If you run a service business, your best product idea is already inside your delivery work
Most product validation is guessing with extra steps. Landing pages, waitlists, surveys, interviews where people politely tell you they'd pay and then don't.
We run a marketing agency, Tactycs, and I've come round to thinking that service businesses get handed a better validation mechanism than any of that, for free, and mostly waste it. The tools you build to survive your own delivery work are pre-validated by definition. You built them because the pain was bad enough to spend billable hours on.
We've turned two of those into products now. The second one is a much stronger case than the first, and the gap between them is the interesting part.
The first one taught us the pattern existed
Local Leads started as an internal tool for running our own outbound. We needed local business leads quickly, built something that got them, and used it on ourselves and for clients for a long time before it occurred to anyone that it was a product.
It works, people pay for it, and "we built this for ourselves and it works" is still the most useful sentence we have when selling it. People trust that in a way they don't trust a feature list.
What it didn't teach us was how to do it on purpose. It happened to us and we noticed afterward.
The second one we watched deliberately
Evergreen AI came out of the same place. SEO delivery at an agency is expensive in specific, repetitive ways: working out what a client's competitors rank for that the client doesn't, deciding what to write against those gaps, writing it, publishing it, then doing the whole thing again next month because everything moved.
So we built an engine to do that. Then, and this is the part I'd repeat, we sat on it.
It ran internally for months across around 15 client accounts before anyone outside the agency could touch it. Real businesses in different verticals, real competitors, real retainers, and our name on every article it published.
That's validation you can't buy, and it has four properties a landing page test doesn't:
- Real sites in different industries instead of a sandbox
- A genuine control, because we knew exactly what the manual version of this work produced and what it cost us in hours
- A quality bar with teeth, because clients who get fed AI slop fire their agency, and they'd be right to
- Twelve month horizons, which is roughly how long SEO takes to prove anything either way
The results are why we shipped. Our longest-running account on it went from an average position of 42 to 7 over twelve months, organic clicks up 650%, impressions up 1,257%, 9.5x the keywords ranking. Their Search Console export, not our dashboard.
By then, opening it up wasn't much of a leap. We had a year of evidence that it beat what we'd been doing by hand, on accounts where being wrong would have cost us the client.
Why this beats the usual approach
A service business gets paid to locate pain precisely. That's the whole job. You aren't guessing which problems are worth money, you're invoicing for them monthly, which means your product backlog is already sorted by willingness to pay before you write a line of code.
You also get something founders rarely have, which is a baseline. We could compare the engine against our own manual process because we'd been running that process for years. Most products launch with nothing to measure against except a competitor's marketing page.
And the agency funds the patience. We could afford to run this quietly for months instead of shipping in six weeks and hoping, because the business paying our salaries wasn't the thing we were testing.
The catch, which is bigger than it sounds
What makes an internal tool good is that you're standing next to it. When it does something odd, you know what to do. Your customer doesn't.
Every internal tool has expertise baked into the operator rather than the software, and productizing one is mostly the work of removing yourself from that loop without the output getting worse. For us that meant a lot of unglamorous scaffolding around quality control, because "Kyle reads it before it goes out" is not a feature you can ship.
Some of it is unglamorous to the point of comedy. One filter in there exists purely to strip em dashes and words like delve, leverage and robust, because those are the tells that make a reader clock a page as machine-written and stop trusting it. I used to catch that by reading everything myself. The software catches it now, on every article, which is the only version of that job that scales.
That part took longer than building the engine did. It's also the part nobody warns you about, because from the inside the tool already feels finished.
Where it landed
Evergreen AI is public now at $149/mo per site. It doesn't replace the agency, it sits underneath it. Businesses that can't justify a retainer run the engine themselves. The ones who outgrow it and want people in the room, we're still here, which is the same shape most agencies could build if they looked at what they've already made.
The framing I'd offer anyone running services: you're not choosing between services and products. Your delivery process is R&D, and it's already funded.
Happy to get into how we ran the internal validation period, or what we got wrong the first time around, if it's useful.
The internal validation point is compelling. Having something run across real client accounts for months gives you a very different signal than getting positive reactions to a landing page.