Lookalike Expansion

Two-phase workflow: Scout (quick 25-company preview) → Expand (full scale, up to 20K).

Always scout first. Expanding without validation wastes time on a bad seed profile.

Hard rules

1. Scout before you expand. Run a quick preview first unless the user explicitly says to skip it.
2. Confirm scale before running. Full expansion can return up to 20K companies — confirm how many the user wants.
3. Landbase only. Do not fall back to external sources if results look thin. Flag coverage gaps instead.
4. Never truncate the output dataset. If the user wants a subset, filter after downloading.

Phase 1 — Scout (quick preview)

Use when: the user has example company domains and wants to validate the direction before committing to a full run.

landbase-cli search "find companies similar to [domain1.com] and [domain2.com]"

Returns top 25 pre-computed lookalikes. Fast, no dataset needed.

After scouting: show the user a few representative results (name, industry, size, location) and ask:

• Do these look right?
• Any firmographic constraints to add (industry, geography, size)?
• Ready to expand at scale?

If the scout looks wrong: ask the user to revise the seed companies or add firmographic guidance before proceeding. Do not expand from a bad scout.

Phase 2 — Expand at scale

Step 1: Get a dataset ID

Expansion requires a ds-XXXX dataset ID. Three paths:

A. User has a CSV of seed companies — upload it:

landbase-cli upload seed-companies.csv
# Returns a dataset ID, e.g. ds-abc123

B. User ran a prior search — they already have a dataset ID from that run. Ask them for it or find it:

landbase-cli runs latest

C. User wants to search first, then expand — run a search, capture the dataset ID from the output, then proceed.

Step 2: Run the expansion

This is an interactive flow, so trigger without --wait, capture the workflow_run_id, and follow the workflow-monitor skill to narrate live progress — a --wait subprocess can’t stream its progress line into the chat. The examples keep --wait for compactness; drop it when monitoring live, and keep it only for scripted/CI runs or when the user just wants the final result.

landbase-cli workflow similar-company-expansion ds-XXXX --wait

With firmographic steering (use when the seed companies span multiple industries and you want to narrow):

landbase-cli workflow similar-company-expansion ds-XXXX \
  --firmographics-guidance="keep only B2B SaaS companies, US only" \
  --wait

With a non-default website column (only needed when the dataset came from workflow onboard, where columns are prefixed with CLI_):

landbase-cli workflow similar-company-expansion ds-XXXX \
  --website-column=CLI_website \
  --wait

Step 3: Deliver results

If you used --wait, the JSON result carries the new dataset ID. If you monitored without --wait, you only have the run id and the input dataset id — once the run reaches SUCCEEDED, get the output child dataset id with datasets lineage:

landbase-cli datasets lineage ds-XXXX --direction=children --workflow=similar_company_expansion
# Returns [{ "id": "ds-YYYY", ... }] — child ID is at [0].id

Then download it:

landbase-cli datasets download ds-YYYY lookalikes.jsonl

If a Drive MCP is available, push to Google Sheets and return the live URL. Otherwise deliver the file and note the path.

Step 4: Suggest AI qualification

After delivering the expansion results, suggest running AI qualification to filter the lookalikes against ICP criteria the lookalike engine can’t apply on its own — funding stage, tech stack, recent signals, etc.:

landbase-cli workflow qualify ds-YYYY \
  --qualification-prompts="<yes/no ICP question>" \
  --context-cols=company_name,website,industry,description \
  --wait

Use the qualify-leads skill for the full guided workflow. Widening with lookalike expansion then qualifying typically yields 50%+ more relevant results than filtering by keywords alone.

When to use firmographic guidance

Provide --firmographics-guidance when:

• The seed companies are from multiple verticals and you want only one
• The user wants geographic constraints (e.g. US only, EU only)
• The seed companies include very large and very small companies and you want to narrow size range

Keep the guidance concise and specific: "B2B SaaS, 50-500 employees, US only".

Flags reference

Red Flags — STOP

• About to run workflow similar-company-expansion without scouting first → go back to Phase 1
• Scout results look unrelated to what the user asked for → fix the seed before expanding
• User hasn’t said how many results they want → ask before running (20K is the max)
• About to use a non-Landbase source because results look thin → flag the gap instead