Carrefour Taiwan
customer lifetime value and churn models to focus marketing effort on the customers worth keeping
Carrefour Taiwan built four predictive models on Google Cloud AutoML (customer lifetime value, segmentation, conversion, churn) to focus its advertising on the customers worth keeping, lowering its cost per action by 40% and raising its return on ad spend to 2.64 times its prior level.
Key points
- Four predictive models (customer lifetime value, segmentation, conversion, churn) to target the customers worth keeping.
- Built on Google Cloud AutoML and BigQuery, activation through Google Ads.
- Cost per action -40%, return on ad spend x2.64, app with more than 5 million downloads.
- Evidence level B, status mixed signals.
Objective
Focus marketing spend on the highest-value customers and those at risk of leaving, using predictive models to identify who is worth addressing, in order to lower acquisition cost and raise return on ad spend.
The deployment
Carrefour Taiwan loaded its customer data into Google Cloud AutoML starting in 2019 and built four machine learning models: customer lifetime value, segmentation, conversion rate prediction, and churn rate prediction. By targeting advertising primarily at the high-value customers identified by these models, the retailer lowered its online advertising cost per action by 40% and raised its return on ad spend to 2.64 times its prior level. The infrastructure later migrated to Google Cloud (migration completed in July 2021), with the app passing five million downloads in June 2022. The setup is part of a broader Carrefour group data modernization on Google Cloud, where BigQuery serves as the analytics foundation and the group claims 104 million households reached per year worldwide.
Results Proof B
Two quantified Google Cloud customer stories attributed to named executives (Henry Ting in Taiwan, David Kestermans in Belgium) documenting the predictive models and media results. Official interested sources (the cloud vendor), with no independent press corroboration of the Taiwan figures.
How it works
Documented architectureThe stack in detail
- plateforme Google Cloud AutoML Automated training service on which the four models (customer lifetime value, segmentation, conversion, churn) were built from 2019.
- infra BigQuery Analytics warehouse underpinning the customer data, within the Carrefour group's data modernization on Google Cloud.
- outil Recommendations AI Google Cloud product recommendation service used on the app and the e-commerce site.
- plateforme Google Ads Activation channel: online advertising is focused on the high-value customers identified by the models.
How it runs, concretely
For ops teams-
1Loading the customer data Data team
Customer data is sent into AutoML for model training.
-
2Building the models AI
Four models are trained: customer lifetime value, segmentation, conversion, churn.
-
3Identifying the customers to keep AI
The models flag the high-value customers and those at risk of leaving.
-
4Media targeting Marketing
Online advertising is concentrated on these customers through the Google Ads integration.
-
5Measurement and retraining Data team
Cost per action and return on ad spend are tracked and fed back to the models.
The purchase and behavior history that feeds the CLV and churn models. Without rich enough data per customer, the predictions degrade and targeting falls back on broad, unprofitable segments.
How your customers perceive this type of use
Sourced studiesC'est la famille la moins acceptee : 68% des Americains jugent inacceptable un score financier personnel calcule par algorithme et 67% l'analyse video automatisee d'entretiens d'embauche (Pew Research, 2018). La demande d'explication et de recours est massive : 83% veulent savoir quelles donnees l'IA utilise et 91% veulent pouvoir corriger des donnees erronees (Consumer Reports, 2024). A l'echelle mondiale, seuls 46% se disent prets a faire confiance aux systemes d'IA et 70% jugent une regulation necessaire (KPMG / Universite de Melbourne, 2025).
Acceptance conditions
- Transparence sur les donnees utilisees : 83% des Americains la reclament (Consumer Reports 2024)
- Droit de correction des donnees erronees : 91% le demandent (Consumer Reports 2024)
- Explication de la logique de decision : 44% des consommateurs sont plus enclins a utiliser un agent IA si sa logique est clairement expliquee (Salesforce 2024)
- L'acceptabilite depend du contexte de la decision : 50% des Americains jugent equitable un score de risque criminel pour la liberation conditionnelle, contre 32% pour un score financier applique aux consommateurs (Pew Research 2018)
Red lines
- La decision opaque et sans recours sur l'emploi, le credit ou le logement : 45% tres mal a l'aise pour l'embauche, 39% pour le pret, 39% pour le logement (Consumer Reports 2024)
- Le scoring des personnes a partir de donnees comportementales : 68% le jugent inacceptable pour les offres financieres (Pew Research 2018)
Sources: Pew Research Center 2018 · Consumer Reports 2024 · KPMG / Universite de Melbourne 2025 · Salesforce 2024
How to replicate
Inference, not sourcedData prerequisites
- purchase and behavior history per customer
- enough volume to train CLV and churn models
- reliable linking of purchases to a customer identifier
Org prerequisites
- data team able to train and maintain models
- integration between models and the media platform
- a cost per action measurement loop
Possible stack
- AutoML or equivalent (Vertex AI, cloud ML)
- data warehouse (BigQuery, Snowflake)
- connection to the advertising platforms
The plan, step by step
- Step 1Consolidate purchase and behavior history in the warehouse, linked to a reliable customer identifier.Deliverable: Unified customer table, ready for training.
- Step 2Train a first customer lifetime value model (AutoML or equivalent) and validate it offline against history.Deliverable: CLV scores per customer, evaluated and documented.
- Step 3Add the churn and conversion models, then define the segments to address first.Deliverable: Scores in batch production and actionable segments.
- Step 4Connect the segments to the advertising platform as first-party audiences and launch the targeted campaigns.Deliverable: Campaigns focused on the customers to keep.
- Step 5Compare cost per action and return on ad spend against prior targeting, and set up periodic retraining.Deliverable: Quantified media review and retraining pipeline.
First step: Train a customer lifetime value model on purchase history to identify the customers to prioritize in media targeting.
The same pattern, proven in other industries
Every Man Jack
predictive replenishment timing (predicted next order date)
Maersk
predictive monitoring of refrigerated cargo (customer visibility)
T-Mobile US
Intent-driven decisioning platform: an engine reads customer intent and sentiment in real time, decides the next action, and can execute it, beyond a simple chatbot
Sources
- S1 Carrefour Taiwan Case Study - Google Cloud Interested party archive pending
- S2 Carrefour - Google Cloud Interested party archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.