Form vs. Flow: A B2B AI Study
From a research-backed comparison to a working iOS prototype
Problem and Research Question
B2B order forms are efficient for routine requests. But when buyers have dietary constraints, budget limits, and uncertain guest counts, forms collect whatever users type without helping them think through complexity. I wanted to test whether conversational AI could do something forms structurally cannot: guide users through complexity in real time.
Research question: What kinds of ordering tasks actually benefit from conversational guidance, and under what conditions does an AI chatbot outperform a static form for B2B catering intake?
Method
I structured the project as a Design Science Research study mapped onto a Double Diamond flow, running from literature review and platform benchmarking through iterated prototype development to a 12-participant mixed-methods evaluation.
- Discover and Define: Reviewed literature across conversational AI design, chatbot usability measurement, and B2B digital lead generation. Benchmarked three leading conversational platforms and translated findings into six concrete design decisions for the prototype.
- Develop and Deliver: Designed two instrumented prototypes for the same catering inquiry task: an AI chatbot (Alex) built in Voiceflow with Claude Sonnet, iterated heavily to a stripped-down final version, and a static JotForm baseline. Ran a within-subjects mixed-methods study with 12 participants. Quantitative measures included qualified lead rate, a 7-dimensional order accuracy rubric, and the B2B-Bot Usability Scale-15. Qualitative measures included 3 open-ended questions per participant.
Alex (AI Chatbot)
- Guided intake, proactive constraint checking
- Flags when servings or budget look off
- Personalizes confirmation before submission
JotForm (Static Form)
- Direct input, familiar structure
- Captures whatever the user types
- No real-time feedback
Findings
The honest headline: AI did not win outright. The win was conditional.
- Forms won on simple tasks. For routine, low-complexity orders, participants preferred the form. It felt faster, and they stayed in control. Conversational guidance added overhead where none was needed. The form also captured dietary requirements more reliably and matched the chatbot on overall order accuracy, so this was never a clean AI victory.
- Chat won on complex constraints. When orders involved multiple requirements, dietary restrictions, budget caps, and uncertain guest counts, the chatbot’s proactive checking caught omissions that forms missed entirely. The chatbot performed better on budget compliance (100% vs. 90%) and date/time accuracy (62.5% vs. 50.0%), because the form will happily accept an impossible order, an over-budget total or a midnight delivery, with no way to validate what people type.
- Trust followed reliability, not novelty. The chatbot’s B2B-BUS-15 Trust and Privacy factor scored near-neutral. Qualitative feedback showed this reflected two session crashes rather than skepticism about AI. Operational reliability is a precondition for trust, not a separate dimension. A tool that fails mid-session cannot sustain a credible experience, no matter how good the design is.
From Research to Product
The takeaway was not “replace forms with AI.” It was a staged model: keep the form for simple, routine requests, and route complex, constraint-heavy orders through conversation where real-time checking actually matters.
The staged model wasn’t just a recommendation, I used it as the design foundation for a native iOS prototype built specifically for Event Organizers. The dual-entry homepage directly implements the research finding: AI chat (Alex) for fast, conversational intake where constraint-checking matters; a structured request form for planners who need full control over complex briefs. Both paths converge on the same event brief.
Key design decisions traced back to the study:
- Proactive dietary headcount, because dietary capture was exactly where the chatbot needed to be stronger.
- Budget shown as per-person in real time, because budget compliance was where AI most outperformed the form.
- Event card states tied to lifecycle (Matching → Quotes in → Booked → Done), because the research showed users needed visibility into where their order stood.
Reflection
- This project changed how I think about AI in product design. The question is always whether conversational AI reduces cognitive load, improves decision quality, and earns enough reliability to be useful in real workflows. When it clears that bar, it is powerful; when it does not, a simpler interface is the better design, and knowing the difference is the actual design work.
- The strongest recommendation was a staged model: keep forms for simple requests, route complex ones through the chatbot where its constraint-checking actually matters.
- The prototype turned the staged model from a recommendation into something testable with real users. The next step is a usability study on the iOS flows to validate whether the dual-entry design actually reduces friction for complex events, and whether the form path remains necessary as AI intent-parsing improves.
Full report available upon request.
