Back to blog
use-caseresearchsales

Research 50 companies in one run

3 min read

You have a spreadsheet of 50 companies. You need to know which ones are worth reaching out to. Manually researching each one takes hours.

One workflow run does the whole batch.

What you'll build

A 3-step workflow:

  1. Research agent reads your company list, looks up each one via web search, gathers funding data, employee count, tech stack, and recent news
  2. Analysis agent scores each company on fit criteria you define (size, funding stage, industry match)
  3. Report agent generates a ranked CSV and executive summary

The fast way

Go to Create and describe the pipeline:

I have a CSV of 50 companies. Research each one for founding year, employee count, total funding, and recent news. Score them 1-10 on fit for a B2B SaaS selling to mid-market companies. Output a ranked CSV and a one-page summary of the top 10.

Svortie builds the workflow and agents for you. Before you hit Run, open the first step in the workflow editor and attach your CSV in the agent's advanced settings. Generated agents know to expect a file but don't have it yet.

The manual way

Step 1: Prepare your input. A CSV with at minimum a company_name column. More columns (domain, industry) help the research agent find the right company faster. For small lists (under 20), paste the company names directly in the task prompt. For larger lists, include the CSV contents in the research agent's task description.

Step 2: Create the research agent. Use the Research Agent template. Web search is built in. Task override:

For each company in the list, research: founding year, employee count, total funding, last funding round, key products, and recent news. Write results to /home/daytona/shared/research_raw.csv. Include a source_url column for verification.

Step 3: Create the analysis agent. Task override:

Read /home/daytona/shared/research_raw.csv. Score each company 1-10 on: team size (prefer 50-500), funding stage (prefer Series A-C), product relevance to [your product category]. Write scored results to /home/daytona/shared/scored.csv.

Step 4: Create the report agent. Task override:

Read /home/daytona/shared/scored.csv. Sort by score descending. Write the top 20 to /home/daytona/output/top_leads.csv. Write a 1-page executive summary to /home/daytona/output/summary.md covering: total companies analyzed, score distribution, top 5 standouts with reasoning.

Step 5: Wire it up. Create a workflow with these three agents in order.

Run it

Hit Run. A similar 6-agent competitive research workflow completed in ~16 minutes using all Haiku agents for 176 credits total. Your 3-agent version should be faster and cheaper.

Check the run detail for exact per-step credit breakdown.

Download top_leads.csv and summary.md from the run artifacts when it finishes.

Making it repeatable

Attach a playbook to the research agent with source verification rules ("always include a source URL," "prefer company websites and Crunchbase over news articles"). Output quality becomes consistent across runs.

Schedule it weekly with a fresh company list for automated lead enrichment.

Variations

  • Hiring signals: research job postings instead of funding
  • Tech stack identification: focus on technologies used
  • Competitive analysis: research your competitors' customers
  • Event-driven: trigger via webhook when a new list lands in your CRM
Back to all posts