enriching-conference-companies

Turn a list of conference companies, sponsors or exhibitors into named decision-makers with LinkedIn profile URLs and verified email addresses. Use after scraping a conference roster, or whenever you

SKILL.md

GitHub

Enriching Conference Companies

Goal

Turn "186 sponsor logos" into "the person at each company whose job is the thing I sell, their LinkedIn, and an address that won't bounce."

When to use

  • Straight after scraping-conference-rosters
  • Any time you have company names but no contacts
  • "Who do I talk to at these companies?"

The four layers

Cheapest first. Each one is independently skippable.

# Layer Tool Cost
1 Company website Tavily free
2 Decision-makers by role Tavily free
3 LinkedIn profile URL Exa free
4 Email + SMTP verification Apify paid, off by default

Layers 1–3 are free tiers with no card. Layer 4 is the only one that can surprise you with a bill, which is why it needs --emails explicitly.

Tavily answers "who runs marketing at this company" the way a search engine would. Exa is embedding-based and much better at "which LinkedIn profile is this exact person" — the step where a keyword engine hands you the wrong Sarah Chen with total confidence.

Process

Step 1 — always start with a limit

python -X utf8 -m scripts.enrich_conference_companies \
    --input data/example-sponsors.csv \
    --output data/example-enriched.csv \
    --roles "VP of Marketing" "Chief Marketing Officer" \
    --limit 10

Read those 10 rows before going further. If the names look wrong, the roles are wrong.

Step 2 — set the roles to match what YOU sell

This is the flag that decides whether the output is useful:

You sell --roles
Video, brand, web design "VP of Marketing" "Head of Brand" "Chief Marketing Officer"
Compliance, legal tech "General Counsel" "Chief Compliance Officer"
Recruiting "Head of Talent" "VP People" "Chief People Officer"
Clinical software "Chief Medical Officer" "VP Clinical Operations"
Sponsorship, events "Head of Events" "VP Partnerships"

Default is marketing leadership. Change it.

Step 3 — the full run

python -X utf8 -m scripts.enrich_conference_companies \
    --input data/example-sponsors.csv \
    --output data/example-enriched.csv \
    --roles "VP of Marketing" "Head of Brand" --workers 4

Step 4 — emails, when you're ready to spend

python -X utf8 -m scripts.enrich_conference_companies \
    --input data/example-sponsors.csv --output data/example-enriched.csv \
    --roles "VP of Marketing" --emails --verify-emails --limit 25

Prints a cost estimate before it starts. Always verify before a first send — bounces damage domain reputation and it is slow to rebuild.

Reading the output — this matters

The confidence column is doing real work here. Do not ignore it.

confidence What it means Do this
confirmed LinkedIn URL verified — the surname matches the profile slug Reach out
probable Name found, LinkedIn unverified Check the profile before sending
weak Name found, no LinkedIn match Verify manually first
— with next_action: research-manually No confident contact at all The booth is still worth visiting

Check evidence for the title. It says either title from source (the source stated their real title) or title unconfirmed; matched on search for '...' (the title column is the role we searched for, not a confirmed fact). Never quote an unconfirmed title back to someone.

Three things it deliberately will not do

It will not attach a LinkedIn URL it can't verify. If the surname isn't in the profile slug, the cell stays empty. A wrong LinkedIn becomes a message to a stranger — worse than an empty cell, because a full-looking row never gets checked.

It will not invent a name. Names come from search-result titles and pass a strict shape test with a stopword blocklist. Mining prose produced contacts called "combine design". Conservative on purpose: a missed contact costs one row, a fabricated one costs credibility at a company you're courting.

It will not silently drop companies. No confident contact means the row stays, flagged. Dropping it would overstate your coverage.

Edge cases

  • "No TAVILY_API_KEY set" — free at tavily.com, 1,000/month, no card.
  • Everything comes back weak — usually no EXA_API_KEY. Exa is what confirms LinkedIn matches. Free at exa.ai.
  • Names look wrong for a company — its name is probably a common word and search is matching the wrong thing. Enrich it by hand.
  • All keys rate-limited (Tavily 432) — often temporary rather than a hard cap. Wait, then retry; keys rotate automatically.
  • Big lists — start with --limit 25. 186 companies × 3 roles is 558 searches.

Cost

Layers 1–3: free. Layer 4: roughly $0.02 per contact looked up, plus verification.