contract, and notify your team on Slack - all in one conversation. If it can be built as a Salesforce Flow or Apex action, your agent can call it. Flow actions are accessible to most admins. Apex actions require developer-level access.
Custom Variables
Custom Variables are dynamic values you define once and reference anywhere in the prompt, tools, and conversation flow using a {{variable_name}} syntax. When a call fires, these variables are resolved and injected into wherever you have referenced them.
Two types of variables:
Static variables: set a fixed value (e.g., company_name = 'KLRE'). Set once in the agent config; updates everywhere automatically. This is how you ensure the agent always uses the correct company name in voicemails - set it here, not hardcoded in the prompt.
Object-mapped variables: map a variable to a specific Salesforce field (e.g., user_street maps to Lead.Street, and resolves to the correct value based on which record type the agent is acting on). Useful for prompts that need to reference property addresses, seller names, or other record-specific data.
COMMON MISTAKE: If your agent is leaving voicemails with the wrong company name, check your custom variables. The company_name variable must be set to your client's company name in the agent configuration. The default is 'GetReal Solutions' - always update this before deploying.
Quiet Hours
Set quiet hours to prevent the agent from making calls or sending messages during late night and early morning hours. The default is 9 PM to 8 AM in the agent's time zone. Configure the time zone either as a static value or as an object-mapped variable for multi-market orgs.
Post-Call Data Extraction
After every call (or SMS conversation), the agent can run additional AI analysis on the transcript. Configure these in the Post-Call section of the agent builder. Options include:
Coaching: score the call against a rubric, identify what went well and what could improve.
Lead Scoring: evaluate the lead's motivation, timeline, and qualification level based on the conversation.
Data Capture: extract any remaining field values from the transcript.
Custom Summarization: generate a plain-language summary of what happened in the call.
All post-call extraction results are written to the Comms record and, where fields are matched, to the related Salesforce record.
Speech and Call Settings
Configure in the Settings tab of the agent builder:
Background Sound: add subtle ambient noise to make the call feel less robotic. Tested range: keep near default.
Interruption Sensitivity: how easily the caller can interrupt the agent mid-sentence. Recommended range: 0.6–0.7. Too low and the agent talks over the caller; too high and it stops mid-sentence from background noise.
Responsiveness: controls delay between when the caller stops speaking and when the agent responds. Too fast sounds unnatural; too slow feels laggy.
Voicemail Detection: when enabled, the agent detects voicemail tone and leaves an AI-generated message.
Max Call Duration: backend setting that terminates calls that run excessively long. AI models have no awareness of time; this prevents calls from staying open indefinitely.
Testing Your Agent
Before publishing, test from within the agent builder:
Click the Test Environment section in the agent builder.
Select a test record (a Lead or Opportunity in your org).
Override any variables if needed (e.g., set a test company name).
Set the call direction: Inbound (the agent answers a call) or Outbound (the agent places a call).
Test via conversation mode (type as a text conversation) or have the agent make an actual outbound call to your phone number.
Review the call transcript, field suggestions, and post-call analysis after the test.
Publishing and Assigning Phone Numbers
When the agent is ready:
Click Publish in the agent builder. Published agents can be assigned to phone numbers and used in outreach lists.
Go to Phone Numbers in the RAIN OS app and add a new number. Select by area code using Twilio or Telnyx.
Assign phone numbers to the agent - inbound numbers (calls coming to the agent) and outbound numbers (what callers see when the agent calls out). You can assign multiple numbers to a single agent.
Set Reserved Concurrency - this controls how many lines are reserved for inbound calls. The default is 5. For high-volume outbound dialing, lower this to 1 or 2 to free up more lines for outbound. Example: if you have 1 phone line with 20 concurrent capacity, setting reserved concurrency to 2 means 18 lines are available for outbound, 2 held for inbound.
CONCURRENCY & VOLUME: Each phone line supports up to 20 simultaneous calls. For a goal of 500–1,000 outbound dials per day, add 2–3 phone numbers to the agent. Each line can triple-dial through its list at 20 concurrent calls.
Part 3: Assignment Rules
Assignment Rules are the 'if-then matrix' for AI decision-making. They control which agent runs next best action analysis for a given set of records.
Without assignment rules, a default agent runs on all triggered records. With assignment rules, you can specify:
All records where Status = New → use the Lead Intake Agent
All records where Status = Nurture → use the Nurture Campaign Agent
All Opportunity records in the Acquisition stage → use the Acquisition Agent
This allows you to have different prompts, different conversation flows, and different objectives for each stage of your pipeline - without building a complex multi-prompt IVR tree. Each assignment rule says: when a record meets these conditions and a next best action fires, use this agent.
Assignment Rule Field | What It Does |
Object | Which Salesforce object this rule applies to (Lead, Opportunity, etc.) |
Conditions | Filter criteria for which records this rule applies to (Status = New, Stage = Under Contract, Owner = specific user, etc.) |
Agent | Which configured agent runs the next best action analysis for records matching these conditions |
Assignment rules are reactive - they fire when a record is triggered by an update or interaction. For proactive, scheduled outreach, use the Next Best Action Outreach List instead.
Part 4: Outreach Lists
Outreach Lists are how you deploy agents at scale - batch dialing a list, running a proactive AI campaign, or setting up real-time speed-to-lead calling on new inbound leads. There are two types.
Type 1: Voice Outreach List (AI Dialer)
A Voice Outreach List connects a Salesforce list view to the AI voice agent for batch outbound dialing.
Setting | What to Configure |
Salesforce Object | Which object to dial from - Lead is the most common; Opportunity is supported. Note: the Property object is not yet available as a dialing source as of April 2026 - it is in development. |
List View | The Salesforce list view you want to dial. The system previews the first 20 records for speed, but the actual outreach list loads all records in the list view when you click Create without selecting any checkboxes. Do not check individual names if you want the entire list - selecting checkboxes only loads the checked records. |
Phone Fields | Which phone fields to dial. You can add multiple fields (Phone, Mobile Phone, Phone 2, etc.) as fallback numbers. Add as many as your records contain. |
Agent | Which published agent to use for the calls. |
Phone Number | Which outbound number to call from. |
Immediate Redials | How many times to retry if the call is not answered (before the redial wait time). 2–3 is standard. |
Redial Wait Time | How long to wait before re-attempting the list (e.g., 4 hours). Common configuration: triple tap, 4-hour wait, up to 10 total reattempts. |
Maximum Redials | Total number of times to cycle through the list. |
Refresh Interval | How frequently to check for new records that have joined the list view. Set to Real-Time for speed-to-lead scenarios - as soon as a new web form submission hits the list view, the agent calls immediately. |
Quiet Hours | When to pause dialing. Set per time zone. |
SPEED-TO-LEAD MODE: Set Refresh Interval to Real-Time and point the outreach list at your New Inbound Leads list view. The moment a web form submission creates a lead that matches your list view criteria, the agent calls immediately - no delay, no webhook, no Zap required.
Type 2: Next Best Action Outreach List (Autopilot Mode)
The Next Best Action Outreach List is a proactive, AI-driven campaign engine. Instead of making calls, it runs your assistant agent against a list of records on a regular interval and lets the AI decide what to do next for each record.
This is the 'autopilot' capability. You give it an audience, a goal, and a set of available actions - and it runs forever, determining the best next outreach for each record based on their history, status, and conversation thread.
Setting | What to Configure |
Object & List View | The records you want the AI to process. This can be any list view on any supported object. |
Assistant Agent | The base agent prompt that contains your business logic, voice, tone, and decision framework. |
Outreach Goal / Objective | A 2 - 3 sentence description of what you are trying to accomplish with this campaign (e.g., 'Get leads to respond, take a call, or set an appointment to discuss their property'). |
Additional Instructions | Campaign-specific overlay instructions that override or supplement the base agent prompt for this outreach (e.g., 'We are running a webinar on March 15. Push all outreach toward registering for this event. Every email and SMS should have this as the call to action.'). |
Run Interval | How frequently the agent processes the list - real-time, hourly, every 4 hours, daily. |
Available Actions | What the agent can do: email, SMS, AI voice call, task assignment to a human, or Salesforce action. |
For each record in the list, the agent reviews the full activity history and conversation thread, then decides:
Is any action needed for this record right now?
If yes - what is the best channel (email, SMS, voice call, task for a human)?
What should the content be?
What is the appropriate timing?
AUTOPILOT: This is the evolution of Action Plans - instead of a rigid template sequence, you give the AI your business logic and it determines the right action at the right time for each individual record. For leads sitting untouched for months, a Next Best Action outreach list can re-engage them at scale without anyone having to manually work the list.
Part 5: Comms Records
Every call, chat, and SMS handled by a RAIN OS agent creates a Comms record. This is where you review what happened, listen to recordings, and track cost.
Data Available on Comms Record | What It Contains |
Call Recording | Audio recording of the conversation (when applicable). These are WAV files and consume File Storage in Salesforce - plan for regular cleanup. |
Transcript | Full turn-by-turn transcript of the call. Includes tool call entries showing when the agent ran a tool and what the result was. |
Context Data | What data the agent had access to when the call started - shows exactly what Salesforce data was passed to the agent. |
Field Suggestions | What field updates the agent suggested post-call and whether they were auto-applied or queued. |
Cost | Per-call cost breakdown in credits. |
Call Duration | Duration in seconds. Calls under 6 seconds are typically voicemails or unanswered. |
From the outreach list view, you can also see call summaries and link through to individual Comms records to review full transcripts. Use the Comms tab filtered to 'All' outbound calls to distinguish answered calls from voicemails by duration.
STORAGE MANAGEMENT: Call recordings in WAV format are the primary driver of Salesforce File Storage overages in orgs running high-volume AI dialing. Set a quarterly review cycle for deleting old recordings. For auto-archival to external storage (Google Drive, AWS S3, etc.), contact the 8020CRM team.
Part 6: RAIN Hub
The RAIN Hub is the unified command center for all AI activity in the org. It is accessible from the utility bar at the bottom of every page in 8020CRM and contains four tabs.
RAIN Hub Tab | What It Shows |
Inbox | All new inbound activity aggregated in one place - new leads from web forms, inbound calls, new emails, new SMS messages. Shows AI-suggested score, lead status, speed-to-lead, and quick action buttons. Audible alert when a new item arrives. |
Insights | All field suggestions and next best actions generated by agents and post-call analysis. Allows bulk approve or bulk reject. Items below the confidence threshold appear here for manual review before being applied. |
Copilot | The RAIN Intelligence AI chat interface for the current user. Ask questions about records, run pipeline analysis, build reports - see the RAIN Intelligence Workspace guide. |
Intelligence | Same as the RAIN Intelligence Workspace accessed from the utility bar or App Launcher. |
SMS Assist
SMS Assist is a background AI feature that runs when drip campaign templates are sent. Instead of delivering a boilerplate template, RAIN OS passes the template through AI before sending, personalizing it based on the recipient's record details and conversation history.
The template structure is preserved - the AI adjusts tone, details, and phrasing to match the individual. This helps avoid carrier spam filters and adds a human feel to templated outreach. Enable SMS Assist in Features & Settings.
Web Chat & Voice Widget
Any configured voice or chat agent can be deployed as a web widget on your website. The widget generates a snippet of embed code you place on your site. When a prospect initiates a chat or requests an AI call from the widget, it runs the same agent, follows the same prompt, calls the same tools, and writes the same data back to Salesforce.
Customize the widget with your brand's icon, colors, and logo. Configure whether the widget offers text chat, a request-a-call button, or both.
Common Issues & Fixes
Issue | Fix |
Agent is leaving voicemails with the wrong company name | Check the company_name custom variable in the agent configuration. Update it to the client's company name. Also verify the voicemail message in Call Settings references the {{company_name}} variable rather than a hardcoded name. |
Outreach list only shows 20 records | This is intentional - the system previews 20 for speed. Do NOT select any checkboxes. Click Create without selecting any records and the entire list will load. |
List view filter includes wrong records | Check for extra commas in filter values. A trailing comma in a 'contains' filter causes it to match everything. Remove the extra comma and the filter will work as intended. |
Skip-traced phone numbers not appearing in list view columns | This is a known issue being addressed in an upcoming release. Skip-traced numbers are on the record but may not display in list view columns immediately after import. |
Token count is too high (prompt over 3,500 tokens) | Move supporting content (scripts, objection handling, property types) from the prompt into the Knowledge Base. Reference the knowledge base from the agent configuration. Keep the prompt focused on structure and goals. |
Voice agent running on wrong LLM model | Voice agents run on a static model configured at the agent level - not your BYOK keys for RAIN Intelligence. Go to the agent's model settings and verify the correct model is selected. BYOK keys are for RAIN Intelligence only. |
Post-call field suggestions not appearing | Verify that the fields you want captured are selected in the Context Manager under Data Capture. Fields not selected here may still be captured in the transcript but will not be prioritized in post-call analysis. |
Agent not creating new leads on inbound calls | In the Context Manager, verify that 'Create New Lead' is enabled for inbound calls and that a Record Type is selected. If no record type is specified, lead creation may fail. |
Best Practices
Build separate agents for separate use cases. A lead intake agent, an acquisition agent, a dispo agent, and a foreclosure agent should each have their own prompt. Trying to handle all scenarios in one agent creates prompt complexity that limits performance.
Set company_name and agent_name as custom variables immediately after creating an agent. Never hardcode these in the prompt - you will miss them in voicemail settings and other places.
Use structured prompt mode for all production agents. It is easier to update specific sections and less likely to accidentally break unrelated parts of the prompt when editing.
Keep prompts under 3,000 tokens. Put supporting content in the knowledge base.
Test every agent on a real record with a real outbound call to your phone number before deploying to a production list.
For inbound calls, set the object waterfall to Opportunity first. For REI orgs, the Opportunity has the most relevant financial and qualification data.
For Next Best Action Outreach Lists, start with a narrow audience and a simple goal. Expand once you have validated the agent's output against real records.
Monitor Comms records weekly for calls with short durations (under 6 seconds) - these are voicemails or unanswered calls. If your answered rate is low, review your phone number health and carrier compliance status.
Plan for File Storage growth from day one if running high-volume dialing. WAV recordings are the #1 cause of Salesforce storage overages. Set quarterly cleanup cadence or implement automated archival.
For the Salesforce Action tool: always test your Flows in a sandbox before connecting them to a live agent. An error in the Flow will cause the tool call to fail mid-conversation.
Comments