The problem this solves
The clearest description of an ideal customer is the set of customers who already renew, expand and refer. Turning that intuition into a list is where the work stalls: someone exports the ten logos, guesses at the attributes that connect them, and searches a database with filters that approximate the pattern. The list that comes back is a caricature of the ten - right industry, roughly the right size - and it misses the quieter attributes like the stack they share or the growth stage they were in when they bought.
Fit scoring is what makes a lookalike list usable by a team rather than one analyst. A rep needs to know which of the 200 to call first and why, and a marketer needs to know which segment to build an audience from. A score without reasons is a number nobody trusts; a score with the three attributes that drove it is a decision the whole team can act on.
How the mission runs
- Profile the seed customers. The Prospecting Agent resolves each of your ten domains and pulls firmographics, technographics, funding stage and hiring pattern for every one. It then reads across the ten for what they share: the industries, the headcount range, the technologies that recur, the stage at which they bought. The shared pattern becomes the working ICP, shown to you with the attributes ranked by how consistently they appear.
- Expand with similar companies. The similar-companies capability takes each seed and returns the companies most like it, and the agent merges the ten result sets into one pool, removing your existing customers and anything already in the seed. The pool is intentionally larger than 200 so the scoring step has room to be selective.
- Score every candidate against the pattern. Each candidate is scored from 0 to 100 on the attributes the seeds share, with a weight per attribute that reflects how universal it was among the ten. The top three reasons are written into the row, for example 'runs <technology>, 120-400 employees, raised Series B in the last 18 months'.
- Save the ranked list and mirror it to Sheets. The top 200 are saved as list <name> in score order and copied into a Google Sheet with the score, the reasons and the standard company columns. The list keeps its criteria, so it can be refreshed as new companies start to resemble your customers.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
List <name> holding 200 lookalike companies ranked by a 0-100 fit score, each row carrying its top three reasons plus domain, industry, headcount, HQ, key technologies and funding stage, mirrored in a Google Sheet. A short summary explains the ICP pattern the agent derived from your ten customers, so the scoring logic can be corrected before the list is used.
Make it yours
- Seed with your 25 best customers and your 10 worst churns, and ask the agent to score against the difference between them.
- Restrict the pool to one region or one headcount band when territories are already fixed.
- Add people: the VP of <function> at every company scoring above 70, with verified emails, pushed to HubSpot.
- Turn the top 500 by score into a LinkedIn company-matched audience through the audience playbooks.
Frequently asked questions
How does the agent decide what the ten customers have in common?
It compares every firmographic, technographic and funding attribute across the seeds and keeps the ones that recur in most of them. You see that pattern before the scoring runs and can promote or demote attributes, for example telling it that the shared technology matters more than the industry.
Can I change the weights after seeing the results?
Yes. Adjust the weights in the thread and the agent re-scores the pool without re-running the search, so a second ranking takes seconds and costs almost nothing.
Does the list include my current customers?
Current customers and the seed companies are excluded automatically. Connect your CRM and the agent uses its customer records for the exclusion; otherwise attach a CSV of customer domains.