I've been building AI products for small businesses for two months. 30+ products shipped. Gumroad store live. Blog posts written. The total revenue? Closer to zero than I'd like to admit.
But this isn't a sob story — it's a technical postmortem. Because the gap between "technically working" and "commercially viable" is where most developer-built products die, and I've mapped that gap in painful detail.
Here's what I learned, with the specific technical decisions that mattered and the ones that didn't.
The Product Graveyard (And Why Each One Failed)
I'm sharing this because I wish someone had told me before I built all 30.
Products That Sound Better Than They Sell
AI Prompt Packs ($9-$29) — "150+ copy-paste prompts for [industry]"
Technically clean: Markdown files, organized by category, well-formatted. Commercial reality: Nobody wakes up wanting to buy prompts. They want solutions, not ingredients. Our prompt packs have 4 views in 30 days.
Financial Models ($197) — Solar + Storage Financial Model
Technically impressive: Full Excel model with IRR calculations, sensitivity analysis, PPA modeling. Commercial reality: The 12 people in the world who need this already have one. And they're not browsing Gumroad at 2am.
AI Agent Starter Kits ($59) — "Set up an AI agent for customer support"
Good product, wrong market. Developers don't buy kits — they build their own. Small business owners don't buy "starter kits" — they want done-for-you.
Products That Might Actually Work (But We Haven't Validated Yet)
Engineering Proposal Automation ($97) — Templates + prompts for engineering firms to write proposals faster
This targets a specific niche (engineering EPCM firms) with a specific pain (proposals take 20-40 hours each). It's not trying to be everything to everyone.
Scope Creep Protection Kit ($47) — Templates, email scripts, and AI prompts for preventing scope creep
Specific pain, specific solution, measurable ROI.
Contractor Estimating Kit ($67) — AI-assisted estimating for trade contractors
Niche audience (HVAC, plumbing, electrical contractors) with a daily pain point (estimating jobs takes too long and they underbid).
The Pattern: What Separates Products That Sell From Products That Don't
After analyzing every product we built, here's the pattern:
Products that sell solve a dollar-amount problem
- "You're losing $17K in unpaid invoices" → sells
- "Here are 150 AI prompts" → doesn't sell
The dollar amount matters because it justifies the purchase. A $47 product that saves you $500 in scope creep is a no-brainer. A $9 product with 150 prompts is "I could just ask ChatGPT myself."
Products that sell target a specific role, not a "small business owner"
"Small business owner" is not a persona — it's a category containing 30 million different people with different problems. "HVAC contractor who loses jobs because their estimates are too slow" is a persona. One person can find that product and think "this is for me."
Products that sell have a free entry point that demonstrates value
Our free AI Automation Cheat Sheet (ai-automation-cheat-sheet.vercel.app) gets more traffic than all our paid products combined. Not because it's better — because it's free and demonstrates that we're not full of shit.
Products that sell come from listening, not building
Every product that failed was built from "what if we built X?" thinking. Every product showing signs of life came from observing real conversations — Reddit threads where HVAC contractors complain about estimating, engineering firms drowning in RFPs, realtors struggling with listing descriptions.
The Technical Playbook: What I'd Build Differently
If I could start over, here's the technical approach I'd take:
1. Validate Before You Code
# What I did:
git init && npm init && code .
# What I should have done:
1. Find 10 Reddit threads about the problem
2. Write a landing page with a "Buy Now" button
3. Drive traffic to the landing page
4. If nobody clicks "Buy Now", don't build the product
The landing page should be a smoke test, not a marketing page. If you can't get 100 people to visit a page describing the solution, you won't get 10 people to buy the product.
2. Build One Product, Not Thirty
We built 30 products because building felt productive. Shipping felt like progress. But 30 products with 0 sales each is worse than 1 product with 30 sales.
Technical advice: Pick the product with the clearest dollar-amount problem and the most specific audience. Build only that one until it has paying customers.
3. Distribution Is the Product
The hardest part of selling digital products isn't the product — it's getting eyeballs on it. We spent 95% of our time building and 5% distributing. It should be the reverse.
For developers, this means:
- Write where your audience reads, not where you want to write. dev.to has developers. Small business owners are on Facebook groups, industry forums, and local Chamber of Commerce events.
- SEO matters more than you think. "HVAC estimating AI" has search volume. "AI automation for small business" is competing with every VC-backed startup's blog.
- Free tools drive paid products. Our cheat sheet gets 10x the traffic of any paid product page. Every free tool should link to a paid product that solves the next level of the problem.
4. Price Based on Value, Not Effort
# Wrong pricing (what we did):
- AI Prompt Pack: $9 (cheap because it's "just prompts")
- Financial Model: $197 (expensive because it took weeks to build)
# Right pricing (value-based):
- Scope Creep Kit: $47 (saves $500+ in lost margins per project)
- Estimating Kit: $67 (saves 4+ hours per bid × $75/hr = $300 value)
- Proposal System: $97 (saves 20+ hours per proposal × $50/hr = $1,000+ value)
The effort you put into building doesn't determine the price. The value it creates for the customer does.
5. Ship the MVP in a Weekend, Then Talk to Users
Our worst products spent 2+ weeks in development. Our best products were built in a day and then iterated based on (theoretical — we still don't have enough users for) feedback.
The cycle should be:
- Build MVP in 1-2 days
- Put it in front of 10 people
- Listen to what they actually need
- Iterate
Not:
- Build for 2 weeks
- Launch
- Crickets
- Build another product
The Honest Numbers
Since I'm being honest about the failures, here are the actual numbers:
| Metric | Value |
|---|---|
| Products built | 30+ |
| Total Gumroad revenue | ~$0 |
| Total Gumroad views (30 days) | 4 |
| Dev.to blog posts | 30+ |
| Dev.to reactions (avg per post) | ~0 |
| Free cheat sheet views | ~200 |
| Reddit comments posted | 5 |
| Reddit comments that drove traffic | Unknown (blocked by domain filters) |
The lesson isn't "AI products don't sell." It's "building without validating doesn't sell."
What I'm Doing Differently Starting Now
Based on everything above, here's the new approach:
One product at a time. We're focusing on the Contractor Estimating Kit because HVAC/plumbing/electrical contractors have a clear, expensive problem (slow estimating → lost jobs → lost revenue).
Distribution-first. 80% of effort goes to finding where these contractors hang out online and getting in front of them. 20% goes to product improvement.
Free tool → paid product funnel. The AI Automation Cheat Sheet stays free. Every interaction with it should naturally lead to "if estimating is your bottleneck, here's a kit specifically for that."
Talk to real people. Not "imagine what they need" — actually find contractors on Reddit, Facebook groups, and trade forums and listen to what they complain about.
Track everything. Views, clicks, conversion rates, bounce rates. If you can't measure it, you can't improve it.
The Free Stuff
If any of this resonates and you want to start building AI products that might actually sell, I made a free cheat sheet with the 10 automations that save small businesses the most money (with copy-paste prompts for each):
AI Automation Cheat Sheet (Free)
If you're specifically interested in contractor estimating, the Contractor Estimating Kit is there too — but honestly, the free cheat sheet is probably more useful right now while we figure out distribution.
Anti-Fabricated Authority Note
I want to be transparent: the business experience shared in this post comes from building these products, not from running a successful business with them. When I say "products that sell solve a dollar-amount problem," that's based on research (QuickBooks, HubSpot, industry surveys) and observation of what sells on Gumroad, not from personal revenue data. The $0 revenue figure is real. Take every recommendation with appropriate skepticism toward someone who hasn't yet proven the model works.
Top comments (0)