← Back to all case studies

Owned AI product / iOS + Mac

AI value with economic guardrails.

For Collage Easy, I connected AI editing with entitlements, credits, provider cost, privacy, server quotas and recoverable failure across iOS and Mac.

Conceptual AI editing system with collage tiles, credit controls and result or refund branches
AI guardrailsConceptual visualization, not production UI
OwnershipProduct + implementation
PlatformsiOS + Mac
AccessPro + AI credits
ControlServer quotas
01 / Product problem

An AI action has cost, trust and failure states

A prompt field and a result image do not make a viable AI product. Every action can carry variable provider cost, sensitive user content, upload latency, safety rejection, timeout, rate limit or entitlement ambiguity. The experience needed to remain understandable while the backend enforced rules the client could not safely own.

Value

Useful creative outcome

The flow must stay focused on editing and export rather than exposing infrastructure complexity.

Economics

Variable model cost

Feature-level limits and pricing have to protect margin without making usage feel arbitrary.

Trust

Photos and prompts

Provider secrets stay off-device, while analytics excludes user photos, prompts and image content.

Resilience

Failures should not feel like theft

Temporary provider errors, timeouts and safety rejects require clear states and appropriate credit refunds.

02 / Workflow

Entitlement to outcome

01Choose an AI editing action
02Check Pro access or credits
03Validate and compress input
04Enforce quota and process
05Return result or explain and refund
System boundary

Client UX → entitlement state → server policy → AI provider → result or recoverable failure

The client explains access and progress. The server owns hard usage limits and provider credentials. The product decides which failure states consume a credit and which must reverse the charge.

Explore product guardrailsAccess, quotas, refunds, purchase safety and privacy
03 / Decisions

Product economics encoded as behavior

Access model

Keep the core editor useful for free, add Pro entitlement for ongoing value and use consumable credits for variable-cost AI actions.

Feature limits

Use feature-specific allowances instead of one global number because provider cost and perceived value differ by action.

Hard control

Pair client-side guidance with server-enforced D1 quotas so a modified or stale client cannot bypass policy.

Cost ceiling

Default to a lower-cost 1024 by 1024 processing mode and cap expected feature cost relative to its price budget.

Failure fairness

Refund usage on safety rejection, rate limit, timeout and temporary provider failure when no usable value is delivered.

Purchase safety

Grant credits idempotently from the StoreKit transaction ID so a repeated transaction cannot duplicate value.

04 / Guardrails

Limits that protect users and the business

Global stop

500 AI actions per day

A server-side kill cap limits unexpected provider spend across the product.

Installation limit

100 purchased-credit actions

A daily per-installation cap narrows abuse while preserving legitimate paid use.

Analytics boundary

Consent-aware events

The funnel can measure feature selection and collage export without sending photos, prompts or image contents.

Secret boundary

Provider keys stay server-side

The app receives a product result, not direct authority over the external provider account.

Private product case reconstructed from reviewed product and backend documentation. No conversion or revenue metric is claimed because none was verified for this case.

05 / Outcome

A complete product system, not a demo feature

AI value depends on the states surrounding the model response.

Collage Easy combines the creative workflow with access rules, pricing, cost protection, privacy, failure recovery and measurement across iOS and Mac Catalyst. The outcome is a product that can be operated and evolved, not only a technically impressive interaction.

User outcome

Clear access and recovery

People can understand why an action is available, what it costs and what happens when processing fails.

Business outcome

Bounded AI exposure

Feature quotas, global limits and server policy create controllable cost boundaries.

Engineering outcome

Separated responsibility

Client, backend, store and provider each own the state they can enforce reliably.

Next measurement

Value after export

Track whether successful AI edits improve creation completion, export and repeat use, not only feature taps.