Home/JCP/AI Cataloging

AI Cataloging.

Reliance brands are catalog-bound. Enrichment used to be slow (days to weeks per brand), expensive (~₹100 per SKU when vendor-led), and entirely manual. AI Cataloging removes that bottleneck.

Powered by the AI Cataloging Agent → · 7-step pipeline · 15 days → 5 hours per shoot · 3-layer eval · /agents/ catalog.

A single pipeline — Google AI + Web Grounding — measured against the existing manual processes at Netmeds and Reliance Digital, then deployed to production for Tira × Akind. Below is the chain of evidence in three steps: measured, shipped, how we know.

What's on this page

Before 15 days · After 5 hours.

AI-powered catalog creation for faster, scalable, and consistent commerce. Pipeline: Mannequin Image → AI Photoshoot Studio Visual → AI Enriched Catalog → AI Generated Video.

Scalable & Fast
15 days → 5 hours · per-brand pipeline
Face Consistency
AI-generated models library across ethnicity, age, size, gender
AI Photoshoot Studio
Mannequin → studio-visual on a single template
AI Generated Video
Better sales conversions vs static catalog
AI Photoshoot AI Photoshoot AI Photoshoot AI Photoshoot

Same pipeline, two operating models, consistent uplift.

Web Grounding was measured against two existing catalog-enrichment processes — one fully outsourced to vendors (Netmeds), one run by an in-house team (Reliance Digital). Different starting points, different bottlenecks, but the pipeline delivers in both: time collapses, marginal cost approaches zero, first-pass coverage jumps.

Netmeds Study · vendor-replacement scenario

Netmeds · vendor dependency eliminated.

~95%+ enrichment-time reduction · ~5 days → ~2 minutes per 1,000 SKUs · ₹100/SKU → near-zero marginal cost.

Catalog enrichment at Netmeds was fully outsourced — ~40 hours to enrich 1,000 SKUs across 6 attributes, ~₹100 per SKU. Web Grounding compresses the same job to ~2 minutes, automated. Vendor dependency goes away; cycle time goes from days to minutes.

MetricAs-Is (Vendor-led)With Web GroundingImpact
Time / SKU 40 hours per 1,000 SKUs for 6 attributes · ₹100 per SKU · fully dependent on 3rd-party vendors ~2 minutes per 1,000 SKUs ~95%+ time reduction · near-zero marginal cost · vendor dependency eliminated
Days to clear 1,000 SKUs* ~5 days ~2 minutes Cycle time reduced from days → minutes

* Days to clear 1,000 SKUs assumes the prior vendor-led pace.

Reliance Digital Study · in-house team augmentation scenario

Reliance Digital · same team, more output.

35–40% faster per SKU · +50–64% throughput on the same team hours · 75% first-pass attribute auto-fill.

Reliance Digital runs enrichment in-house — a 5-person team at 90 minutes per SKU. Web Grounding cuts per-SKU time to 55–60 minutes, lifting daily throughput by ~half without adding people. The same hours either unlock more capacity or take cost out — the choice is the operator's.

MetricAs-is (Manual)With Web GroundingImpact
Time / SKU 90 min 55–60 min 35% to 40% time saved
Team throughput / day* 15–20 SKUs 28.5–32.7 SKUs +50% to +64% additional SKUs enriched
Days to clear 100 SKUs* 5.0–6.5 days 3.5–4 days Faster cycle time
Attribute fill rate by web grounding 75% High first-pass coverage

* Team throughput / day and Days to clear 100 SKUs assume current team strength.

Pick the variable to hold constant
A · Capacity unlock

Keep the team constant.

Team strength of 5 members. Each member ships more.

Additional SKU enrichment / member
+7.5 to +12.7 / day
≈ +37 to +64 SKUs / week / member
B · Cost / time takeout

Hold output at 15–20 SKUs / day.

Same throughput, fewer person-hours.

Person-hours saved
7.5–11.7 h / day
Annualized (250 days): 1,875–2,917 h saved

What both studies show: the pipeline collapses time and cost regardless of the operating model. The next step is whether it actually ships into a live storefront. ↓ See the Tira × Akind release.

From measured to deployed.

The same Web Grounding pipeline benchmarked above is now live inside Tira's production environment for the Akind brand, in collaboration with Google. The Tira team consumes the enriched attributes in production, coordinated with the Akind brand. This is the first release-noted milestone; it validates that the pipeline doesn't just measure well — it ships.

Tira Akind Google Live · in production
Key results · Akind brand
Time taken for enrichment
~14 sec
Google AI · for all SKUs
vs. Tira team manual · 10–15 min per SKU
Attributes filled
332 / 382
With Google AI · 87% overall fill rate
Mandatory · 100% · Non-mandatory · ~91%
Manual baseline (Tira team) · 262 / 382
Accuracy vs. ground truth
89%
Google AI
Why this matters

Tira maintains catalog data in production with the AI enrichment pipeline · less manual effort · validated reliability · scope to extend to full Tira catalog in the next phase.

Next phase · expand AI enrichment to the entire Tira catalog.

The 89% accuracy claim above is measured against a controlled ground-truth set per attribute (see §04 Methodology). The same pipeline that produced it was featured on the Google Cloud Next '26 keynote stage — the catalog-as-intelligence-layer foundation behind JIIA on JioMart. ↓ §03 Showcased.

Catalog enrichment with Gemini.

Live · keynote 29-Apr-2026

Catalog Cloud now combines Gemini multimodal vision with web grounding and CAM sheet intelligence to process millions of product images and generate structured attributes that source systems didn't reliably carry — colour, style, material, usage context (occasion, intent fit), and deeper product semantics beyond titles and descriptions.

A SKU stops being a database row and becomes a queryable entity an AI can reason over. This is the foundation that makes systems like JIIA on JioMart work at scale.

Inputs

Three signal sources

Gemini multimodal vision (product images) · web grounding (open-web reference data) · CAM sheet intelligence (vendor catalog masters).

Output

Structured attributes

Colour · style · material · usage context (occasion, intent fit) · semantic enrichment beyond title/description fields.

Reach

Designed as single source of truth

Designed to feed Reliance Digital · JioMart · Tira · AJIO · Netmeds with the same attributes, vocabulary, and channel-agnostic API.

Live Tira × Akind   ·   Measured Netmeds, Reliance Digital   ·   Build at-scale rollout

What this delivers
Why this matters

Reliance Retail was selected as one of a handful of customer deployments featured on the Google Cloud Next '26 opening keynote alongside Vodafone, CME, Macquarie Bank, Colgate-Palmolive, and Citi. The catalog-enrichment layer is what makes JIIA's natural-language commerce work — without rich, queryable attributes the agent has nothing to reason over.

Attribute Accuracy.

Every accuracy number on this page (including the Tira pilot's 89%) comes from this method: each enriched attribute is compared to a ground-truth value and bucketed into one of three verdicts.

Exact match or similar

Ground truth and enriched value both have the same data.

Partial Match

AI response has correct values but is missing some values from ground truth. Specifically useful for multi-select attributes.

No Match

AI response is missing all values in ground truth.

Sample analysis · observations on the No Match bucket
Sample calculator · how a verdict is reached
Ground truth (attribute · value) Enriched value Verdict Explanation
Primary_colour · Gold Primary_colour · Gold Exact match or similar Ground truth and enriched value both have the same data, 'Gold'.
Concern · Pores, Dullness, Blemishes Concern · Dullness Partial Match AI response only mentions Dullness, missing Pores and Blemishes.
Benefits · Lightweight, Moisturising, Nourishing, Smoothening Benefits · Sun Protection No Match AI response is missing all of the benefits.
← View the full JCP track