Cutting customer onboarding effort dramatically with AI-assisted workflows
How my teams at Flipdish rebuilt customer onboarding around AI-assisted menu creation — reducing manual effort, shortening time-to-live for new restaurants, and lowering early-stage client churn.
The challenge
Flipdish provides online ordering, kiosks, and restaurant management software to thousands of restaurants. Before a new customer can take a single order, their entire menu — categories, items, descriptions, options, modifiers, pricing — has to exist accurately in the platform.
That setup work was largely manual. Menus arrive as PDFs, spreadsheets, photos, and links in every imaginable format, and turning them into structured platform data was slow, repetitive, and error-prone. The cost wasn’t just internal effort: every day a restaurant waits to go live is a day it questions its decision to sign — and slow onboarding showed up directly in early-stage churn.
My role
As Engineering Manager I lead multiple Agile teams at Flipdish, reporting directly to C-level leadership. I owned this initiative end-to-end: framing the problem with stakeholders, shaping the architecture, planning and prioritizing delivery across sprints, and reporting outcomes against business metrics — not shipped features.
The approach
- Map the bottleneck before touching AIWe traced the onboarding funnel step by step to find where the hours actually went. Menu creation dominated — making it the highest-leverage target rather than the flashiest one.
- Build AI-assisted menu creationWe introduced LLM-powered workflows that take the menus customers actually send — documents, images, links — and draft structured menu data automatically: categories, items, modifiers, and pricing ready for review instead of manual entry from scratch.
- Keep humans on the quality gateAI drafts, people approve. Onboarding staff review and correct the generated menus, so accuracy stays high while the repetitive transcription work disappears — and their corrections show us exactly where to improve the workflows next.
- Ship iteratively, measure relentlesslyWe rolled the workflows into real onboarding pipelines sprint by sprint, measuring onboarding effort and time-to-live against the manual baseline and reporting the deltas to leadership.
The results
- A major improvement in onboarding efficiency — new restaurants go live with markedly less manual effort per account.
- Lower early-stage client churn — faster time-to-first-order keeps new customers engaged through the riskiest phase of the relationship.
- Onboarding capacity scales without headcount — the same team handles more new customers, and the work shifted from data entry to quality review.
What made it work
Business metric first, technology second. The goal was never “add AI” — it was cutting onboarding effort and churn. Framing it that way kept leadership aligned and gave the teams a clear definition of done.
Human-in-the-loop is a feature, not a compromise. Keeping people on the approval step made the system trustworthy enough to put in front of real customer data on day one, and their corrections became the improvement loop.
AI-assisted delivery compounds. The same teams use GitHub Copilot, Claude, and ChatGPT across the development lifecycle — so the people building AI workflows were also shipping faster because of them.
“His engineering skills are exceptional, he has strong judgment, writes clean and scalable code, and consistently thinks a few steps ahead when it comes to architecture and long-term impact.”
Facing a similar onboarding or delivery bottleneck?
I help product organizations apply AI where it moves business metrics — as an engineering leader, fractional CTO, or consultant. Let’s talk about what that could look like for your team.