🤖 AI-Powered Sales Intelligence
Welcome to the complete functional, technical, and user documentation for the AI-Powered Sales Engine in iSales CRM. This section explains the core AI capabilities, how they support real-world sales workflows, and how they connect to the application's APIs
AI Sales Features — Overview and Architecture
The AI Sales Engine is built into the CRM to help sales teams do three things faster:
- Understand information — summarize calls, notes, images, emails, and prospect profiles.
- Decide what to do next — identify risks, prioritize follow-ups, and recommend actions.
- Take action — enrich records, create structured notes, run qualification calls, and update the sales workflow.
You do not need to open a separate AI tool. AI results appear inside the screens where sales work already happens, including the Dashboard, Lead Details, Deal Details, Add Lead drawer, Add Outcome drawer.
How the AI Works — In Four Simple Steps
Step
Stage
What Happens
AI Flow
Example
1. Capture
CAPTURE
The CRM receives information from a URL, conversation, uploaded image or existing activity history.
URL, call, notes, image, or questionnaire answers
A representative enters https://stripe.com or uploads a meeting whiteboard.
2. Understand
UNDERSTAND
AI extracts text, transcribes audio, reads the company profile, or combines related CRM history.
AI reads or transcribes the information securely
The system reads the image and identifies requirements, dates, and action items.
3. Analyze
ANALYZE
The analysis layer identifies important sales signals such as pain points, inactivity, buying intent, deal risk, and likely next steps.
AI finds risks, blockers, intent, and next steps
A negotiation that has been inactive for 30 days is marked as high risk.
4. Assist
ASSIST
The result appears in a practical CRM format that the user can review, edit, save, or act on.
CRM shows a summary, score, alert, note, or recommended action
A recap appears on the Deal Details page and a follow-up appears in Your Radar.
Where Each AI Capability Appears
AI Capability
What It Helps With
Where Users See It
AI Follow-Up Intelligence
Finds overdue, stalled, and high-risk opportunities and recommends the next action.
Dashboard, Your Radar, Suggested Next Actions
Discussion Recap
Summarizes the latest customer interaction and agreed commitments.
Lead Details and Deal Details
AI Outcome Summary
Explains project context, pain points, blockers, sentiment, and chance of closing.
Lead Details, Deal Details, and table-level AI tooltips
Image-to-Outcome Extraction
Converts whiteboards, screenshots, and handwritten notes into editable CRM notes.
Add Outcome drawer
Simple Architecture View
┌──────────────────────────────┐
│ SALES INFORMATION │
│ notes · images │
└──────────────┬───────────────┘
│
▼
┌──────────────────────────────┐
│ SECURE AI PROCESSING │
│ Backend proxy · Groq LLMs │
│ OCR · transcription · SSE │
└──────────────┬───────────────┘
│
▼
┌──────────────────────────────┐
│ USEFUL CRM RESULTS │
│ Recaps · summaries · scores │
│ alerts · notes · next actions│
└──────────────────────────────┘
The central AI analysis and proxy layer connects the main AI services, including image extraction and cold-call session analysis. The proxy layer keeps external service credentials away from browser clients and returns results to the appropriate CRM screen.
Purpose and Business Value
Traditional CRM workflows are slowed by manual research, fragmented notes, missed follow-ups, and unstructured cold outreach. The AI Sales Engine addresses these bottlenecks by embedding intelligence directly into daily sales operations:
- Eliminates manual research: Enriches company headcount, industry, corporate profiles, and prospect information.
- Ingests meeting information instantly: Extracts structured notes from screenshots, whiteboards, handwritten notes, and chat exports.
- Detects revenue risk proactively: Surfaces stalled deals, overdue calls, inactivity, and slipping opportunities.
- Preserves sales continuity: Converts historical interactions into concise recaps and actionable next steps for the current record owner.
What Users Should Know
- AI-generated content is a starting point for review, not a replacement for professional judgment.
- Users can review and edit AI-generated notes before saving them to a lead or deal.
- AI recommendations are based on the information available in the CRM and may be incomplete when notes, transcripts, or profile data are missing.
- Sensitive audio and third-party service requests are routed through authenticated backend proxies.
AI Follow-Up Intelligence
What This Feature Does
AI Follow-Up Intelligence continuously checks the health of every active Lead and Deal. It turns CRM activity into a simple priority list so representatives know who needs attention, why it matters, and what to do next.
How the Follow-Up Architecture Works
CRM activity and deal data
↓
Health evaluation engine
(days in stage · last response · close-date proximity)
↓
Priority and severity decision
↓
Your Radar + Suggested Next Actions
↓
Reschedule · Log Outcome · Change Status · Open Record
↓
Updated CRM activity and follow-up history
The engine uses three primary signals:
- Time in the current stage — identifies deals that may be stuck.
- Time since the last customer response or representative activity — identifies records that may be going cold.
- Distance from the target close date — identifies opportunities that need immediate action to protect revenue.
Example of the User Experience
Priority
Example Alert
Recommended Response
🔴 High
Lead #104 at Acme Corp has had no touchpoint for 14 days and is overdue by 48 hours.
Open the lead, log the outcome, or reschedule with a reason.
🟡 Medium
A $45,000 Tech Global deal has remained in Negotiation for more than 30 days.
Review the deal, contact the customer, and update the stage or next follow-up.
🟢 Low
Lead #202 completed an introductory call and is waiting for a proposal.
Send the proposal or create the next scheduled follow-up.
Your Radar — Proactive Alert Feed
Your Radar is the risk-focused part of the feature. It helps users act before a lead or deal is forgotten.
- Severity dots: 🔴 High, 🟡 Medium, and 🟢 Low.
- Dynamic health scoring: Evaluates stage duration, the last customer response, and target close date.
- Instant actions: Opens drawers for Reschedule with a mandatory reason or Log Outcome without leaving the dashboard.
- Direct navigation: Opens the full Lead or Deal Details page.
- Owner awareness: Administrators can see the representative responsible for the flagged record.
Suggested Next Actions — Daily Task Hub
Suggested Next Actions converts recent discussion recaps and call notes into practical tasks. For example:
“Client requested revised pricing with dedicated developer terms. Send the updated quote.”
- Action: Opens the related Lead or Deal or its action drawer.
- Clear: Dismisses the recommendation after the task is completed.
- Stateful completion: The cleared item remains dismissed until new notes or activity create a new recommendation.
- Priority order: Urgent and overdue actions appear before routine tasks.
Lead Follow-Ups vs. Deal Follow-Ups
Follow-Up Type
Sales Stage
Typical Work
Information Used
Lead Follow-Up
Discovery and qualification
Introductory calls, qualification, first outreach
Lead activity, notes, contact response, and next-touch date
Deal Follow-Up
Proposal through contract signature
Proposal reviews, negotiations, technical demos, and contract discussions
Deal stage, monetary value in currencies such as $ or ₹, activity history, and target close date
What the User Can Do Next
- Open the Radar alert or Suggested Next Action.
- Review the record context and recent activity.
- Select Log Outcome, Reschedule, Change Status, or Open Record.
- Save the update so the next recommendation can be recalculated.
Discussion Recap
How the Recap Architecture Works
Discussion Recap gives a representative a quick, reliable briefing before a call or meeting. Instead of reading every email, note, attachment, and previous outcome, the user can review the latest customer context in one place.
How the Follow-Up Architecture Works
Calls · emails · meeting notes · outcomes · attachments
↓
Interaction history collector
↓
AI recap and commitment extraction
↓
Last Recap Summary on Lead or Deal Details
↓
Representative reviews the context and takes action
The recap is built from timestamped CRM activity and is designed to preserve both what was discussed and what must happen next.
Example Recap Card
Recap Field
Example
Touchpoint
22 Sep 2026 · Zoom Discovery Call
Contact
Michael Chang · CTO, RetailTech Inc.
Discussion
Evaluated React and Node.js architecture for a legacy migration.
Timeline
Target kickoff in Q4 with three dedicated engineers.
Commercial Context
$35,000 budget approved with a fixed milestone structure.
Agreed Next Step
Deliver a customized NDA and sprint milestone roadmap by Thursday at 3:00 PM.
What Users Get from the Recap
- Instant pre-call context: Provides a short briefing of previous commitments and eliminates repetitive questions such as “Where did we leave off?”
- Clear commitments: Separates discussion points from agreed next steps so the representative knows what to complete.
- Seamless rep reassignment: When ownership changes because of offboarding or territory reassignment, the new owner receives historical continuity without reading raw email chains.
- Audited interaction history: Cross-references timestamped notes, outcomes such as Interested, Rescheduled, and Meeting Done, and uploaded attachments.
Where It Appears
- Lead Details: Shows the latest prospect context before discovery or qualification activity.
- Deal Details: Shows the latest commercial, technical, or contract discussion before the next deal action.
- Activity history: Uses the same logged notes and timestamps that support the CRM audit trail.
AI Outcome Summary
What This Feature Does
AI Outcome Summary turns a large amount of customer history into a clear sales-health briefing. It helps representatives understand the opportunity, identify blockers, estimate likelihood to close, and choose the next commercial or technical action.
How the Summary Architecture Works
Lead or Deal history
(notes · calls · meetings · emails · files · stage activity)
↓
Structured summary · pain points · chance score · next step
↓
Lead Details · Deal Details · table-level hover tooltip
Example of the User Experience
AI Result
Example
Overall Summary
Enterprise prospect needs a full-stack rebuild with AWS microservices.
Decision Context
Decision maker is comparing two vendors and requires SOC2 compliance.
Pain Points and Blockers
Legacy database timeouts above five seconds, EU data-residency requirements, and a $60,000 Phase 1 budget cap.
Chance Score
78% likelihood to close · High intent · Pricing alignment verified.
Suggested Next Step
Provide the security whitepaper and schedule a technical deep dive with a Solutions Architect.
What Each Result Means
- Overall Executive Summary : Project scope, technical requirements, decision context, and commercial boundaries.
- Client Pain Points and Blockers : Technical vulnerabilities, organizational bottlenecks, compliance needs, competitor concerns, and budget constraints.
- Chance Score Assessment : A 0–100% estimate based on engagement signals, customer intent, and stage duration.
- Suggested next step: A practical action that can move the Lead or Deal forward.
Where Users See It
- Lead Details: Full summary for prospect qualification and outreach planning.
- Deal Details: Summary and risk context for active pipeline opportunities.
- Lead listing and Deal Table : Hover over the AI icon to scan a short summary without opening the record.
Review Guidance
AI output should be reviewed against the original notes and customer context. Users can validate, edit, and add information before saving a note or acting on the recommendation.
Company Web Scraping and AI Profile Auto-Enrichment
What This Feature Does
Company Web Scraping and AI Profile Auto-Enrichment reduces manual research when a representative creates or updates a Lead, Contact, or Company. The representative enters a company website, and iSales CRM retrieves available public company information and displays it in the form for review before saving.
It can populate the company profile in seconds while keeping the representative in control of the final information.
Company Scraping Architecture
User enters company website
↓
URL normalization and validation
↓
Duplicate company check
↓
Existing company? Link to the existing account
↓ No
AI web-scraping service receives the URL
↓
Company fields are parsed and normalized
↓
Add Lead / Contact / Company form is populated
↓
Representative reviews, edits, and saves
Example Workflow
Step
What Happens
1. Enter the website
The representative enters a company website, such as https://stripe.com.
2. Check the website
iSales CRM removes unnecessary trailing characters, validates the URL, and confirms that the domain is usable.
3. Check for an existing company
The CRM checks whether the company already exists in the organization's records.
4A. Existing company
The new Lead or Contact is linked to the existing Company account. This prevents duplicate company data.
4B. New company
The CRM triggers the AI scraper
5. Fill in company details
Available company information is placed into the form automatically.
6. Review and save
The representative checks the information, makes corrections if needed, and saves the record.
In simple terms: Enter the website → the CRM checks the company → existing companies are reused or new companies are enriched → review the details → save.
What the User Sees
- Open the Add Lead, Update Lead, or Contact form.
- Enter a company website in the Company Website field.
- Leave the field or wait for the background lookup to start.
- Review the company information returned by the scraper.
- Correct or complete any missing information manually.
- Save the Lead, Contact, or Company record.
The AI does not silently save scraped information. The representative reviews the populated values before they become part of the CRM record.
Company Fields That Can Be Populated
CRM Field
Example
How It Is Used
Company Name
Stripe, Inc.
Stores the legal or display name of the company.
Number of Employees
7,000
Cleans the available headcount into a numeric value for reporting.
Company LinkedIn
https://linkedin.com/company/stripe
Links the CRM account to the official company LinkedIn page.
Company Description
Global financial infrastructure platform for businesses.
Provides a short overview of the company's business and services.
Headquarter Address
510 Townsend St, San Francisco, CA
Stores the available headquarters or registered office address.
Only information available for the submitted domain is populated. Missing values remain editable so the representative can enter them manually.
Key Functional and Technical Mechanics
1. Real-Time Trigger and URL Normalization
The lookup starts from the onBlur event on the Company Website field
Before the request is sent, the form normalizes the URL by trimming trailing slashes, such as https://example.com/ → https://example.com, and validates supported formats such as https://, http://, or www. using the configured validation pattern.
2. Duplicate Account Prevention Shield
Before dispatching the scraping request, the form runs checkCompanyExistence(website).
If the domain already exists in the organization's database:
If the domain already exists in the organization's database:
- The new record is linked to the existing parent Company.
- Redundant company fields are disabled with
isDisabled = trueso the existing account remains the source of truth. - The user is informed: “This company already Exists! If you want to add more contact click on add more contact button”.
This protects the workspace from duplicate company accounts and keeps contacts, leads, deals, ownership, and communication history connected to the correct organization.
3. Scraping and Field Population
For a new domain, handleWebScrapping processes the response and maps the available data into the CRM form:
- Company Legal Name : Uses available OpenGraph tags, title tags, and corporate metadata.
- Employee Size : Cleans the returned value with numeric filtering for consistent analytics and reporting.
- Corporate LinkedIn URL : Finds and binds the official company LinkedIn page when available.
- Company Description : Provides an overview of the company’s products, services, or About Us content.
- Registered Address : Extracts the available headquarters or registered office address.
4. Graceful Fallbacks and Partial-Data Warnings
The form remains usable when a domain has anti-bot protection, limited public information, or incomplete metadata:
- If the website cannot be read, the CRM displays: “The website data is not available, kindly enter the data manually”.
- If a specific field is unavailable, the form keeps the other successful values and shows a targeted helper message, such as “Number of employees data not available, please enter manually”.
- Representatives can always correct, complete, or replace the returned values before saving.
Business Value
- Less manual research: Representatives do not need to copy company information from several pages.
- Cleaner account data: URL normalization and duplicate checks protect the organization directory.
- Faster lead creation: Company and contact context can be available while the representative creates the Lead.
- Better preparation: Company information and LinkedIn intelligence give the representative useful context before outreach.
- Human review remains required: Scraped and AI-generated values are suggestions and should be checked before saving or using them in a customer conversation.
Image-to-Outcome Extraction
What This Feature Does
Image-to-Outcome Extraction converts visual information into editable CRM notes. It is useful when important details are captured on a whiteboard, in handwritten notes, on a business card, or in a screenshot of a WhatsApp, Slack, or Skype conversation.
How the Image Architecture Works
Image selected in Add Outcome drawer
↓
AddOutComeDrawer.tsx validates the upload
↓
POST /summarize-content sends the image to the vision and OCR service
↓
Extracted text is formatted as HTML
↓
React Quill receives the content through noteContent
↓
Representative reviews, edits, and saves the outcome
Supported Inputs
.png.jpg.jpeg.webp
Typical inputs include whiteboard diagrams, handwritten meeting notes, conference business cards, and chat screenshots.
Example of the Generated Notes
<p><strong>Discussion Notes from Whiteboard Architecture:</strong></p>
<p>• Client requires PostgreSQL database migration with a GraphQL API layer.</p>
<p>• Deadline: 15 December for beta deployment.</p>
<p>• Action: Schedule a follow-up demo on Monday at 11:00 AM.</p>
What the User Sees
- Open the Add Outcome drawer.
- Select Upload & Extract.
- Upload a supported image.
- Watch the status message: AI is analyzing your input…
- Review the generated notes in the Quill editor.
- Correct or add context where needed.
- Save the outcome and schedule the next follow-up.
The AI does not silently save extracted text. The representative can review and edit the generated content before it becomes part of the Lead or Deal activity history.
Technical Architecture, API Endpoints, and Components Matrix
API Endpoints Reference
Feature
Backend Service / Purpose
Lead AI Summary
Generates the overall summary, recap, pain points, and chance score for leads.
Deal AI Summary
Generates the executive summary and risk assessment for active deals.
Image Extraction
Vision OCR endpoint that extracts text from uploaded images.
Web Auto-Enrichment
Scrapes a company URL for name, LinkedIn, headcount, industry, and description.
Secure Processing and Data Flow Principles
- AI requests are routed through backend proxies rather than exposing third-party credentials in browser clients.
- Audio recordings are served through authenticated proxy endpoints.
- Server-Sent Events provide real-time status for LinkedIn analysis.
- AI-generated content remains reviewable and editable before a representative saves it to a CRM record.
- AI recommendations complement, rather than replace, representative and manager review of sales records.
