I would like to suggest extending the Intelligence module with AI-assisted processing of incoming emails.
A significant amount of time is currently spent manually reviewing emails that EspoCRM linked to the wrong Parent record or could not categorize correctly. Technical email parameters such as sender address, subject, Message-ID, and thread headers are useful, but they are not always sufficient to understand the actual purpose and context of a message.
The proposed functionality would combine deterministic rules with LLM-based context analysis.
Suggested concept
Administrators should be able to create AI email-processing rules for each Group Email Account. A rule could contain:
The AI should be able to use the email subject, body, previous messages in the thread, and selected CRM context to:
Example 1: An external support system
An external support platform sends a follow-up email about an existing issue. Because the external platform changes the subject or does not preserve the original email-thread headers correctly, EspoCRM creates a new Case.
The AI could recognize the external ticket number, customer, product, problem description, and previous Case context, and then link the email to the existing Case instead of creating a duplicate.
Example 2: Subscription expiration notifications
A manufacturer sends automatic notifications about subscriptions that will expire soon. EspoCRM may currently link these emails only to the related senders Account.
An AI rule could recognize the customer, product, subscription, and expiration date, and then:
Example 3: Support request field completion
An incoming support request may contain the product name, serial number, site, urgency, and problem description in unstructured text.
The AI could extract this information and populate the configured Case fields before assigning the Case to the support team.
Important safeguards
This would turn the Intelligence extension into a practical email-triage and data-entry assistant, while keeping administrators in control through explicit rules and confidence thresholds.
A significant amount of time is currently spent manually reviewing emails that EspoCRM linked to the wrong Parent record or could not categorize correctly. Technical email parameters such as sender address, subject, Message-ID, and thread headers are useful, but they are not always sufficient to understand the actual purpose and context of a message.
The proposed functionality would combine deterministic rules with LLM-based context analysis.
Suggested concept
Administrators should be able to create AI email-processing rules for each Group Email Account. A rule could contain:
- traditional conditions, such as sender, recipient, domain, subject pattern, email headers, or body keywords;
- natural-language instructions describing how the email should be interpreted;
- a list of CRM records and fields that the AI is allowed to examine;
- expected structured output;
- actions that EspoCRM should perform;
- a confidence threshold for automatic processing.
The AI should be able to use the email subject, body, previous messages in the thread, and selected CRM context to:
- identify the correct existing Parent record;
- link the email to an existing Case, Account, Opportunity, or another supported entity;
- create a new Case, Task, Opportunity, or another configured record;
- extract structured information into CRM fields;
- populate missing Case or support-request fields;
- assign the record to a team or user;
- flag uncertain emails for manual review.
Example 1: An external support system
An external support platform sends a follow-up email about an existing issue. Because the external platform changes the subject or does not preserve the original email-thread headers correctly, EspoCRM creates a new Case.
The AI could recognize the external ticket number, customer, product, problem description, and previous Case context, and then link the email to the existing Case instead of creating a duplicate.
Example 2: Subscription expiration notifications
A manufacturer sends automatic notifications about subscriptions that will expire soon. EspoCRM may currently link these emails only to the related senders Account.
An AI rule could recognize the customer, product, subscription, and expiration date, and then:
- link the email to the correct Account;
- create a follow-up Task; or
- create or update an Opportunity for the renewal process.
Example 3: Support request field completion
An incoming support request may contain the product name, serial number, site, urgency, and problem description in unstructured text.
The AI could extract this information and populate the configured Case fields before assigning the Case to the support team.
Important safeguards
- Deterministic rules and exclusions should take priority over AI decisions.
- Administrators should explicitly define which entities and fields may be sent to the AI provider.
- Automatic actions should be performed only above a configurable confidence threshold.
- Lower-confidence results should be placed in a review queue.
- Each action should have an audit log containing the matched rule, selected Parent, confidence score, model, and extracted values.
- Administrators should be able to configure whether a rule may only suggest an action or execute it automatically.
- Manual corrections should ideally be usable as feedback for improving future classification.
This would turn the Intelligence extension into a practical email-triage and data-entry assistant, while keeping administrators in control through explicit rules and confidence thresholds.
Comment