Parsing Property Portal Lead Emails Into Clean Records Before Any AI Touches Them

Property portals often send leads to small agencies via email notifications. Before incorporating AI into the follow-up process, it is crucial to convert these emails into clean, structured records. This initial step, although not glamorous, lays the foundation for efficient lead automation in real estate. The desired format includes fields such as portal name, received timestamp, name, phone number, email, listing reference, message, and raw email ID.

To accommodate different email formats from various portals, each portal requires a dedicated parser. These parsers must be written based on the sender's email address, with routing determined by the "From" header. The parsing process involves utilizing regular expressions to extract relevant information from the email body.

The provided Python code snippet demonstrates a parser for portal A. It defines a function called "parse_portal_a" that takes the email body as input and returns a dictionary containing the extracted details. Regular expressions are used to match specific patterns, such as name, phone number, email, listing reference, and message. If a pattern is found, its corresponding group is extracted and returned after removing any leading or trailing whitespace.

To handle multiple portals, a dictionary called "PARSERS" is created, mapping the sender addresses to their respective parsing functions. When parsing an email, the code retrieves the appropriate parser based on the sender's email address. If no parser is found, the email is routed to a human for further handling.

Before storing the parsed records, validation checks are performed. The code ensures that the required fields, such as listing reference, are present. Additionally, it checks if at least one of the optional fields, such as phone number or email, is provided. If validation fails, the email is sent to a person for review, along with a reason indicating the validation issues.

To maintain flexibility and adaptability, it is essential to store the raw email ID. This allows for re-parsing in case a portal modifies its email format unexpectedly. Additionally, phone numbers should be normalized to a consistent format to facilitate accurate matching against existing contact records. Failure to handle phone numbers correctly can lead to missed follow-ups or miscommunication.

It is crucial to be explicit about validation failures and not rely on silent failures. If a parser fails to extract a phone number, the lead should not be silently discarded but rather directed to a person for manual resolution. This ensures that no lead is left without a viable contact method.

AI should only be employed for processing the free-text message portion of the lead. Tasks such as extracting information like budget, move-in date, or the customer's intent (e.g., "wants to view this weekend") can benefit from AI techniques. These extracted details should be stored in separate fields, preserving the deterministic parsing results while leveraging AI for more complex tasks.

Finally, it is advisable to write tests for each portal using saved real emails, with personal details removed. This ensures that the parsing logic works as expected and handles various email formats consistently. Once the intake step is complete, subsequent processes such as deduplication and follow-up timers become significantly easier, as all leads arrive in a standardized format.