Why do software products fail even when the product is good?

I've been through it more than once. After my insurance software stalled, my co-founder and I did the post-mortem, and the answer wasn't the product. It was sales. As he put it, we solved every other problem except the one that mattered most.

My history, for context: insurance agency software that reached about 15 customers over two years, then a year on an AI product that ran out of money, then my newsletter software, which I built because my own team needed it.

What actually went wrong?

• Nobody was selling full time. No one was putting in 20 to 30 hours a week in front of buyers.
• The market was mature. In a mature market an incomplete product loses, because buyers expect every feature the competitors already have.
• "No competitors" can be a warning. My co-founder's point: it can mean there's no proven market.
• Marketing never worked. I used to be an agent, and I still struggled to get in front of agents. Cold email got shut down, Meta ads didn't work, and referrals didn't spread. Two years of that.

"Marketing is everything. I don't care how good your product is. I don't care how valuable it is to the readers. Nothing matters if you can't put it in front of them."

The biggest startup mistake is building something people don't strongly want. It can be awesome and still not be a big enough pain point. Right behind it is having nobody to sell to. A product with a great idea and no audience has no buyers. That's a big reason I built local newsletters. I can reach 120,000 people instantly.

The best product often doesn't win anyway. Nobody likes Salesforce, yet it's one of the biggest companies in the world. I don't know their exact strategy, but I think part of it was going so hard on sales. They have something like 10,000 salespeople. How you package and sell matters as much as what you can build.

Why does a product that's 95% done still fail?

Because the last 5% can make it worthless. I funded a white-label AI product for about a year. It looked great in demos. Then clients used it and asked why their customers could see API keys, or why a user saw one set of custom GPTs and not another. The app worked fine for one person. Setup and user management were impossible for customers.

Demand wasn't the problem. Most projects stop because nobody wants it or nobody will pay. In this case people wanted it. It just wasn't good enough to give them.

Part of that was who built it. My developer partner knew a crazy amount about AI and could fix and automate things behind the scenes at a big company. That isn't the same skill as building a product non-developers can use. As a non-technical founder, I had to take his word for it.

The last 5% takes as long as the first 95%. That's been true on all three of my builds. One example from my newsletter software: the AI APIs can't read a web page from a link the way the ChatGPT app can, so we had to build our own tooling for something the apps do automatically.

Can I sell an app AI built in one shot?

Probably not to a lot of people, not without working on it hours a day. AI will do amazing things fast, then break something and send you down a rabbit hole fixing three other things. A simple one-shot app can work for yourself. Selling at scale to million-dollar businesses takes a lot of little details. greetform keeps getting more complicated because million-dollar businesses use it.

Cheap building creates its own trap. When building was hard, you'd never build a feature without understanding its value. Now it's easy to say yes to everybody, and things get bloated and confusing. I worry about that a lot. Shane Brady said the same on my podcast: the hard part with early customers is telling a scalable request from a one-off, or you end up with 400 features on an app that should do one thing.

Will people actually use it?

Adoption is its own risk. My insurance software was cheaper than a competitor's, and I still heard it: great concept, but my team's not going to use it. If the value depends on an owner making employees use another system, that's a huge risk, no matter how good the app is.

I have a version of this fear for my newsletter software. Operators may say it's cool and helpful, but they don't want another $30 to $50 software when building the newsletter in their email platform only costs them a couple of hours.

Painkillers sell more easily than vitamins. lightbreak is a vitamin: nice, but you aren't making money because of it. greetform makes the customer money. A buddy of mine is already proving that, and he's not even doing it right.

How much should I spend before I have proof?

Less than I did. My biggest mistake with my first AI product was putting too much money in before proof. Good feedback isn't proof. People paying you and wanting to work with you is proof. With my newsletters I'm spending a lot too, but I have that.

I also went too hard too early. We started in early 2024. If we'd started in mid-2025, with the new tools, the whole game would have been different. My rule: if you could spend $20,000, have it not work, and it wouldn't change your life, go for it. If you'd regret it badly, don't.

Builders from outside an industry make a related mistake. One vendor offered me a cut to promote a directory priced at $5,000 to solo operators who struggle to spend $50 or $100. Startup founders who'd never run a local newsletter pitched me software I wouldn't have used for free. Because I'm doing this 15 times a week, I at least know what I'm building matters.

What do I do differently now?

Define the sales motion before building more features. For my newsletter software that means three things: something exists, it's sitting in front of a buyer, and there's a defined price.

When should I shut a product down?

When the math has changed and won't change back. I officially shut down that insurance software after we'd put a lot of money and effort into it.

"We started it pre-AI, so it went after a small market, which at the time made sense, and everything that made sense then makes the opposite of sense now."

AI collapsed the cost of building competitors. The target audience was too small and the legacy players had a 10-year head start. The workload versus the opportunity was too small, and the opportunity cost for all of us was too high.

Why fall for the sunk cost fallacy? I would never start that product today. That settles it.

Should I wait for one big app or ship small pieces?

Ship the pieces. I gained roughly three to four months by keeping my website, lightbreak and greetform as separate products that integrate, instead of waiting until lightbreak could control the website. The website already worked.

Integration isn't the same as reliable end-to-end publishing, though. The demos at the time still had incomplete social visuals and no social scheduler.

For what to build instead, see niche versions of old software and doing it by hand before you build.