From days of manual setup to minutes — powered by production-grade AI.
Launching a new loyalty program took days of configuration. We built an AI system that generates a fully configured program from a brand's website URL, with retryable LLM calls, validation, and audit trails.
Context
Every time Perk onboarded a new client, someone had to manually configure the program: upload brand assets, pick colors, create challenges, configure products, set up tiers. This took days per client and required developer involvement for most steps. The goal wasn't to add a chatbot. It was to build AI infrastructure that could reliably analyze a brand's online presence and generate a production-ready loyalty program configuration — automatically.
- •Program setup took days of manual configuration per client
- •Required developer involvement for most steps
- •Brand assets, colors, challenges, products, tiers — all manual
- •AI output needed to be production-ready, not suggestions
- •Must work reliably across diverse brand websites
Our role
We designed and built 'Assist' — an AI module that takes a brand's website URL and generates a fully configured loyalty program. Not a prototype. A production system with retryable calls, validation, audit logging, and feature-flagged rollout.
"There's a level of professionalism and consistency in that output being high quality and trustworthy. Instead of just saying yes and doing the thing, you make sure it's lining back to its purpose."
What changed
Program setup went from days to minutes. The AI pipeline handles brand analysis, color extraction, challenge generation, product identification, and reward suggestions — all automatically.
- ✓Program setup reduced from days to minutes
- ✓9-step onboarding wizard with save/resume capability
- ✓Automatic brand analysis from any website URL
- ✓Challenge, reward, and tier configurations generated automatically
- ✓Feature-flagged rollout for controlled adoption
Areas we took ownership of
Why this matters
This is the difference between 'we integrated an LLM API' and 'we built AI infrastructure that operates reliably under real conditions.'
Retryable calls with custom validation. Error feedback piped into retry prompts. Full audit trail. Feature-flagged rollout. Background job processing. Production AI means the system works when nobody is watching, not just during demos.
Outcome
Program setup went from a multi-day manual process to minutes of automated generation plus human review. The AI infrastructure we built — retryable calls, validation, audit logging — is reusable across any future AI features Perk adds to the platform.
If you're building AI features into an existing product and need them to work reliably in production, this is the kind of work we do.