Building a salesperson that works weekends

One of the things I like most about building software is that it lets you compensate for personal limitations.

I don’t mean technical limitations. After more than twenty years working in technology, I’ve learned that almost any technical problem can be solved given enough time. I mean a different kind of limitation.

In my case, selling.

I never considered myself particularly good at generating leads, finding clients, or doing outreach. It’s not something I especially enjoy, and it’s not a skill I’ve exercised over the years. My career stayed much closer to building systems, leading teams, and solving technical problems than to selling anything.

Over the last few months, though, I started spending part of my weekends on Blest Learning, a platform focused on corporate English training. And the same problem that hits so many projects showed up almost immediately: having a product is one thing, finding clients is something else entirely.

The question that started circling in my head was fairly simple. If models today can write code, summarize documents, and analyze information, could they also help me with something I personally don’t do well?

I wasn’t thinking about replacing a professional salesperson, and I definitely wasn’t trying to build the next sales-automation startup. I just wanted to know if it was possible. And as tends to happen with most projects born at SparkIO, curiosity mattered more than any business plan.

The first instinct was to build a “sales agent.” But I quickly realized that framing was too broad, because the actual problem was made up of several independent tasks. First, I needed to identify companies that might be a good fit. Then, find some kind of contact there. Then figure out who they were, what they did, and whether they could realistically benefit from something like Blest Learning, or even Blest, our platform for managing English-language institutes. Finally, I needed to put together a reasonable first message and, later, handle the inevitable follow-ups.

What’s interesting is that none of those tasks, on their own, looks particularly hard. The difficulty is volume. Finding a company takes a few minutes. Researching it takes a few more. Finding a contact can take considerably longer. Drafting a personalized email adds another chunk of time. Doing that process a hundred times turns into a project on its own. Doing it a thousand times is basically impossible for someone who also has a full-time job, a family, and other commitments.

That’s where I started experimenting.

Today the system runs in stages, using different tools to discover companies, gather public information, build profiles, draft emails, and suggest follow-ups afterward.

I tried a few options for email validation, including MillionVerifier, ZeroBounce, and NeverBounce. So far NeverBounce has felt like a reasonable place to start, mainly because it lets you buy credits in small volumes without committing to anything bigger. I also learned quickly that validating emails costs money, so it’s worth filtering as much as possible before reaching those services — plenty of clearly invalid addresses can be ruled out with simple validation rules in code. That kind of optimization seems minor, but when you’re funding the experiment out of your own pocket, it starts to matter.

After roughly a week of testing, I managed to build a list of about 150 companies that looked potentially interesting. Of those, I got around 20 valid emails.

That’s not an impressive number. It’s not enough to build a sales machine, either. But it’s far more than I would have managed manually in the same amount of time, and that’s exactly what I find interesting about this.

We often talk about AI in terms of replacement, full automation, or autonomous agents. My experience so far has been much more modest. I’m not looking at a machine that sells on its own. I’m looking at a tool that lets me do something I probably wouldn’t do otherwise. That difference matters.

If I ever get to a database of a thousand companies and a hundred valid contacts, it will still be necessary to build relationships, understand needs, and earn trust. No model is going to do that for me. But getting to that point already represents an enormous amount of administrative work that can genuinely benefit from automation.

I still don’t know if this will end up generating clients for Blest Learning. I don’t know if it will turn into a standalone product. And I don’t know if the response rate will justify the effort.

But that uncertainty is part of why I enjoy these projects.

To me, SparkIO has always been more of a lab than a business — a place to test ideas, learn new technologies, and explore questions that genuinely make me curious. In this case, the question is fairly simple: can someone with technical knowledge build, on weekends, a system that helps them compensate for a skill they never quite developed?

I still don’t have the answer.

But so far, the experiment has been a lot of fun.

Related Sparkio Products

Leads B2B Leads by Spark.io helps companies identify high-potential prospects, enrich lead data, and streamline outbound sales efforts using AI and automation. Currently in early validation, focused on building practical solutions for real business growth

Founder Interactive business history simulator inspired by the stories and decisions behind the world's most influential companies.

AppGrid Platform designed to help independent developers connect their products with potential customers.