My AI Agent Found Investors by Scanning Their Instagram Photos. Here’s the Full Pipeline. – Asian Efficiency

When I started raising money for Paddle Society — a padel club I’m building in Austin — I ran into a targeting problem that no investor database really solves.

My ideal investor wasn’t just any high-net-worth person or any C-level executive. I wanted people who played racket sports. Tennis players, padel players, squash players. Athletes (current or former). People who understood what it meant to show up at a club, build community around a sport, and spend real money on it. That’s the investor who gets the business viscerally.

The problem: you can’t filter for that.

LinkedIn has job titles, industries, geography. Apollo has similar structured data. But “plays tennis on weekends” isn’t a field in any database.

So I built an agent to find it anyway.

The Pipeline: Four Steps

Step one — Pull the contacts. I started with Apollo to find C-level executives in target industries, then cross-referenced with LinkedIn to build a working contact list. This part is straightforward. It’s the data collection layer.

Step two — Scan for sports mentions. The first agent scans every publicly available digital footprint for each contact: LinkedIn bio, Twitter, personal website, any interview or press coverage. If they’ve ever mentioned tennis, padel, squash, or any racket sport anywhere online, they get flagged and scored.

Step three — Analyze their photos. This is the part that surprises people.

Even if someone has never written “tennis” in any public profile — never mentioned it in a bio, never tweeted about it — they might still play every Saturday morning. And if they do, there’s a decent chance a photo of that Saturday morning is sitting on Instagram.

The agent scans their Instagram photos. It’s doing visual reasoning: if this person appears in a photo holding a tennis racket, or appears to be playing on a tennis or padel court, they get flagged as a qualified lead.

That’s the insight that shifts how people think about AI for prospecting. It’s not just about what people say. It’s about what they show up doing in photos.

Step four — Map connections and draft intros. A separate agent logs into my LinkedIn account via a virtual machine — think ChatGPT Operator, a browser running in the cloud. You can watch it clicking through screens like a human would. For each prospect on the list, it scans my LinkedIn connections, identifies every mutual connection I have with that person, and drafts a warm intro request.

The draft might be a short text message to my mutual connection, or an email. Either way, it’s ready to send. I review, personalize slightly if needed, fire it off.

The Cost Breakdown

Total cost to identify, research, and generate intro requests for 100 contacts: roughly $30-40 in LLM credits.

Not $30-40 in staff time. Not $30-40 per contact. Thirty to forty dollars for the whole batch.

Compare that to hiring a VA to research LinkedIn manually, licensing specialized investor databases, or the time it would take to build a list like this by hand. The economics are completely different.

Why This Works as a Pipeline (Not Just a Prompt)

What makes this effective isn’t any single step. It’s the combination.

No one data source gives you what you need. Apollo tells you job titles. LinkedIn tells you what people write about themselves. Instagram tells you what they actually do. Each tool has a different view of the same person.

The agent chains them together. This is what I mean by multi-tool native — routing different parts of a workflow to the tool best suited for it, rather than forcing everything through one platform.

Apollo is good at structured contact data. Visual AI models are good at analyzing photos. A virtual machine is good at authenticated browser navigation. An LLM is good at drafting personalized messages. Each layer does its job.

Beyond Padel Society

Lucas Siegel, co-founder of Yuna (AI mental health platform, 50,000 users across 155 countries), heard this on a call and said he wanted to build the exact same pipeline for Yuna’s sales prospecting. Different product, same architecture: find qualified prospects by combining signals that no single database captures, map the mutual connection layer, draft the warm intro.

That’s the pattern that makes this worth paying attention to. The investor-targeting agent I built for Padel Society wasn’t a one-off — it’s a template for any situation where your ideal prospect has a defining characteristic that doesn’t live in a CRM field.

What does your ideal customer show up doing in photos?

Thanh Pham is the founder of Asian Efficiency and runs the Two Hour Workday program. He helps business owners use AI to build systems that do what spreadsheets and databases can’t.