— Surface 06 · Proof · Giant Eagle Q1 2026

A $9.3B grocer.
216 stores. Already in market.

In Q1 2026, Giant Eagle — a $9.3B regional grocer running primarily on Azure + Databricks — validated Delectable AXP across 216 stores. Delectable bridged the Azure side to GCP, ran the enrichment on Vertex + BigQuery, and converted a non-GCP customer into a Tier-2 Vertex AI consumption account. 70,000 products enriched from 8 attributes each to 42 · catalog grew from 560k data points to 2.94M in 4 weeks · Gemini Pro reasons across it. The numbers below are what John's team can take into Kroger, Albertsons, Wegmans, and Publix next week.

Catalog scope
70k
SKUs enriched · every product, every store. Nightly refresh on BigQuery + Vertex.
Attribute depth
8 → 42
Per-SKU dimensions. Now Gemini knows: gluten-free flag, lineage, sodium tier, allergen graph — 42 dimensions.
Total data points
2.94M
From 560k baseline — 5.25× grounding multiplier per Gemini call. Higher relevance, lower hallucination, identical Gemini cost.
Time to delivery
4 wk
Weeks, not quarters. SOW signed Monday → 70k SKUs enriched + Vertex catalog live → 4 weeks later.
— Downstream outcomes · what Gemini alone could not do

The agent actually shipped.

The mandate from Giant Eagle's CMO was simple: stop debating proofs-of-concept, ship a real agentic basket. Delectable enriched the catalog in 4 weeks, joined Eagle Eye loyalty + mParticle into the Shopper HyperGraph, then handed it to Gemini Pro. Here's the math on production traffic.

Basket lift
+18%
Measured AOV uplift on Delectable-served carts vs. control. Statistically significant at p<0.01.
Response latency
490 ms
vs. 4.0s Gemini-only baseline. 8× faster · sub-second perception · acceptable for chat UI.
Per-call efficiency
$0.0024
vs. $0.033 raw Gemini. Same Gemini Pro model · fewer tokens · grounded inputs. The unit economics that flip this from shelved POC to shipping product — driving 1.2M calls/mo that wouldn't otherwise exist.
Projected ROI
460%
Year-1 on $1.0M Phase 1 investment. Basket lift × volume × margin = the recoverable.
— The 4-week sprint · what actually happened

Day 1 to Day 28. No drama.

Wk 1

Catalog ingest

Read-only access to Giant Eagle's Salsify PIM. 70k SKUs streamed into BigQuery enrichment_warehouse dataset in 14 hours.

Wk 2

Food HyperGraph applied

Each SKU matched against 142k canonical ingredients. ~50k Gemini Pro calls for edge cases. Daily quality review with Giant Eagle's data team.

Wk 3

IA ontology layered

Dietary, allergen, occasion, cultural vectors applied on top. 27 cross-cutting taxonomies. Food scientist QA pass on the dietary flags.

Wk 4

Production cutover

Output to BigQuery retail_analytics + Vertex Retail Search catalog + Merchant Center feed. Gemini Pro grounded in the new corpus. First agentic basket served at 14:32 EST.

— Cross-cloud capture · the Google quota story

Giant Eagle ran on Azure.
Delectable made the workload run on GCP.

The slide an FSR pitches to a grocer who "is already on Azure" or "standardized on AWS." Delectable's multi-tenant platform has the bridges built — we ingest from Azure Synapse, Databricks, Redshift, Snowflake, S3, ADLS — and run the agentic-basket workload on Vertex + BigQuery + Cloud Run. The grocer's existing data stack stays where it is; the new workload — the one Google is competing for — lands in the grocer's new GCP project. Even a non-GCP customer becomes a Vertex AI consumption account.

— Where Giant Eagle's data lived

Azure + Databricks.

Salsify PIM feed landing in Azure Data Lake Storage. Loyalty + transaction data in Databricks Delta tables. Merchant Center feed pipeline running on Azure Functions. The grocer had years of investment in the Azure stack — no plan to migrate it.

— Where the new workload runs

Vertex + BigQuery + Cloud Run.

Delectable's ingest bridge pulls from ADLS + Databricks into BigQuery enrichment_warehouse. The Food + Shopper HyperGraphs run on Memgraph. Gemini Pro + Vertex Retail Search resolve every basket. Cloud Run autoscales the agent. Every Gemini call lands in a GCP project Google bills.

The FSR talking point → "We don't need you to move off Azure. We need one project — your new agentic-basket workload — to land on GCP. Delectable's bridge does the rest." Net Vertex AI consumption captured per Tier-1 grocer Delectable lands on a non-GCP-native customer: $1.6M+ / yr, that would otherwise stay zero on Google's quota.
— Quote · Giant Eagle production benchmark · March 2026

"Delectable lets Google deliver Gemini-quality reasoning at sub-second latency, at a fraction of the token cost — protecting Google Cloud's margin while making the grocer's basket actually ship."

Delectable × Giant Eagle production benchmark · March 2026 · 1.2M requests

Same playbook · ready for the next grocer.

→ Co-sell playbook → Catalog enrichment