Building Tools Inside an Agency

I didn't set out to build software. I set out to run a marketing agency.

But after a few years of enterprise work I kept hitting the same wall. The tool we needed didn't exist, or the closest thing was overbuilt, underbuilt, or built around somebody else's workflow. At some point you stop waiting.

The first one was a dashboard for multi-location client data. Then an automation layer for review management. Then a reporting framework that showed the numbers clients actually asked about instead of the ones that were easy to export. None of them started as a product idea. Every one started as something breaking on a Tuesday.

Agencies are in a strange, good position for this

Startup convention says build the product, then find customers. An agency has that backwards, which turns out to be an advantage.

You have the customers first. You're inside their operations, watching the same problem recur across accounts, in a position to notice which parts are specific to one client and which are structural.

That's a level of understanding that's hard to buy. You don't need discovery interviews to learn that an operator doesn't want a general dashboard — they want to know which locations are slipping, which listings are wrong, and what to do first. You know that because you were the one making the list by hand.

The second advantage is a real testing environment. Everything we build runs on live accounts with live data long before we'd call it a product. SiteScore was an internal audit script for months. By the time it looked like software, the edge cases had already found us.

The third is boring and decisive: agency revenue pays for it. Nothing has to justify itself against somebody else's fund timeline. We can build slowly and correctly, which is a genuine luxury.

What came out of it

SiteScore — automated site auditing across performance, SEO, accessibility and content quality. We were doing these by hand for every prospect and losing hours to it.

AdPulse — campaign monitoring and anomaly detection. Across enough accounts and campaigns, you need something watching pacing and performance drops while you're asleep.

BrandMap — multi-location brand consistency tracking. With 21 locations and 80-plus directories each, NAP drift is constant and invisible until it costs you rankings. BrandMap watches for it.

The n8n content pipeline — research, drafting, editing and publishing as one operated system. This is what let us ship 40-plus long-form posts without the team hating their jobs.

Same pattern every time. Hit the pain, build the fix, notice later that the fix had value on its own.

The part that's genuinely hard

Agency work is service revenue — time and expertise. Tool work is leverage — build it once. Those two things compete for the same people, and the competition is real, not a framing problem.

Three things I've learned about holding them together.

Don't starve the agency. It funds the tools and it generates the feedback that makes them good. Pull too many people onto product work and you lose the clients who made the tools worth building.

Build for your own delivery first. Every tool we've kept makes our actual work better — SiteScore made prospecting faster, BrandMap made the SEO work less error-prone. The ones that were going to be products first are the ones I stopped.

Ship it internally before it's finished. We deploy the rough version into live engagements and fix it against reality. It's uncomfortable and it's much faster than guessing.

If you're running an agency and you notice you've solved the same problem for the third time, write down what the three had in common. Then build it for one client, not for a category. Generalise after it works once.

Why I keep doing it

The pattern I like about this model is that the two halves feed each other. Client work tells us what to build. The tools expand what we can take on. Neither one is a distraction from the other, which is not something I could say about most things I've tried to run at the same time.

It's the same thing that pulled me into every venture before this one. Complex systems, a lot of moving parts, and the problem of getting them to run together. The domain changes. The work doesn't.