Full AI Agent–Powered RFQ Portal

Context & My Role

The RFQ (Request for Quotation) portal is the procurement heart of HikeOn's eCommerce ERP. Before this project, the RFQ process at HikeOn was almost entirely manual: buyers emailed vendors, tracked responses in spreadsheets, and manually compared pricing across hundreds of SKU variants. It was slow, error-prone, and impossible to scale.

This project was the most ambitious AI design challenge on the platform — designing a conversational AI Agent that sits at the centre of the RFQ workflow and automates the parts of procurement that don't require human judgment.

The Challenge

Traditional RFQ processes break at scale. When a buyer is managing 50+ active RFQs across multiple vendors and thousands of SKU variants, the administrative overhead consumes the time that should go into vendor relationships and strategic decisions.

The deeper problem: buyers weren't making bad decisions because they lacked data. They were making rushed decisions because they spent most of their time managing process, not analysing options.

"The AI Agent's job isn't to replace the buyer's judgment. It's to give the buyer back the time to actually use it."

Process & AI Initiative

This project required the deepest discovery work of any feature on the platform. Spent 4 sessions mapping the existing RFQ process end-to-end with the procurement team — documenting every manual step, every decision point, and every place where information was delayed or incomplete.

Identified 3 categories of work in the RFQ process

• High-judgment work (vendor selection, negotiation strategy, final approval) — keep human
• Medium-judgment work (variant configuration, initial pricing targets) — assist with AI
• Low-judgment work (follow-ups, status tracking, format standardisation) — automate fully

The AI Agent was designed to own the third category entirely, assist with the second, and prepare information for the first.

Design process

• Mapped RFQ workflow and categorised every step by judgment level
• Defined the AI Agent's scope — what it handles autonomously vs. what it escalates
• Designed the conversational interface: natural language input, structured output
• Built vendor shortlist logic with the AI team: matching criteria, scoring model, data sources
• Designed the approval workflow UI — clear handoff points from AI to human decision
• Prototyped the full flow in Figma, tested with 2 buyers across 3 rounds

Key Solutions

A conversational interface where buyers type or dictate requirements: product category, quantity ranges, target pricing, and delivery timeline. The AI Agent generates a complete RFQ draft — with variant breakdowns, suggested vendors from the approved list, and benchmark pricing from historical data — in under 30 seconds.

The draft is fully editable before sending. Buyers review, adjust, and approve with a single click. Vendor follow-ups are automated: the system tracks response deadlines and sends reminders without buyer intervention.

Once responses arrive, the AI consolidates them into a comparison view with a recommended shortlist ranked by a configurable scoring model (price, lead time, reliability history). Buyers make the final vendor selection from a structured decision view rather than a spreadsheet.

The approval workflow has two tiers: standard RFQs proceed automatically to the buyer's review queue; high-value or out-of-policy RFQs escalate to procurement management with a full audit trail attached.

Impact

End-to-end RFQ cycle time reduced by 65%. Buyers now handle 4× more active RFQs per day without additional headcount. The administrative work that previously consumed the majority of a buyer's day has been reduced to review and approval.

This is the fifth live AI feature on the platform — and the most strategically significant. It demonstrates that the AI initiative isn't limited to surfacing insights; it can own complex, multi-step workflows end to end while keeping humans in the decision seat where it matters.

Buyers describe the change as "finally doing the job I was hired to do" rather than managing process.

What I Learned

The most important design principle I developed on this project: the AI Agent's output needs to be immediately auditable. Buyers accepted AI-generated RFQ drafts only when they could see exactly what data drove each recommendation — vendor history, pricing benchmarks, policy rules. When the reasoning was visible, trust followed quickly. When it wasn't, even good recommendations got rejected.

The second lesson: design for the exception before the happy path. The most important screens in this flow aren't the ones that work smoothly — they're the escalation screens, the rejection flows, and the override states. Those are where buyer trust is won or lost.