We’re excited to announce our newest feature for spam protection called “Lead Scoring”. Simply pass your business context in plain English along with your API calls and get all the power of modern LLMs in your API response.
This allows you to define any requirements and scoring system for your spam checks, per API call. LLMs are not yet a complete replacement for traditional spam checking techniques. We’ve found a combination of existing heuristics, machine learning, and filtering is best. These datapoints are also forwarded to the LLM to provide full context and achieve a best of bost worlds approach.
Here is an example prompt and response we’ve found to work but of course you’re free to come up with something that works for you! Come up with something that works for you. The score threshold is completly arbitrary and allows us to respond with a simple above_threshold boolean for ease of integration with your stack.
Example Request:
Context: This is a submission from a lead form for a Real Estate B2B SaaS in search of enterprise customers. The total score should be above 80 only for high quality leads interested in purchasing. The score should be low for spam and promotions.
Score Threshold: 80
And this is a real response:
Above Threshold: False
Score: 15Reason: This lead has a very low score because the message appears to be a promotional or spam message promoting a marketing SaaS rather than a genuine inquiry about a Real Estate B2B SaaS product. The content is highly promotional, unsolicited, and lacks specific interest in real estate solutions, indicating low intent and quality. The email and message content suggest a marketing spam. The location (Stockholm, Sweden) does not significantly impact the score, but confirms the lead is from a European country, which is common for high-value enterprise B2B leads. Overall, the lead’s focus on promotion and lack of targeted details result in a low score of 15.
Software 3.0 is a term coined by Andrej Karpathy. The future of software is plain English.