Case study 04 · SMM AI Assistant · AI content-planning SaaS
Five weeks to production.
Billing that adds up.
SMM AI Assistant is our own SaaS for small businesses and social-media specialists: it researches a brand, writes a social-media strategy and prepares a weekly content plan with posts, images and storyboards. Dmytro built it alone and took it to production with billing in five weeks.
- Product
- AI content-planning SaaS — our own product
- Our role
- Founder and sole engineer (Dmytro): architecture, development, DevOps, design, QA and product
- Period
- 4 August 2026 → today; in production since 8 September
- Users
- Solo entrepreneurs and social-media freelancers; dozens at launch
from first commit to production with billing
engineer, from architecture to design and QA
margin over AI costs built into pricing
automated end-to-end tests
The challenge
Every AI call costs real money. The product had to meter each one, charge users fairly, never pay twice for the same call and keep a margin on every plan — from the first release.
Results
- In production on 8 September 2026, five weeks after the first commit, with payments, ads and attribution connected.
- Charges are checked against the OpenAI bill every day; on its first morning the check caught a 3.8% discrepancy.
- Dozens of users at launch; advertising is still ahead.
What we did
- 01
Strategy first, then weekly content
Research and strategy come first, then a weekly plan of posts, images and storyboards, generated in the background.
- 02
Billing that protects the margin
Credits, five Stripe plans and limits per plan, with prices set at three times the AI cost.
- 03
Ready for real users
An onboarding questionnaire, a calendar of posts and storyboards, strategy import from PDF or CSV, and Ukrainian and English interfaces.
- 04
Growth tools from day one
An admin panel with a Telegram support bot, consent-based marketing attribution and short links.
- 05
Tested end to end
112 automated end-to-end tests run against the real database and background worker.
Under the hood
Money that reconciles itself
What users are charged and what OpenAI bills have to match automatically.
- Every call is metered in credits, based on the tokens it used and the models the user's plan allows.
- A daily job compares charges with the OpenAI bill and raises an alert when they differ by more than 2%.
- Background jobs run on queues in Postgres, so a deploy never leaves work stuck or pays for the same call twice.
How we worked
Our own product. Every change goes through a pull request with AI review, the author's own review and automated checks, and the main branch deploys automatically.
Stack
Start a conversation
AI features that pay for themselves?
Tell us what you're building, modernising or trying to unblock. One contract via a Lithuanian entity, EU VAT invoicing — we reply within one business day.