Kol Discovery

KOL Discovery

Build a ranked KOL list by combining a post-data half (LinkedIn post search -> aggregate by author -> deterministic engagement scoring) with a web-research half (the agent finds conference speakers, newsletter/podcast hosts via web search). The agent does keyword generation and web research; the bundled scripts do the post fetch and the scoring math.

When to use

• "Find KOLs / influencers in [industry]" / "who are the thought leaders for [domain]?"
• "Run KOL discovery for [client]" — authority voices, not complainers.
• Upstream of linkedin-message-writer / linkedin-outreach for influencer engagement.

How it works (agent + scripts)

1. (agent) Intake domain, audience, seeds, exclusions, scoring thresholds. Generate 15-25 authority/thought-leadership keywords (industry terms, "future of...", conference/ event terms, content-creator signals) + KOL title keywords + vendor-exclusion patterns. Present for approval before running (controls Apify cost).
2. Post search — for each approved keyword, pull LinkedIn posts (bundled post_search.py, same as linkedin-post-research: Apify if APIFY_API_TOKEN set, else keyless serp degrade). Concatenate all keyword results into one posts JSON.
3. Score — aggregate by author and rank (bundled score_kols.py).
4. (agent) Web research — find known KOLs (top-[industry]-influencer searches, conference speakers, newsletter/podcast hosts). Write them as web_kols.json ([{name, linkedin_url, source, notes}]).
5. Merge — pass --web-kols web_kols.json to score_kols.py; overlaps are flagged source: both (the strongest KOLs); web-only entries appear as source: web-research.

How to run

Run a post search per keyword and collect into one file, then score:

# Step 2 — collect posts (loop keywords; append into posts.json yourself, or run per batch)
python3 ${SKILL_DIR}/scripts/post_search.py \
  --keywords "future of rpa,agentic automation,intelligent automation" \
  --max-items 50 --sort-by relevance --output json > ${WORKSPACE}/posts.json

# Step 3+5 — aggregate, score, merge web-researched KOLs
python3 ${SKILL_DIR}/scripts/score_kols.py \
  --posts ${WORKSPACE}/posts.json \
  --web-kols ${WORKSPACE}/web_kols.json \
  --min-posts 2 --min-total-engagement 50 --top-n 50 --output csv

Outputs

Ranked KOL list: {rank, name, linkedin_url, headline, kol_score, total_posts, total_reactions, total_comments, avg_engagement, top_post_url, top_post_preview, source(post-data|web-research|both)}.

Credentials / env

• env.required: none — the serp post-search degrade + web research backbone need no key.
• env.optional: APIFY_API_TOKEN — if set → LinkedIn posts-search actor (accurate engagement metrics for the post-data scoring half); else → keyless serp (the default; no engagement counts → weaker scoring, lean on web-research KOLs).

Notes & edge cases

• Search authority keywords, not pain-language — the goal is voices who shape the conversation.
• Always run a small/test pass first (few keywords) before the full run to control Apify cost.
• Too many irrelevant authors -> tighten domain keywords + exclusions; too few -> lower --min-posts / --min-total-engagement. Overlaps (source: both) are the strongest KOLs.