Leads Sit Unassigned? Fix Routing with an AI Workflow

New enquiries arrive through forms, ads, email, WhatsApp and social messages, but ownership is unclear. A lead may sit in an inbox for hours, reach the wrong salesperson or get entered into the CRM twice.

An AI routing workflow can solve the operational gap without trying to close the sale itself. It reads the enquiry, extracts the useful details, applies simple rules, assigns the lead and creates a visible next step for a human.

Define the routing decision before choosing the AI

The most important question is not which model to use. It is what decision should happen after a new enquiry arrives?

For example, a property business might route enquiries by:

• Location or project
• Budget range
• Buying or renting intent
• Preferred language
• Urgency
• Whether the person has provided a phone number

An e-commerce store may use different fields:

• Product or category mentioned
• Order number
• Pre-sale question or support request
• Delivery country
• Wholesale or retail intent

Keep the first version narrow. If the workflow has to understand every possible message, it becomes difficult to test and easy to distrust. Choose three to five routing fields that genuinely affect ownership or response priority.

Write the fallback rule

AI will sometimes find incomplete, contradictory or unclear information. Decide what happens before the workflow goes live.

A sensible fallback might be: if confidence is low, assign the enquiry to a general sales queue and label it “needs review”. Do not let the system silently guess a salesperson, budget or customer intent.

Build the workflow around a single intake event

In n8n or Make, the workflow should begin whenever a new enquiry is created. That event could come from a website form, a shared inbox, a CRM, a lead ad integration or a webhook from another tool.

The basic sequence is:

1. Receive the new enquiry and record its source.
2. Normalise the text, name, email, phone number and timestamp.
3. Check whether the person or email already exists in the CRM.
4. Ask the AI to return structured fields, not a paragraph.
5. Apply deterministic routing rules.
6. Create or update the CRM record.
7. Assign an owner and create a follow-up task.
8. Notify the responsible person in the agreed channel.
9. Log the result for reporting.

The AI should handle interpretation. Your automation platform should handle actions and business rules. For example, the model can classify “I need a two-bedroom apartment near the metro” as a residential enquiry and extract location preference. The workflow should decide which team receives residential enquiries for that location.

This separation makes the system easier to audit. You can change team assignments without changing the prompt, and you can improve the prompt without accidentally changing your sales territories.

Require structured output from the model

Do not ask an AI agent to “read this lead and decide what to do” without defining the output. Use a fixed schema with permitted values.

A practical lead triage output could include:

• intent: sales, support, partnership or other
• language: Arabic, Turkish, English or unknown
• product: one of your approved product categories
• urgency: low, normal or high
• location: approved location or unknown
• missing_fields: list of information still needed
• confidence: high, medium or low
• summary: one short sentence for the owner

Use controlled values wherever possible. “High urgency” should not mean whatever the model happens to interpret on a given day. Define it in the workflow, such as an existing customer reporting a payment problem or a buyer requesting a quotation for a time-sensitive purchase.

If your platform supports validation, reject outputs that contain invalid categories. Send them to a review path instead of allowing malformed data into the CRM.

Add routing rules that humans can understand

The routing layer should be visible and predictable. A simple example might be:

• Arabic enquiries go to the Arabic-speaking sales queue.
• Wholesale enquiries go to the partnerships owner.
• Existing customers with an active order go to support.
• High-value or high-urgency enquiries create an immediate notification.
• Low-confidence records go to a shared review queue.
• Leads with no contact method are tagged for data completion rather than assigned as ready-to-contact.

These rules can be implemented with filters and branches in n8n or Make. The AI does not need authority to override them.

For example, if a message contains both a product question and an order number, your workflow can use the order number as the stronger signal and route it to support. The AI may identify both intents, but the business rule decides which team owns the next action.

Make the handoff impossible to miss

Creating a CRM record is not the same as creating accountability. Every routed enquiry should have an owner, a status and a next action.

A useful task might say: “Reply to Ahmed about the Turkish-language quotation request for Product A. Confirm delivery location and quantity.” This is more useful than a notification saying “New lead received”.

Choose one primary notification channel. It could be email, Slack, Microsoft Teams or a CRM task queue. Avoid sending the same alert to five places unless the enquiry meets a clearly defined escalation condition.

You can also add a response timer. For example, if a high-priority lead has no status change after two business hours, the workflow can notify a manager. This is an example policy, not a universal benchmark; set the threshold according to your team and customer expectations.

Measure routing quality, not just volume

Your first dashboard should answer operational questions:

• How many enquiries entered the workflow?
• What percentage received an owner?
• How many went to fallback review?
• Which sources create the most incomplete records?
• How long did assignment take?
• How many assigned leads received a status update?
• Which classifications are often corrected by staff?

Review a sample of routed records every few days at the beginning. Compare the AI classification with the human correction, then update the categories, examples or rules causing confusion.

Do not judge the workflow only by how many leads it processes. A fast system that sends qualified buyers to the wrong team can create more work than it removes.

A practical implementation checklist

Before switching on automatic routing, confirm that:

• The intake sources and trigger events are listed.
• The AI output uses fixed fields and approved values.
• Low-confidence and missing-data paths are defined.
• CRM duplicate checks happen before creating a new record.
• Every successful route creates an owner and next action.
• Notifications have one clear primary destination.
• Corrections are logged for prompt and rule improvements.
• A human can pause the workflow without losing new enquiries.

How ADMOV can help

ADMOV can design and build an AI lead triage workflow in n8n or Make around your existing forms, inboxes, CRM and notification tools. We can define the routing schema, add confidence and fallback controls, connect multilingual intake where needed, and create a reporting view that shows where enquiries are being delayed.

The goal is not to replace your sales team. It is to remove the avoidable delay between an enquiry arriving and the right person taking ownership. Book a free call to review your lead-routing process.