Finding the right clients with AI agents

Here’s something every salesperson knows in their gut: the pitch is not the hard part. The hard part is figuring out who to pitch to. Get in front of the right company and the conversation almost runs itself. Get in front of the wrong one and the best pitch in the world goes nowhere.

In this post I’ll walk through how I’ve started handing that “who” question to AI agents. When I say AI agent, I just mean a program you can give a goal to in plain English, that then goes and does the busywork for you: opening websites, reading them, and writing down what it found, the same way an assistant with a laptop would. The result: where one person used to research about ten companies a day, we now scan hundreds in a matter of hours.

Let me use one running example. Say we sell a product that companies buy for their staff as a perk, an employee-benefit. There are thousands of companies out there. I have limited hours in the day. Which ones do I actually call?

Why the research is the whole game

Sales is a numbers game with brutal math. Here’s roughly how it goes:

Reach out to 100 companies, maybe 10 reply, 5 or 6 show real interest, and 1 or 2 actually buy. So to close a couple of deals, the top of that funnel has to be wide. You need those 100 good companies in the first place.

And that’s the trap. If finding each good company takes you real time, then all your time goes into filling the top of the funnel, and you never get to the part that pays. Worse, if the research is rushed, bad fits slip through, and you burn your best hours on meetings with companies that were never going to buy. So the question isn’t “how do I pitch better”, it’s “how do I find the right 100 companies without it eating my whole week”.

Who’s even worth calling?

Before any tools, just think it through like a person would. For an employee-benefit product, the companies most likely to buy are the ones that already care about keeping their staff happy, and can afford it. A few signals point at that:

• Do they treat their people well? You can read this off employee-review sites like Glassdoor and AmbitionBox (both are sites where current and former employees rate their company), and off places like Reddit where people talk candidly about where they work.
• Did they just raise money? A company that recently got funding is often thinking about growing the team and keeping people from leaving, which is exactly when a benefits product lands well.
• Do they pay enough? If a company’s salaries are very low, our product may be out of their range. You can sanity-check pay from Glassdoor, from levels.fyi (a site where employees share their salaries), or from what people say on Reddit.
• Where do the good people go? A neat shortcut: pick a few top colleges, and see which companies their graduates end up at. That alumni trail is a ready-made shortlist of desirable employers to start from.
• Does leadership talk about it? If a company’s HR head (the CHRO) posts on LinkedIn about employee wellbeing, that’s a strong sign they’d care about what we sell, and a clue for how to reach them.

Once you’ve got a shortlist, there’s a second question: who do you actually contact, and how? If the founder or HR head is active on LinkedIn, that’s often the way in. If not, a cold email might be the fallback.

The catch: doing this by hand is painfully slow

Now line that up against the clock. To do the research above for one company properly, checking the review sites, cross-referencing pay, hunting down the HR head, I’d spend at least half an hour. Realistically, one person gets through about ten companies in a day.

Remember the funnel: I need 100 good companies at the top to close one or two deals. At ten a day, that means basically 100% of my time goes into research, and none into selling. That’s the wall. It’s not that the work is hard, it’s that there’s too much of it for one person with one laptop.

And it’s not just slow, it’s uneven. When a human does it, the quality swings: a careful afternoon, a rushed evening, a different person with a different idea of a “good fit”. In our case that inconsistency was the expensive part: it put us in meetings with the wrong companies.

The trick: it’s really just a table

Here’s the mental model that makes the whole thing click. Everything I just described is really one big table. Each row is a company. Each column is one thing I want to know about it: what they pay, whether they treat staff well, whether they just raised money, who to contact.

For the example, picture a made-up mid-sized Chennai company, let’s call it Acme Co., rather than a giant like Google (a company that big is the wrong place to start: too sprawling, and every office is different).

Filling in one cell is a small, boring, well-defined task: “go find Acme Co.’s Glassdoor rating.” Do that for every cell and the table fills up. And “small, boring, well-defined” is exactly the kind of task you can hand to an AI agent.

Handing the table to agents

So instead of me filling the table by hand, I set up agents to do it. Two ideas make it work.

First, I write down, once, how to research a single company: which sites to check, what to look for, how to weigh it. Think of it as a recipe, written in plain English, that anyone (or any agent) could follow. In agent-speak this reusable recipe is called a skill, but “recipe for researching one company” is all it means.

Second, I use one agent as the manager. It takes my list of companies and hands each one to a fresh worker that follows the recipe. That manager-agent is called an orchestrator, which is just a fancy word for the agent that farms the work out and collects the results.

The output is one note per company, and stacked together they become the filled-in version of that same table:

All of this lives in Obsidian, a note-taking app. Each company is a plain note, and Obsidian shows the whole folder of notes as one table, so the agents and I are looking at the same thing.

Scoring without pretending it’s math

One thing I want to be honest about: the final score can’t be a tidy formula. “Employee happiness” isn’t a number you multiply by funding. It’s a judgment call. If I were doing it by hand, I wouldn’t compute a score, I’d read everything and sort each company into a rough bucket: strong fit, maybe, or not worth it.

So that’s exactly what I have the agent do. It reads the evidence it gathered and sorts each company into High, Medium or Low, the same call I’d make, at a scale I never could. That’s why the table above shows buckets, not a made-up number out to two decimal places.

Why this is worth it

Here’s the payoff. Before, one person could research about ten companies a day, and the quality wasn’t consistent, which meant wasted meetings with the wrong companies. With agents running in parallel, we scanned hundreds of companies in a matter of hours, every one researched the same way. The only real limit now is how much I want to spend on running them, not how many hours are in my day.

Two things make it genuinely better than just hiring someone to do the research:

• It thinks at my level. The recipe carries my judgment about what a good fit looks like, so every company gets researched the way I would do it, not filtered through a new hire still learning the ropes.
• I can tune the strategy by editing one file. Say I notice Glassdoor is thin for smaller companies but AmbitionBox is rich. I just open the recipe and change that line. Every future company is researched the new way, instantly. No retraining, no meeting.

That’s the whole idea. The slow, unglamorous part of sales, finding the right companies to talk to, is really just a big table waiting to be filled. Write down how you’d fill one row, then let agents fill the rest while you get on with actually selling.