The cheapest part of running an AI SaaS is the AI. Pricing is the part developers keep getting wrong.
Cheap APIs were supposed to mean cheap software
Instead they've created a strange market where AI founders price themselves into oblivion — because they can't believe their own margins.
You see it in every "Show HN" launch lately. A solo developer ships a Claude-powered tool, posts it for $5 a month, and within six weeks they're closing it down. Not from lack of users. From unit economics that made customer support cost more than the subscription. From a price that was set against the API bill instead of the customer's pain.
This is the trap I almost walked into. This is how I climbed back out.
The stack
If you read my last post on building an AI summarizer in one night with Claude + Next.js, this is the same stack — but a different problem.
I'm building HireRoom — an AI CV tailorer for job seekers. Paste the job posting, upload your master CV, get a tailored .docx back in 30 seconds. Claude API costs me pennies per CV. Pricing turned out to be the hardest part of the whole project.
The wrong math
I opened a spreadsheet. Wrote down what one active user costs me to serve. Roughly $1–3/month in Claude API calls on average. Maybe $5–8 for heavy users running 50+ CVs through the system.
I followed the standard cost-plus formula every developer instinctively reaches for:
- Cost per user: $2/month
- 4× markup
- Final price: $8/month
I sat with that number for a day. It felt off. Not because the math was wrong — the math was fine — but because $8/month felt like apology pricing. I was apologizing to imaginary customers for charging anything at all.
The shift: change the question
Instead of "what does it cost me to serve this customer?" I asked "what is the customer already losing without me?"
I imagined a specific customer: a software engineer in a layoff cycle, applying for 20 roles a month. Each application "costs" 30 minutes of CV tailoring — re-ordering bullets, surfacing keywords, hiding lines that look irrelevant for this role. Skip the tailoring and recruiters auto-reject in 6 seconds. Do the tailoring and the time stacks up fast.
Run the math on his side:
- 30 min/CV × 20 applications/month = 10 hours/month of tailoring tax
- At $50/hr opportunity cost (a conservative software-engineer rate): $500/month of pain
- Plus the jobs he never applies to because by application 6 he's burned out and the pipeline thins
- Plus the emotional cost of weeks-long job searches stretched into months
So his pain is conservatively $500 a month. Mine to fix: $1–3.
Suddenly, charging $29–49/month felt obvious. It's 6–10% of his monthly pain — enough that he sees the value, leaves him with $450 of net time-savings, and gives me 95%+ margin. Same product. Same costs. Different anchor. (Job seekers anchor lower than B2B because their use is bursty — concentrated during a job search, then they cancel. The 30% rule that works for SMBs over-prices for individuals; price for the discount they expect on consumer subscriptions, not the full enterprise mark-up.)
That was the shift: stop pricing your costs, start pricing the customer's pain. The numbers I'd been staring at — Claude calls, server bills, my own time — were bookkeeping. The number that actually mattered was the one on his side of the table.
Three anti-patterns I had to unlearn
1. Calculating cost-per-user
It's a useful number for unit economics, but it has nothing to do with what you charge. Cost-per-user tells you when you're losing money. It doesn't tell you what to charge.
2. Discounting to "get the first customer"
I caught myself drafting a "founding member" pricing tier at $9. The instinct was to remove every barrier so someone — anyone — would say yes.
The reality: bargain-hunters churn the fastest, complain the loudest, and don't help you understand whether your product solves a real problem. The first customer who pays full price is worth ten who got a discount.
3. Justifying the price by listing features
"It does X, Y, Z — that's why it's $99." Wrong. Customers don't pay for features. They pay for the absence of pain. Justify your price by naming the pain you're removing, in their terms, in their currency.
None of these three track the customer's pain. The customer's pain is the only thing pricing should track.
Try it + see the code
Live work I've shipped with this stack:
- AI Summarizer: ai-summarizer-next.vercel.app
- Portfolio: yramstech.com
HireRoom — the AI CV tailorer I'm pricing in this post — went live this week. Waitlist's open: first 100 signups get it free at launch. Follow my dev.to for the build journey.
If you're pricing an AI product right now and the numbers don't feel right — uncomfortable on your side, suspiciously generous on theirs — that's the signal you nailed it.
Don't fix the feeling. Trust the math.
