Founder-led build · AIBYTERS · Sales operations
LeadHarvest: from manual prospecting to a scored pipeline
How I turned slow, inconsistent prospect research into a standard pipeline (search, enrich, validate, score) that hands the sales team a prioritized, ready-to-call list.
01The problem
Building prospect lists by hand is slow, inconsistent, and hard to hand off. Someone searches Google Maps, opens each business website, hunts for an email or WhatsApp number, pastes it into a spreadsheet, and then guesses which leads are worth calling first.
Quality depends on who did the research that day, and results skew toward wherever the researcher's browser thinks they are.
02My role
- Mapped the manual research process step by step and redesigned it as a standard pipeline: search → enrich → validate → score → export.
- Defined the lead-quality scoring rules so that prioritization is consistent, not a judgment call.
- Designed the output around how a sales team actually works: separate lists by priority and contact channel.
- Built and iterated the tool from v1 to v2 based on speed, accuracy, and reliability gaps found in real runs.
03What was delivered
- Location-accurate search. Each city and sub-area (e.g. Gulshan, Banani) is geocoded separately, so results come from the right neighbourhood wherever the tool is run.
- Contact enrichment. Each business website is checked for email, Facebook, Instagram, and WhatsApp, with a fast lightweight request tried before a full browser.
- Validation. Worldwide phone-number validation adapted to the target country, plus email filtering that removes false positives and prefers business domains.
- 0–100 lead score. Valid phone +30, valid email +25, real website +20, Facebook / Instagram / WhatsApp +10 each, rating ≥ 4.0 +5, 10+ reviews +5.
- Sales-ready Excel output with separate sheets: All Leads, High Priority (≥ 70), Standard (40–69), Phone Only, Email+, and Rejects (< 40).
- Two modes: keyword search ("hotels in Dhaka") or a CSV of named companies to enrich an existing account list.
04How it improved (v1 → v2)
- 3–5× faster, with up to 4 browser pages running in parallel.
- No lost work. Progress is checkpointed after every lead (previously every 5), and re-running the same command resumes automatically after a crash.
- Fewer bad contacts thanks to stricter email filtering and country-aware phone validation.
- Stable long runs. Randomized pacing between requests avoids rate limits.
05Results
The real win is operational: prospecting went from a skill that lived in one person's head to a repeatable process anyone on the team can run.
Let's find the bottleneck in your operations.
A 30-minute call is enough to map where work gets stuck and agree on a practical first step.