Why we capped Pro at $5 and refused to add quotas

We're a job-search tool. Our users are people whose income just disappeared, or is about to. The last thing I want them looking at, in week three, is a paywall that says “you've used 47 of your 50 monthly tailorings.”

The math of metered pricing

Every job-search SaaS in our reference set runs a quota model. LazyApply sells “750 applications per month.” Sonara sells tiers by “applications submitted.” Teal sells by saved jobs and tailored résumés. The reasons are obvious: it caps cost exposure when a power user blows through your inference budget, and it makes upsell easy.

The reasons it's wrong are quieter. Job search is bursty — you do nothing for two weeks because of a family emergency, then run 80 applications in a Tuesday because a layoff hit. A monthly quota fits neither rhythm. And the conversion event the quota forces — “upgrade or stop” — happens at the worst possible psychological moment, because the user is already burnt out from rejection.

What $5 has to defend against

Flat $5 unlimited only works if four things hold:

1. Cheap inference, by default.We default to Gemini 2.5 Flash Lite and GPT-4o Mini. Both deliver tailoring quality close to the frontier models at <10% of the cost. We have a hard whitelist (LLM_TIER_WHITELIST) that refuses to start the API if a more expensive model is configured without an explicit override flag.
2. Caching where it counts.Anthropic prompt caching is wired into our tailoring path; Vertex context caching keeps a per-user résumé cache so we don't re-tokenise it across 30 tailorings a day. Cache hit ratios on a typical Pro user run 60-80%.
3. Concurrency limits.A user can't fire 200 parallel tailorings; we serialise per-user and rate-limit. This costs us nothing and prevents the worst-case bills.
4. Account aging. Most abuse is signups in the first 24 hours. Conservative limits in week one flatten that curve without affecting real users, who barely notice them.

The result: our blended cost per Pro user is well under a dollar a month, even for power users firing 30 tailorings a day. Five dollars works.

What we give up

Two things. First, we don't have an obvious upsell path. There's nothing to upgrade to. That's a feature, not a bug — pricing pages with five tiers and a comparison matrix are a tax on the user's decision time. We have one button: $5/month. Decide in three seconds.

Second, we forfeit the high-ARPU power users who would happily pay $30 for unlimited everything. Some competitors will outprice us at the top. That's fine — we'd rather have ten thousand users at $5 than a thousand at $50. The bigger pool is harder for a competitor to dislodge, and it's where the product gets the most signal for improvement.

Refunds are part of the price

The auto-apply category has a chargeback rate roughly 5x the SaaS norm — between 1.5% and 2.5%. Most of those chargebacks come from users who can't figure out how to cancel, or who tried the product, didn't click with it, and used their bank as a complaint channel.

We ship a one-click refund button inside the app. No support ticket, no screening questions. Click, full refund, subscription cancelled. The cost of refunding a fraction of users cleanly is a hundred times lower than the cost of one chargeback — Stripe charges $15 in dispute fees alone, and chargeback rates above 1% threaten your processor relationship.

If you're building a similar product

The pricing question gets framed as “what will the market bear?” That's the wrong question. Ask instead: “at what price do my users tell their friends about me?” For a product whose users share the same Slack channels and Reddit threads, word-of-mouth is the only acquisition channel that compounds. Five dollars is a recommendation. Forty-nine dollars is a complaint.

— Aayush