Cater AI: Balancing Guidance and Control in Catering Requests
What I built and tested
I created two prototypes for the same catering inquiry task. Alex, built in Voiceflow with Claude Sonnet, guided users through their requirements, checked serving quantities and budgets, and personalized order confirmations. The JotForm baseline let users enter information directly, without real-time constraint checks.
Before building, I reviewed research on conversational AI, chatbot usability, and B2B lead generation, and benchmarked three conversational platforms. This informed six design decisions and iterative simplification of Alex. The process followed Design Science Research and the Double Diamond framework.
The study compared the interfaces using one catering scenario. I assessed qualified lead rate and order accuracy across seven criteria, alongside a chatbot usability questionnaire (B2B-BUS-15) and three open-ended questions to understand participants’ experiences.
What the study revealed
Participants described the form as faster and easier to control when order requirements were clear, with menu details visible together. Overall order accuracy was similar: 76.8% for Alex and 80% for JotForm. JotForm captured dietary requirements more reliably, meeting them in 80% of submissions compared with Alex’s 50%.
Participants valued Alex’s serving-count warnings, budget estimates, and clarification questions. Budget compliance was 100% for Alex versus 90% for JotForm; date and time accuracy was 62.5% versus 50%.
Alex’s Trust and Privacy score was near-neutral. Feedback linked weaker trust to session failures; technical interruptions prevented two participants from submitting. Reliability was essential to credibility. These exploratory results reflect eight scoreable chatbot submissions and ten form submissions.
From research to iOS design
I translated the findings into an iOS homepage with two entry points. Alex helps organizers clarify requirements, while the form gives them direct control when they already know what they need, even for complex orders. Both paths lead to the same event brief. The choice reflects both the complexity of the request and how much guidance the organizer needs.
The prototype explicitly asks how many guests have each dietary requirement to address gaps in Alex’s dietary capture. A real-time per-person budget display builds on its stronger budget compliance. Event cards show Matching, Quotes in, Booked, and Done to make progress visible beyond intake.
Reflection
This project taught me to choose an interface based on the support users need. Conversational guidance is valuable when it reduces cognitive load and helps people resolve uncertainty, but that value depends on reliable operation and accurate capture of requirements. When conversation adds unnecessary effort, a simpler interface can be the better design.
