Fynd Commerce
Agentic AI
Enterprise SaaS
B2X Commerce
Led the entire Fynd Commerce ecosystem with it’s shift to supervised autonomy: merchant copilot + AI storefront creation.
Year
2023 - 2026
Services
End-to-End Product Design, AI UX Strategy, Platform Architecture, Research, Design Ops
At a glance
The Platform: Fynd Commerce is an AI-native unified commerce platform. At scale, it serves 300M+ customers, powers 20K+ stores and 2,300+ brands, managing 720B total inventory units and processing 3.2M orders annually.
My role: Head of Design / Design DRI. I own the Fynd Commerce Platform end-to-end (across OMS, logistics, payments, catalog, storefronts, and extensions), with a focused charter to drive the platform's AI transformation layer.
Team: I lead a team of 5 product designers across distinct platform modules. I personally owned the AI transformation layer end-to-end — Sidekick, AI Theme Editor, and Order Flow Builder, from UX architecture to final UI. Catalog, Payments, Logistics, and Storefronts I directed and reviewed as Design DRI, with individual designers holding module ownership.
Case Study Scope: Rather than breaking down every platform module, this case study focuses specifically on my charter to architect Fynd's AI-native transformation layer. However, it also highlights my foundational work as the Design DRI for OMS, showcasing core platform differentiators—such as the deterministic Order Flow Builder—that uniquely enable Fynd to resolve complex orders and logistics and scale for massive enterprise clients.
Problem (why AI, now)
At FCP’s massive scale, merchant operations were failing for two predictable reasons:
High-frequency tasks are manual (search, configuration, exception handling).
High-stakes tasks can’t be fully automated without trust, reversibility, and auditability.
Before proposing the AI layer, I spent 3 weeks shadowing merchant ops teams across 4 brands (Diesel, GAS, Superdry & Celio). The pattern was the same everywhere: power users had built elaborate workarounds in spreadsheets for tasks the platform was supposed to handle. That told me the problem wasn't feature gaps — it was that the platform assumed a level of technical fluency most merchants didn't have. That's what led to the Supervised Autonomy framing.
I framed the AI transformation as a product system problem: How do we drastically reduce time-to-launch and time-to-resolution while maintaining absolute correctness across inventory, pricing, fulfillment, and storefront integrity?
Strategy: “Supervised Autonomy” (agentic, but accountable)
I designed FCP 2.0 around a single, radical operating principle: AI should propose. Humans should approve. Systems should execute.
Instead of replacing the user, we transitioned them from manual operators to AI supervisors. This became a repeatable, system-wide UX loop:
Observe (context + state)
Propose (action plan)
Preview (diff / impact)
Approve (permissions + confirmation)
Execute (API/workflow)
Audit + rollback (traceability)
Intervention 1: Sidekick (Merchant Copilot)
I led the UX architecture for Sidekick, turning the static admin dashboard into an “actionable workspace.”
The Execution Engine: Designed the interaction model to convert natural language directly into complex operational workflows. Sidekick actively executes multi-step tasks—such as creating products, orchestrating email campaigns, and building smart collections—rather than just returning text responses.
Context-Aware Intelligence: Architected the system to explain the "why" behind its actions, grounding every AI proposal in live platform data (e.g., real-time inventory states, pricing rules, and fulfilment constraints).
Architecting Trust (The Guardrails): To make AI execution safe at an enterprise scale, I designed a strict "proposal → preview → approve" loop supported by core system guardrails:
Role-based permissions that dynamically restrict the agent's action surface based on user access.
Preview-first UI for any write operation, displaying clear before/after database diffs before execution.
System audit trails paired with 1-click state rollbacks for fully reversible actions.
Intervention 2: AI Theme Editor (Prompt → Storefront)
Our ecosystem is undergoing a massive architectural shift from Vue to React. I focused the AI Theme Editor experience on turning raw "prompt outputs" into merchant-safe, shippable React UI—scaling from full storefront generation down to surgical component edits.
Macro Generation (Prompt-to-Storefront): Integrated a full-scale generative UI engine natively into the editor, allowing merchants to generate entire, multi-page e-commerce storefronts from a single text prompt, slashing design-to-live time by 40%.
Micro Generation (Zero-to-Code Custom Sections): Designed an interface for merchants to construct bespoke UI components via natural language. The AI acts as a full-stack developer, instantly generating the required Data Source (GraphQL/REST), HTML template, and Custom CSS behind the scenes, while retaining no-code configuration over layout and behavior.
Architectural Safety: Enforced strict design constraints (responsive behavior, brand tokens, content bindings) within the new React architecture so the AI could safely orchestrate layouts without ever breaking the storefront's underlying integrity.
The Deterministic Foundation: Order Flow Builder
While the future of FCP leans heavily into Agentic AI, enterprise commerce still demands strict, deterministic logic for fulfilment. I conceptualised and designed the zero-to-one MVP for a node-based Order Flow Builder within the OMS to let sellers construct custom operational journeys.
Visualising the "State Tree": I architected a drag-and-drop canvas where operations teams could map out exact fulfilment states (e.g., Placed → Manual Invoice → Pack), assigning strict admin approval loops to sensitive transitions.
Power-User Tooling: To remove the engineering bottleneck, I designed contextual side-panels that exposed Form Builders, JSON payloads, and dynamic "Pre/Post-hooks." This gave sellers code-level control over their logic wrapped in a highly usable interface.
The Boltic.io Evolution: The UX paradigm, State Tree architecture, and execution engine of this builder proved so universally scalable that it transcended the OMS. It was spun out to become the UX foundation for Boltic.io—a dedicated big-data and enterprise automation platform. Today, Boltic utilises this exact visual node-logic to connect databases and execute over 477 million workflows with a 105% YoY growth rate.
Platform Core: OMS & Logistics
Before layering Agentic AI onto the platform, the underlying digital plumbing had to be flawless. While my purview as Design DRI spanned the entire Fynd Commerce ecosystem—architecting end-to-end workflows across Catalog, Payments, Refunds, and Storefronts—the OMS and Logistics modules required the heaviest operational lifting. To illustrate this foundational scale, here is a curated showcase of three USP-defining features I re-engineered within the OMS to make real-world retail complexity operable across 3,000+ physical stores:
Omnichannel Order Ingestion: Designed unified flows enabling merchants to manually inject offline, walk-in traffic directly into the digital OMS. This bridged a critical omnichannel gap, seamlessly converting traditional retail footfall into trackable digital order volume.
Click & Collect Orchestration: Architected the end-to-end UX for in-store pickup. I translated complex backend logic into simple, front-line operational interfaces that manage store readiness, strict pickup SLAs, and real-time exception handling.
Multi-Piece Shipments: Designed the visual logic for complex fulfilment edge-cases, enabling split deliveries and partial fulfilment while maintaining unified, frictionless communication for the end customer.
The Impact: These UI and algorithmic redesigns drove 30% reduced delivery timelines, 27% reduced delivery costs, and a 50% reduction in RTO (Return to Origin) rates.
Tradeoffs and constraints (what made this real)
I made deliberate product calls to keep the AI trustworthy and shippable:
Trust over Magic: I explicitly chose an approval-first UX over full background automation to protect merchant trust and limit the "blast radius" of AI hallucinations.
Architecture as a Constraint: Treated the Vue-to-React codebase migration as an enabling constraint, designing the AI tooling to operate strictly within the bounds of the new React theme ecosystem.
Scoping: Intentionally scoped early AI agent capabilities to “high-frequency, reversible tasks” before expanding into higher-stakes financial actions.
What shipped vs what’s next
Shipped / In Active Rollout:
Prompt-to-Storefront Engine: Integrated a generative UI engine that allows merchants to orchestrate entire, multi-page storefronts (setup, layout, styling, media) from a single text prompt—taking brands live in minutes instead of weeks.
Zero-to-Code Component Builder: Shipped an AI tool that translates natural language (e.g., "testimonial carousel") into production-ready GraphQL, HTML, and CSS. An inline "Enhance Prompt" feature automatically engineers basic inputs to guarantee perfect code generation.
Embedded Generative AI: Scaled auto-generation across the ecosystem—from product catalogs to extensions. It instantly generates descriptions, interactive layouts, and images, slashing time-to-live.
AI Sidekick (Supervised Autonomy): Shipped a conversational execution engine that actively performs complex tasks (creating products, orchestrating campaigns). It operates strictly on a "proposal → preview → approve" model to guarantee user trust.
Execution Foundations: Established the core UX architecture ensuring full auditability and repeatability for all platform AI actions.
Next on the Roadmap:
Expand Sidekick’s action surface area across OMS, catalog, and storefront ops.
Standardise agent permissions, preview diffs, and rollbacks across all platform modules.
Transition from “assistant” to autonomous “agent” in bounded domains where absolute correctness can be guaranteed.
Learnings
I'd surface the audit trail UI in Sidekick's first release, not the second. Merchants hesitated to adopt the execution engine when they couldn't immediately see what it had done.
Module ownership only works when the design system is tight enough to absorb independent decisions. Ours wasn't early on — I'd invest in system alignment in sprint 1 itself.
Merchant teams who'd been burned by over-promised automation needed more trust-rebuilding than I'd planned for. I'd now treat merchant trust as a launch metric, not a post-launch problem.
