Design Challenge · 4 días · Senior Product Designer
StockAI
Replenishment reimagined: from manual setup to AI-driven decisions
A 4-day Senior Product Designer challenge. A full redesign of the replenishment flow in a retail-intelligence SaaS — from a manual 2-step form to an experience where AI proposes, the user decides, and the system executes. Goal: 25 min → under 5 min per decision.
- Formato
- Design Challenge
- Rol
- Senior Product Designer · Solo
- Duración
- 4 días
- Plataforma
- Web
- Stack
- Figma · Shadcn UI · Poppins + Inter · AI tooling
El brief
Un producto que prometía IA y entregaba un formulario
StockAI is a retail-intelligence SaaS that helps merchandising teams allocate inventory across stores for thousands of SKUs. Its Replenishment feature was blocking sales conversion: it promised an "AI-driven" product but forced users to configure every replenishment by hand through a 2-step form — no proactive recommendations, no transparency, no visible financial impact. The result: stalled growth, low adoption, and a blocked sales process. Users saw StockAI as just another ERP, not an intelligence tool.
El cambio de paradigma
De configurar a aprobar
El producto no fallaba por mala UX — fallaba porque la IA era invisible. El rediseño parte de un solo movimiento: cambiar el rol del usuario. Dejar de ser quien configura para pasar a ser quien decide.
Antes — el usuario como operador
Después — el usuario como decisor
Discovery
La auditoría que originó el rediseño
I audited the existing flow against three lenses: value proposition (does it deliver on "AI-driven"?), user efficiency (does it cut time and cognitive load?), and business outcomes (does it create measurable impact?). I found 7 problems ranked by severity — 2 critical (invisible AI across every screen, heavy manual setup in steps 1–2) and 3 major (no visible revenue impact, zero AI transparency, no bulk actions). In parallel I mapped the personas from the brief: Maya (Inventory Manager, daily use, frustrated by repetitive clicks, can't see financial impact, depends on the planner for the big decisions) and David (Merchandising Planner, weekly use, distrusts black-box AI, has to justify decisions to the CSCO, still finds Excel faster for analysis).
De los 7 problemas, estos cinco —los críticos y mayores— guiaron el rediseño.
La IA era invisible
No aparecía en ninguna pantalla del flujo. El producto se veía como un ERP, no como inteligencia.
Configuración manual pesada
Los pasos 1 y 2 obligaban a configurar a mano cada replenishment, sin una sola recomendación.
Sin impacto de revenue visible
La pantalla de revisión mostraba unidades, nunca el impacto financiero de la decisión.
Nula transparencia de la IA
Cuando había cálculo, no se explicaba la lógica. Imposible confiar en una caja negra.
Sin acciones masivas
Cada tienda se aprobaba de a una. Inviable para quien gestiona decenas de tiendas.
Las personas
Dos usuarios, un solo producto
Maya y David usan StockAI de formas opuestas. Servir a ambos sin duplicar el producto fue el marco de todo el rediseño.
Maya
Inventory Manager
Uso diario · sesiones de 15-30 min
- Ejecuta el día a día del inventario
- Frustrada con los clics repetitivos
- No ve el impacto financiero de sus decisiones
- Depende del planner para las decisiones grandes
David
Merchandising Planner
Uso semanal · sesiones de 1-2 horas
- Define la estrategia de allocation
- Desconfía de la IA tipo caja negra
- Debe justificar cada decisión al CSCO
- Excel le sigue resultando más rápido para analizar
Decisiones
Tres decisiones que definieron el rediseño
A paradigm shift: from configuring to approving
The product was not failing because of bad UX. It was failing because the AI was invisible. I shifted the flow from "configure → calculate → review" to "AI proposes → user decides → system executes." That completely redefines the user's role: from operator to decision-maker.
Dual mode, one product
Instead of building two products, the same flow adapts: Inventory Managers enter through "Quick Approve" from the dashboard (review in seconds); Merchandising Planners enter through "Smart Setup" with AI-prefilled defaults (tailored configuration when it is needed). Both paths converge on the same Review & Approve screen — which cuts build cost and the learning curve.
One core moment, 60% of the effort
In a time-boxed challenge I prioritized depth over breadth. The home dashboard and the quick path stayed at medium fidelity. The Review & Approve screen is the only one at high fidelity, because that is where the product wins — or loses — the demo and the conversion.
Ejecución
Tres pantallas que reconstruyen el flujo
AI Insights Dashboard — the urgent first
I replaced the static list with an actionable home. A yellow banner up top surfaces 3 urgent opportunities worth +$340K in revenue impact right away. 4 KPI cards translate the data into context: active replenishments, pending AI reviews, revenue opportunity, and AI acceptance rate. Status pills (Critical / Review / Approved) and confidence bars replace the flat list — users know where to start without reading.
Quick Path — Inventory Manager
Urgency color-coded on each card's border (red critical, yellow review, green routine). An inline approve button on every card — Maya never has to open the detail page. The "Approve all (12)" bulk action closes 12 stores in one click. A mechanism that turns 30-minute sessions into 5-minute ones.
Smart Setup — Merchandising Planner
AI first, manual as backup. A yellow "AI Suggests" card above the form shows scope (West Coast), category (Footwear), frequency (Weekly), and estimated impact (+$340K · 87 SKUs) — all prefilled from patterns. Power users can still adjust every field. No blank initial empty state.
AI Review & Approve
The screen where StockAI proves its worth. Revenue impact ($340K) and overstock avoided ($85K) instead of operational data. Confidence as a first-class signal: percentage + color bar, filterable and sortable. Inline transparency: expanding a row reveals the 4 data points behind the logic (sales velocity, stock level, seasonal pattern, similar stores). Bulk approval with guardrails — it only applies to items above 80% confidence; below that threshold, individual review is required.
$340K
Impacto en revenue de la decisión
$85K
Sobrestock evitado
60%
Del esfuerzo del challenge, en esta pantalla
Resultados
El impacto proyectado
25 → 5
Min per decision (goal)
+60%
Replenishments per user / week
>70%
Target AI acceptance rate
4 days
Challenge duration
Métricas objetivo definidas en el challenge — un ejercicio de diseño, no datos de producción.
Aprendizajes
Lo que me llevo
- 01
If a product is sold as AI-driven but the experience is an ERP-style form, the gap with a spreadsheet disappears — and conversion goes with it.
- 02
Building trust in AI takes transparency, not hiding it: visible confidence, logic you can expand inline, and an override that is always within reach.
- 03
Serving two personas does not mean two products. The path adapts the entry point; the core moment converges.
- 04
In a time-boxed challenge, one pixel-perfect core moment beats five average screens. That is what wins demos and turns trials into customers.
¿Seguimos con otro proyecto?