How to score and qualify leads with LLMs

Background

The launch of our newest LLM, Refuel-LLM-2 came with good news and bad news. The good news was that we had more inbound and demand than we knew what to do with. The bad news was that not all inbound was equal.
‍

It quickly became apparent that we needed a mechanism to qualify and filter out the leads that were not the best fit for Refuel. However, setting up a full blown CRM would be next to impossible at the time/effort needed to set up a such a system.
‍

We quickly realized that we needed a workflow for automatically vetting and qualifying leads.

Challenges with manual lead qualification

‍
Maintaining our status quo, and continuing with manual lead qualification, would have multiple cascading consequences.
‍

• We risked outgrowing our lead qualification frameworks (BANT, MEDDIC, CHAMP etc.) as our company and market evolved
• The time required to manually vet, score, and qualify each lead would be a huge burden
• Manually qualifying would be next to impossible during new launches and times of high demand
• Using heuristics based qualification could lead to inconsistent lead qualification
• Ultimately, the time and manual effort spent researching leads meant that we weren’t spending that time with customers that we were uniquely qualified to help.
  ‍

How to score and qualify leads with AI and LLMs

We realized that Refuel would be the ideal solution for our challenge at hand.
‍

1. We uploaded a dataset of our historical inbound submissions

‍

2. In natural language, we defined a task that outlined what the inbound submission looked like, and rules that defined a good lead and a bad lead. Refuel was also configured to search the Internet for the lead and enrich with additional data.
‍

‍

3. We ran the model and provided feedback on the outputs to improve prompting. Refuel uses few-shot prompting to improve the prompt quality as more data gets passed through while human feedback is added in the loop.
‍

‍

4. We deployed the model as an endpoint
‍

‍

5. We then connected the endpoint with Zapier
‍

‍

6. We now get pinged any time we receive a qualified lead, along with a confidence score!
‍

‍

What are the outcomes of LLM based lead scoring and qualification?
‍

1. We were able to reduce ~2 hours/week spent researching and qualifying leads
‍

2. We eliminated bias and ensured consistency in our lead qualification methodology
‍

3. The entire lead qualification process was automated, and plugged into our existing work tools (Slack)
‍

4. We are able to quickly and seamlessly adapt our qualification methodology and approach based on the state of the business
‍

5. We were able to source additional data points on the lead, beyond what they filled out in the inbound form. Many companies struggle with finding the right balance between collecting enough data points on the inbound lead without adding too much friction to completing the form. The approach found the right balance between the two

‍