Skip to content
Build Notes
Build Notes

I Built 4 SaaS Products With AI. Total Revenue: $279.

Creation got commoditized. Distribution got captured. A field report on a year of AI products, the $279 they made, and why the build was never the hard part.

I Built 4 SaaS Products With AI. Total Revenue: $279.

Last year I shipped four SaaS products, a Kalshi trading bot, a virtual product, and a small ad business that two AI models designed from scratch. Total revenue across all of it: $279.

Then it gets worse. $250 of that came from the trading bot. It placed one bet it was sure about. The data later showed the signal was a hallucination, the kind of confident mistake AI makes all the time. It won anyway, by luck. The other $29 was a single sale of a virtual product. The four SaaS products I actually built, the ones I tuned for months, made nothing. Zero.

I spent a long time assuming I’d built the wrong things. I hadn’t. That was the lesson.

The receipts

What I builtWhat it doesRevenue
serpdeltaGoogle Search Console analytics$0
itbrokesite monitoring (now becoming askbowtie)$0
spendravenanalytics for paid ads$0
bestthriendsThreads insights and scheduling$0
Kalshi botcustom prediction-market trader$250
yourpathreading.coma virtual product$29

The few people I got to try the SaaS products gave real, unprompted, positive feedback. Not “this is nice.” More like “when can I keep using this.” So the build wasn’t the failure. The code worked. The UIs were clean. The thing did the thing.

That’s what makes it useful as a lesson. When the product is fine and the result is still zero, the problem is somewhere you weren’t looking.

What actually went wrong

Three things, and none of them were the build.

Commodity. Every one of those tools already exists, in better-funded form, with a sales team and three years of reviews. The demand was filled before I started. I wasn’t opening a market. I was joining a queue.

No edge. My only differentiation was “looks a bit better” or “could be cheaper.” That’s not an edge. That’s a coupon. An edge is something a competitor with a bigger budget can’t copy by Friday. I had none. This is the same wall I hit on a client’s hosting section: clean content, decent SEO, and still no reason for it to win. If you can’t name the one thing your thing says that nobody else can, you don’t have a product, you have a page without a moat.

No persistence. Real products are pushed for years by people who refuse to stop. I shipped, posted a few times, and moved to the next idea. Without that grind, a good product just collects dust where nobody can see it.

And nobody could see them. I’d built five sites and was a ghost to every crawler and AI looking at them. Being good is invisible if being found never happens.

The experiment that proves the point

Here’s the one that settled it for me. I wanted to know what the best models would do completely on their own, so I let Codex and Claude Code design a business together, one model instructing the other. The plan they came up with: find a low-cost, high-volume keyword on Google Ads, build a reusable calculator around it, run paid traffic, monetize with affiliate links.

They built all of it. The concept, the site, the ad copy, the campaign. I mostly watched.

It failed on math. Affiliate revenue landed near 2%. Click-through stayed under 5%. The unit economics never closed. Two of the most capable models alive ran an entire business end to end, and it still died, because the build was never the bottleneck. Distribution was. The economics were. The edge was.

The part that’s getting harder

Now the uncomfortable update. The “get found” problem isn’t holding steady. It’s getting worse, and AI is the reason.

Start with search. When Google shows an AI summary at the top, people stop clicking through.

No AI summary: users click a real result about 15% of the time.

AI summary present: that drops to 8%, and only 1% click a source cited inside the summary.

That’s Pew’s behavioral data from 2025. Ahrefs measured the same thing from the other side: the top organic result loses about 58% of its clicks when an AI Overview shows, and that figure got worse through the year. Google’s own framing, from Sundar Pichai, is that AI Overviews increased engagement. He gave no numbers. Read both claims together and the trade is clear: Google’s engagement went up, the clicks small sites live on went down.

Then there’s the recommendation problem, which hits new products hardest. When you ask an AI to recommend a tool, it reaches for the names it has seen most. Research across seven frontier models found a consistent bias toward established, high-exposure incumbents, and the bias survives standard debiasing because it’s baked into the training data itself. That work studied investment and brand recommendations, not SaaS specifically, so I won’t overstate it. But the mechanism is obvious enough to reason from. An AI asked for “the best monitoring tool” has no incentive to dig through thousands of solo-built products that might be dead, half-finished, or a scam. The proven, multi-million-dollar option is right there. It’s the safe answer. AI defaults to the safe answer.

So the two free channels a small builder used to have, organic search and word-of-AI-mouth, both tilt toward whoever is already big.

That leaves paid. And paid is climbing out of reach. Average Google Ads cost-per-click rose 12.88% year over year, up across 87% of industries, with 2026 sitting near $5.42 a click and five straight years of increases. Paid search is quietly becoming the new organic: the place you have to be, priced so only the well-funded incumbent can afford to outbid you.

I’m not calling search dead

I want to be honest about the size of this, because the doom version is wrong. AI search is still tiny. It drives roughly 0.32% of web traffic today against about 42.75% from regular Google organic. Nobody’s distribution has been replaced yet.

But the direction is the whole point. AI referrals convert higher than normal search and the share is climbing fast. This isn’t a takeover that happened. It’s one that’s loading.

AI didn’t make building hard. It made building easy, then quietly made being found the only thing that matters.

What I’m betting on now

I don’t have a clean answer. Anyone selling you one is selling you something.

What I’ve changed is where I spend. Building used to feel like the work. Now I treat it as the easy 20%. The real job is edge and distribution, and that’s where the time goes. The rough bets:

1
Own a channel nobody can outbid me for. An audience, a list, a community. Rented attention gets expensive; owned attention compounds.
2
Get listed where small builders still surface. Directories, niche aggregators, places a human goes specifically to find the non-enterprise option.
3
Chase search that still rewards the work. Long-tail queries and smaller engines where good SEO and a real answer still beat brand size.
4
Serve people who don’t want the safe answer. There are buyers who actively distrust the giant incumbent. That distrust is an opening.

None of these are proven. They’re where I’m pointing next, written down so I can check later whether I was right.

The takeaway I’d hand my past self: AI will help you build almost anything. It will not help one person find it. Treat the build as the cheap part, because it is, and spend your real effort on the part the machines are quietly making harder.


Related:

Ben Peetermans

Ben Peetermans

I build MVPs and run ad campaigns with AI, and I write these notes from real projects. Twenty years shipping for the web. If you want someone who works with these tools every day to build yours, that is what I do.