Workflow X-Ray / Builder case study
Building around exact workflow economics.
Owners cannot price a manual workflow before deciding to change it. Generic AI advice skips the measurement step.
- Customer workflow details are confidential business information.
- The economics must be exact, not plausible.
- The report must never state something the customer did not say.
The build
Key decisions
- 1
One full-transcript interview, then a single structured extraction whose every figure carries the customer's own words.
- 2
All money math in code: the model proposes facts, and code computes.
- 3
Model routing is restricted to zero-data-retention providers with the named fallback model claude-haiku-5.5, so a provider outage never strands a paid report.
- 4
The paywall sits after the customer confirms the workflow summary and sees the free preview — payment buys the written report, not access to their own numbers.
- 5
The report route verifies payment server-side on every view rather than trusting a success redirect.
Making long structured drafts reliable took staged generation, output budgets, and a repair pass design.
Sensitivity classification is handled by extraction with an abstention path rather than keyword matching.
Current state
Workflow X-Ray is live, the report is $49, and an end-to-end purchase has been verified in production.
Continue to the product
For hiring conversations, collaborations, or product questions: hello@agenticartists.com