AI System for Preparing Multichannel Leads for CRM Routing

The problem

When sales enquiries are spread across a website, different form tools and messaging channels, it becomes difficult to see every lead in one place.

If the same person makes contact through several channels, duplicate CRM records may be created. The sales team then has to determine by hand whether the contact is a new opportunity or a continuation of an existing conversation.

Free-text enquiries differ in product, service, budget, timing and purchasing intent. Simple forms made up of fixed fields do not always capture this context.

When clear rules do not determine which team or sales representative should receive a lead, records remain in a shared queue or are routed to the wrong person.

What we set out to improve

• Time between receipt of a lead and creation of its CRM record
• Number of leads waiting without an assigned sales representative
• Rate of records routed to the wrong team or sales representative
• Number of duplicate CRM records for the same person
• Rate of uncertain classifications sent for human review

How the system works

1. 01 Receive a new enquiry from permitted website, form and messaging channels
2. 02 Convert channel-specific fields into a common lead schema
3. 03 Standardise identifiers such as telephone numbers and email addresses
4. 04 Search existing CRM records for the same person or company
5. 05 Extract the subject, product or service interest and a needs summary from the enquiry text
6. 06 Prepare classification and routing suggestions from predefined rules
7. 07 Separate uncertain, incomplete or conflicting records for human review
8. 08 Create the approved CRM record or update an existing record
9. 09 Have the relevant sales team or representative confirm ownership of the lead
10. 10 Log the source, classification and routing decision

Methods we used

• Data normalisation that combines the different fields from each channel into one CRM schema
• Format standardisation and exact-match checks for telephone number and email fields
• A two-stage decision structure that separates deterministic rules from AI classification
• A structured output schema that extracts only defined fields from free text
• A routing table based on product, service, region or sales-team rules
• A human review queue for low-confidence and conflicting records
• Channel and record identifier checks that prevent the same event from being processed again
• Source and processing logs for every decision written to the CRM

Where people stay involved

Data and security

The system should process only the fields required for the lead record; channel access and CRM write permissions should be restricted by role. Unnecessary personal or sensitive data in free text should be removed before content is sent to AI. Every record creation, update and routing action should be retained in an auditable processing log.

Who this suits

Frequently asked questions