→

Back

SUCCESS STORY · ASTORE (ACCOR GROUP)

Astore｜The industrialized product benchmark on Snowflake

KPC has industrialized Astore's product benchmark on Snowflake. A scanned invoice becomes a quantified recommendation in just a few minutes based on actual data, not estimations.

Snowflake

Generative AI

Data Products

Data

Data Governance

Digital BU

The data impact

Days → min.

Duration of a product benchmark

750k

Astore catalog references

~20

Sales representatives using the tool

-15%

Measured savings on a quote compared to the Astore catalog (real case: €115,000 vs €97,000)

The data impact

Days → min.

Duration of a product benchmark

750k

Astore catalog references

~20

Sales representatives using the tool

-15%

Measured savings on a quote compared to the Astore catalog (real case: €115,000 vs €97,000)

The data impact

Days → min.

Duration of a product benchmark

750k

Astore catalog references

~20

Sales representatives using the tool

-15%

Measured savings on a quote compared to the Astore catalog (real case: €115,000 vs €97,000)

RESULTS

Results for Astore

From days to minutes: commercial decision-making accelerated

An uploaded invoice, a quantified recommendation in a few minutes. The sales representative focuses on the customer interaction. Excel analysis is eliminated from the workflow.

Results for Astore

Recommendations on real data

Comparisons are based on actual invoices and the Astore catalog, not on estimates. In a real-case scenario, the measured savings reached up to 15%.

Results for Astore

AI in production, invisible

Not a POC: an application used daily by sales representatives with no technical background. The AI works behind the scenes.

RESULTS

Results for Astore

From days to minutes: commercial decision-making accelerated

An uploaded invoice, a quantified recommendation in a few minutes. The sales representative focuses on the customer interaction. Excel analysis is eliminated from the workflow.

Results for Astore

Recommendations on real data

Comparisons are based on actual invoices and the Astore catalog, not on estimates. In a real-case scenario, the measured savings reached up to 15%.

Results for Astore

AI in production, invisible

Not a POC: an application used daily by sales representatives with no technical background. The AI works behind the scenes.

RESULTS

Results for Astore

From days to minutes: commercial decision-making accelerated

An uploaded invoice, a quantified recommendation in a few minutes. The sales representative focuses on the customer interaction. Excel analysis is eliminated from the workflow.

Results for Astore

Recommendations on real data

Comparisons are based on actual invoices and the Astore catalog, not on estimates. In a real-case scenario, the measured savings reached up to 15%.

Results for Astore

AI in production, invisible

Not a POC: an application used daily by sales representatives with no technical background. The AI works behind the scenes.

Context & central challenges

Les enjeux

A central purchasing body that must prove its savings, invoice in hand

Astore, the Accor group's central purchasing body, pools volumes to offer hoteliers prices lower than those of their usual suppliers. To convince a non-client hotelier, its sales representatives compare the latter's invoice to the Astore catalogue of 750,000 references, and calculate the potential savings.

This comparison used to be done by hand, in Excel, line by line: several hours, sometimes several days per request, with a result that depended on each individual's experience.

Go from several days to a near-immediate result per request

Making it usable by non-technical sales representatives, integrated into the sales cycle

Moving away from reliance on individual experience for consistent results

Base every recommendation on real data, not on estimations

Context & central challenges

Les enjeux

A central purchasing body that must prove its savings, invoice in hand

This comparison used to be done by hand, in Excel, line by line: several hours, sometimes several days per request, with a result that depended on each individual's experience.

Go from several days to a near-immediate result per request

Making it usable by non-technical sales representatives, integrated into the sales cycle

Moving away from reliance on individual experience for consistent results

Base every recommendation on real data, not on estimations

Context & central challenges

Les enjeux

A central purchasing body that must prove its savings, invoice in hand

This comparison used to be done by hand, in Excel, line by line: several hours, sometimes several days per request, with a result that depended on each individual's experience.

Go from several days to a near-immediate result per request

Making it usable by non-technical sales representatives, integrated into the sales cycle

Moving away from reliance on individual experience for consistent results

Base every recommendation on real data, not on estimations

Our approach

The entire pipeline in Snowflake, without data movement

From raw document to business deliverable, with no extraction off-platform. AI is placed solely where it creates value: extraction, normalization, matching, and remains invisible to the end user.

AI extraction & reading

From PDF invoice to actionable data

The invoice is loaded and automatically read by AI extraction. Product descriptions are normalized by an LLM to be comparable to the catalog.

Cortex AI

AI Parse Document

LLM Complete

Matching & scoring

Catalog reconciliation and prioritization

Each product is matched against the catalog via semantic search (Cortex Search), combined with SQL fuzzy matching and a LLM. The recommendations are then automatically scored and prioritized.

Cortex Search

SQL Fuzzy Matching

LLM Complete

Application & report

Business validation and PDF report

The sales representative validates the recommendations in a Streamlit application hosted in Snowflake. They then generate a PDF report with price differences and savings.

Streamlit (SPCS)

Snowflake

PDF Report

Our approach

The entire pipeline in Snowflake, without data movement

From raw document to business deliverable, with no extraction off-platform. AI is placed solely where it creates value: extraction, normalization, matching, and remains invisible to the end user.

AI extraction & reading

From PDF invoice to actionable data

The invoice is loaded and automatically read by AI extraction. Product descriptions are normalized by an LLM to be comparable to the catalog.

Cortex AI

AI Parse Document

LLM Complete

Matching & scoring

Catalog reconciliation and prioritization

Each product is matched against the catalog via semantic search (Cortex Search), combined with SQL fuzzy matching and a LLM. The recommendations are then automatically scored and prioritized.

Cortex Search

SQL Fuzzy Matching

LLM Complete

Application & report

Business validation and PDF report

The sales representative validates the recommendations in a Streamlit application hosted in Snowflake. They then generate a PDF report with price differences and savings.

Streamlit (SPCS)

Snowflake

PDF Report

Our approach

The entire pipeline in Snowflake, without data movement

From raw document to business deliverable, with no extraction off-platform. AI is placed solely where it creates value: extraction, normalization, matching, and remains invisible to the end user.

AI extraction & reading

From PDF invoice to actionable data

The invoice is loaded and automatically read by AI extraction. Product descriptions are normalized by an LLM to be comparable to the catalog.

Cortex AI

AI Parse Document

LLM Complete

Matching & scoring

Catalog reconciliation and prioritization

Each product is matched against the catalog via semantic search (Cortex Search), combined with SQL fuzzy matching and a LLM. The recommendations are then automatically scored and prioritized.

Cortex Search

SQL Fuzzy Matching

LLM Complete

Application & report

Business validation and PDF report

The sales representative validates the recommendations in a Streamlit application hosted in Snowflake. They then generate a PDF report with price differences and savings.

Streamlit (SPCS)

Snowflake

PDF Report