About the client
Polar Night Software developed LeadSense, an AI-based platform designed for prospecting. LeadSense offers automatic analysis of company websites to identify potential customers using a range of customizable markers relevant to each client. The intelligent platform supports businesses with comprehensive insights and improves prospecting process by providing an efficient, automated way to discover high-quality leads. With LeadSense, organizations can customize their search criteria and refine their strategies to maximize customer acquisition.
Problem
Prospecting for niche B2B segments took too long and yielded inconsistent lead quality. Off-the-shelf tools couldn’t capture the nuanced signals we cared about, and manual research didn’t scale across regions.
Approach
We treated LeadSense as an R&D case with real usage. We defined clear buyer-fit signals, built an incremental pipeline that discovers companies by region, analyses websites with LLMs using templated prompts, and generates short, structured summaries for outreach. A human-in-the-loop review fed corrections back into prompts and thresholds. We instrumented cost, latency, and freshness so the system could run continuously under provider limits.
Outcomes
Consistent coverage across target geographies without blind spots
Structured lead summaries aligned to buyer criteria, ready for outreach
Shorter research-to-outreach cycle compared to manual prospecting
Lower duplicate rate via stable identity keys and idempotent jobs
Operational guardrails for quotas, retries, and cost visibility
Project overview
Partnership Goal
Prove a repeatable, AI-assisted prospecting approach we can tailor for clients - reducing manual research while keeping quality under human control.
Context & goal
We wanted a faster path from discovery to first outreach - without sacrificing relevance.
Internal experiments showed that generic tools missed domain-specific signals. We framed LeadSense as an in-house pilot to codify those signals and turn them into a steady, auditable pipeline rather than one-off lists.
Success meant reliable coverage, clear summaries, and a feedback loop that improves quality over time.
Decisions that shaped the MVP
Start narrow, instrument everything, and ship in safe slices.
We tiled regions to avoid gaps, normalised identities to prevent duplicates, and analysed sites with prompt templates bound to our buyer criteria. Each stage exposed metrics - time, cost, freshness - so we could tune batch sizes and backoff without breaking flow.
Summaries followed a fixed schema, which made review quick and downstream tools easier to integrate.
Human-in-the-loop impact
Quality rises when humans correct the edge cases the model can’t see.
Reviewers flagged misclassifications and requested re-checks; those signals updated prompts and thresholds rather than becoming one-off fixes. Over time, less manual cleanup was needed and the research-to- outreach cycle shortened.
With guardrails in place, the pipeline runs continuously and stays within provider limits.
How our team turned ideas into a working, scalable solution
What we worked on
Services
Technologies
Have a similar project in mind?
This case study resonated with your needs? We'd love to discuss how we can apply our expertise to solve your unique challenges and deliver exceptional results.
Aku Kajan
Commercial Director
Ready to stay ahead?
Let's explore how we can transform your business.
Get in touch via email
We'll reach out to discuss your needs. No spam!