Expansion Signal Spotter

Expansion Signal Spotter

A weekly scan over a customer account list. The keyless sweep + sub-skill signal sources are deterministic; signal aggregation, the expansion play, and the talk track are the agent's.

When to use

• "Which customers are ready to expand?" / "Find upsell opportunities in our accounts."
• "Run the weekly expansion signal scan." / "Who should I pitch [new tier] to?"

How to run

Step 1 — filter accounts + keyless web sweep

python3 ${SKILL_DIR}/scripts/account_signals.py \
  --accounts ${WORKSPACE}/accounts.csv \
  --min-account-value 1000 \
  --exclude "ChurnCo,PausedInc" \
  --output ${WORKSPACE}/signals.json

Filters by min value / exclusions up front (exclude churn-risk/paused/disputed accounts to keep the list actionable), then sweeps each account for hiring, leadership-change, and public-news hits.

Step 2 — add job + funding signals (sub-skills)

For each account, run job-posting-intent / job-scraper over the account domain (new relevant roles) and funding-signal-monitor (funding announcement). Use PHANTOMBUSTER_API_KEY (+ cookie) to monitor key_contacts profiles for promotions if available. Merge these into each account's signal_stack.

Step 3 — aggregate + write plays (you, the agent)

Aggregate signals per account; multi-signal accounts rank highest. Per opportunity, fill signal_stack, expansion_play, talk_track, and rank. Render the weekly list and persist (store prior-week snapshots so "new since last scan" is computable; degrade to a single scan if no durable store).

Outputs

signals.json — [{company, domain, tier, account_value, primary_contact_linkedin, web_signals{hiring,leadership,news}, signal_stack, expansion_play, talk_track, rank}].

Credentials / env

• Required: none — the serp backbone runs keyless.
• Optional: APIFY_API_TOKEN (job-posting/LinkedIn-scale signals); PHANTOMBUSTER_API_KEY + cookie (LinkedIn profile monitoring of key contacts); SUPABASE_URL + SUPABASE_SERVICE_ROLE_KEY or an Airtable key (week-over-week history — degrades to a single scan); ANTHROPIC_API_KEY only if the talk-track LLM isn't platform-provided.

Notes & edge cases

• Multi-signal accounts are the strongest — weight them above single-signal hits.
• Store prior-week snapshots so "new since last scan" is computable; degrade to a single scan without a durable store.
• Proxy + throttle LinkedIn; respect Apify costs (optional).
• Exclude churn-risk/paused/disputed accounts up front (the --exclude filter).