# Lead classification and handoff

Extract details from incoming leads, review uncertain classifications and hand records to your business systems.

## Preserve the source of each lead

Collect incoming emails or form submissions with their original identifier, date and source. Define a small set of business categories and the fields needed by the person receiving the lead. In a property workflow, these might include location, property type, contact preference and missing information.

The output is a classified record with supporting evidence and a suggested handoff. It should remain possible to return to the original message.

## Build the processing flow

1. Read new messages through your existing email or CRM integration.
2. Remove irrelevant formatting while retaining the complete business message.
3. Ask the model to assign a category and extract only supported fields.
4. Apply deterministic validation to dates, required identifiers and duplicates. Send ambiguous cases to review.
5. Pass the approved record to the destination system and store whether that handoff succeeded.

## Protect against quiet mistakes

A wrongly discarded lead may be more costly than an extra manual review. Keep an explicit uncertain category and review samples from both accepted and rejected messages. Compare results against examples your team has labeled, including short messages, forwarded threads and genuine enquiries that resemble spam.

Do not treat a model's confidence number as a calibrated probability. Evaluate the actual false-positive and false-negative behavior on your own representative messages.

## Connect the systems

Use [text models](https://seedrouter.ai/text-generation) for classification and extraction, and [n8n](https://seedrouter.ai/integrations/n8n) or your application code for message retrieval and handoff. SeedRouter does not grant access to a mailbox, CRM or property database.

For the industry context, see [real estate workflows](https://seedrouter.ai/industries/real-estate). Keep model-call costs, review time and the cost of missed enquiries separate when evaluating the workflow. [Current prices](https://seedrouter.ai/pricing) cover the model-rate component.

## Related resources

- https://seedrouter.ai/models/gpt-6-sol
- https://seedrouter.ai/models/claude-opus-5-5
- https://seedrouter.ai/integrations/n8n
- https://seedrouter.ai/integrations/openclaw
- https://seedrouter.ai/industries/real-estate
- https://seedrouter.ai/blog/claude-opus-5-5-prompting-guide