Audience overlap is not the same as audience fit
Partnership lists are built from follower counts and category adjacency. That produces a partner whose audience looks similar to yours on paper but has no reason to care about the products you sell. The co-marketing campaign generates no conversions, and both parties lose a quarter.
Intent is what has to overlap. Two audiences can share every demographic attribute yet be in completely different buying moments. You see that only by reading what each audience actually engages with.
Another question that people do not ask early enough is what advantage the partner receives. A partnership that offers an imbalance of benefits does not fail at launch. It fails silently by the third email.
The assessment should also include adjacency risk. A partner who is only one product decision away from competing with you is a different situation from one who could never do so.
How the agent is built in Agent Studio
The system prompt states the fit criteria as intent overlap, not demographic similarity. If it does not, the agent defaults to the easier comparison.
Prompt skills include what you can offer a partner and what you need from one, so the case addresses both directions, not only your own.
A Scrappey node and an HTTP Request node gather public material. A flow assembles it and passes it to the agent through the Agent Chat node. A Loop node walks a candidate list when you are screening in volume.
Each candidate gets one brief case, including the candidates you recommend against, with reasons. Those negative entries take the most time.
Use only public information, as specified in the system prompt.
A person always owns outreach decisions. The agent runs the screening and does not make contact. Do not configure it to contact anyone.
What this agent is built from
- System prompt: states fit criteria as intent overlap, not demographic similarity.
- Prompt skills: state what you can offer a partner and what you need from one.
- Scrappey (Util): gathers public material. HTTP Request covers the rest.
- Loop (Control): walks a candidate list. Agent Chat (Util): screens each candidate one at a time.
- Notion (Integration): receives one case per candidate, including the negative ones.
- Human decision: the agent screens and never contacts anyone.
- When an ActionFlow is enough: Scraping a directory into a candidate list is a workflow. Screening is the agent's job.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.