Using AI to extract information from sales leads

Getting all your leads into some kind of CRM system is crucial if you want to make as many sales to those leads as possible. It's best if this happens automatically - rather than having someone type them in. But this can be tricky, especially if you get leads from more than one source.

A client of mine in the motor trade gets thousands of leads every month from:-

• Google Ads
• Facebook
• Their website
• The various car manufacturers (Ford, Renault, Suzuki etc)
• Used car listing websites e.g. AutoTrader
• And several other sources.

None of these sources add leads directly to their CRM system, instead they deliver them by email.

Until now, I've relied on Zapier's email parser to extract contact information like first name, last name, email address and phone number from the lead delivery email and put it into the CRM.

Here's how it works:-

1. We give each lead source a unique @zapier email address to deliver leads to.
2. We set up a template that shows Zapier how to extract the lead's name, number etc from the message.
3. Zapier sends the extracted contact information to the CRM every time a new leads arrives.

We've handled about 100 000 leads this way. It has generally worked ok - but there have been some frustrations:-

• Some of the lead sources only allow for one email address - usually the login email address, so we can't use the @zapier address. We've worked around this by setting up automatic forwarding to our email parser, but it's clunky.
• Every lead source sends an email in a different format. That means we have to set up, test and tweak a new template every time we add a lead source.
• Some sources use several different email formats e.g. one format if the lead called, another if the lead filled in an enquiry form and something else if they used live chat.
• The lead sources change the format of the lead delivery email, usually without notification. This confuses the system and it can't extract the right information.
• Every now and again Zapier's email parser forgets how to extract the info we need. I have no idea why it happens.

Enter AI...

It turns out that ChatGPT (an AI-based language processing tool) is really good at extracting data from emails. It took a little tweaking to get the instructions right, but after that the AI scored 100% on a test of 1 000 lead delivery emails in various formats.

Here's the non-technical explanation of how it works:-

1. A lead source delivers the lead by email.
2. We send the message to ChatGPT using an API. (An API is a way for 2 computers to talk to each other.)
3. ChatGPT extracts the contact information and sends it back in a standard format and the lead gets added to the CRM.

I was so excited about this technology that I just had to write about it. I hope it's as interesting to you as it is to me.

Related articles

106 quotes. 17 sales.

The data gap between advertising and sales

Why many RCICs don’t know if their Google Ads are profitable

The invisible Google Ads problem: zero-second calls

Why the phone rings but the workshop stays empty

Computer code rusts

Topics you'll find on this site