GhonaimStudio

Home / Work / W-02

AI Lab: funnel, billing and access lifecycle

iPhone Photography School — Online photography school, 800K+ students, Latvia

Built as senior engineer at IPSWeb app2025Latvia, remote

The AI Lab sales page as it appears to a visitor arriving from an ad
Industry
Online education
Size
Platform serving 800K+ students
Region
Latvia, remote
Stack
Laravel, Vue, Braintree, Meta CAPI, BigQuery, Sigma
Timeline
2025, shipped in stages
Status
Live

The problem

The school wanted a new paid AI product line, sold to its own audience and to cold traffic from Meta ads. That meant an acquisition funnel that converted, subscription billing that could handle introductory pricing without producing wrong receipts, and community access that switched on when someone paid and off when they cancelled. All of it had to be live for a launch date.

What we built

A quiz funnel that converts

A Vue quiz and acquisition funnel with its controller, state and navigation, tracking and tests. Marketing designed the flow; we owned the engineering end to end. It converted about four times better than the previous flow in a 30-day A/B test across Meta ads and launch-email traffic.

Tiered subscriptions on Braintree

Introductory pricing that rolls into a plan, plan suffix and frequency handling, first-cycle discounts, recurring receipts, and webhook receipt routing with deferred linkage. The launch ran without billing incidents.

Access that follows billing

A queued cancellation and entitlement revocation pipeline: Braintree cancellation and access-expiry dates drive chunked revocation jobs and asynchronous member creation in the Circle community, so nobody keeps access they stopped paying for and nobody loses access they paid for.

Ad tracking that does not double count

Meta Pixel plus the Conversions API, with per-product server-side routing and composite event-ID de-duplication written atomically. Meta Event Match Quality improved by about 20%.

Live campaign reporting

Airbyte into BigQuery into Sigma, so the team could see campaign spend against conversions while the launch was running, not a month later.

  • Laravel
  • Vue
  • Braintree
  • Circle
  • Meta Conversions API
  • Airbyte
  • BigQuery
  • Sigma

Screens

The AI Lab sales page as it appears to a visitor arriving from an ad
The AI Lab sales page, the end of the funnel.

How we worked

The whole launch was built on an AI-assisted workflow: Claude Code as the primary tool, Codex as the second, MCP servers for Sentry, GitHub and Meta, and parallel agents for the repetitive parts. The same method produced a second A/B win on a smaller sales page, at about 1.5x conversion.

Results

~4x

conversion lift on the new funnel

30-day A/B test across Meta ads and launch-email traffic.

+~20%

Meta Event Match Quality

After server-side events with composite event-ID de-duplication.

~1.5x

conversion on a second funnel

A quiz mini sales page for another product, same A/B method.

What's next

The billing, access and tracking pieces built for this launch are now the pattern the platform reuses for new products.

Have a similar problem? Tell us