Finding Luxury SKUs in 60 Seconds, How 9AI Slashed Search Time by 96% for The House of Things
Business Impact at a Glance
The House of Things : Udaipur-based luxury furniture & décor marketplace (30k+ SKUs)
Operations Adaptation AI Growth Office
Design team spent hours manually matching client references to products; backlog hurt conversions
- — 96%faster product match
- — ₹7.5L monthly OPEX saved
- — +18% upsell rate
- — 100% catalog coverage
“The new visual search lets us wow clients in minutes instead of days.”
Pain Narrative
THOT curates 30k+ designer SKUs, but clients arrive with mood-board images or vague briefs (“mid-century caramel lounge chair”). Designers had to trawl Magento, open dozens of tabs, and still risk missing hidden gems. Average shortlist took 18 minutes, backlog ballooned, and premium clients lost patience—threatening brand equity and revenue.
Legacy Tasks- —Manually tag visual cues, guess keywords
- —Scroll 100+ product pages per query
- —Update option sheets & quotes by hand
AI-Powered Visual Search Platform
Discovery, SOP mapping, success metrics
Multi-modal embedding pipeline, Magento adapters
“Visual Search Bot” orchestrating ingestion → ranking → analytics
The platform ingests every SKU from Magento nightly, generating CLIP embeddings that blend product photography with enriched attributes (category, form factor, material, colour, era).
Designers drop a reference image or description; an agent detects objects, fuses text & vision vectors, applies weighted similarity, and returns top-10 matches plus look-alikes for upsell.
Usage, hits, and feedback loop to fine-tune weights are logged in the lead module for continuous improvement.
Strategy that Delivered Results
- — 8 designer interviews, 2 week shadow-work
- — 14 manual steps → consolidated into 4 automated stages
- — Pilot (5 users) → phased rollout (25 users) in 4 weeks
- — Lunch-&-learn workshops; in-app thumb-up/down training loop
- Input ingestion: Magento web-hooks, nightly CSV diff
- Embedding: OpenAI CLIP + custom attribute JSON
- Query engine: Vision boundary detection → vector search (PgVector)
- Portal: Next.js front-end, role-based auth
- Security: VPC-isolated, AES-256 at rest, audit trail to ELK
Business Impact at a Glance
| Metric | Pre-9AI | Post-9AI | Δ Impact |
|---|---|---|---|
| Avg. search time per query | 18 min | 0.8 min | ▼ 96% faster |
| Designer hours on search | 200 h/wk | 20 h/wk | ▼ 90% |
| Monthly labour spend | ₹10L | ₹2.5L | ▼ ₹7.5L |
| Client shortlist accuracy | Subjective | 95% Top-10 match | ▲ Precision |
| Cross-/upsell conversions | 8% | 26% | ▲ 18 pp |
Start with the audit.
Keep everything. Risk nothing.
The 4-week audit is your entry point. Even if you don't proceed, you walk away with the most detailed SOP your company has ever had, a validated AI strategy, and solution mockups. All yours to keep.
The first step is always a conversation.