Vodafone
Industrialized churn prediction: cancellation models trained per market on a group MLOps platform, with standardized templates replicated from country to country to trigger proactive retention actions
Vodafone industrialized churn prediction across more than 8 countries through AI Booster, its MLOps platform on Vertex AI, cutting model time-to-market by 5x and moving from PoC to production in 4 weeks instead of 5 months.
Key points
- Industrialized churn prediction, one model per market replicated from country to country.
- AI Booster MLOps platform on Vertex AI, XGBoost baseline, unified data on BigQuery.
- Time-to-market cut by 5x, PoC to production in 4 weeks instead of 5 months.
- Evidence level B, confirmed status, thousands of models per day across 8+ countries.
Objective
Retain customers before they leave, market by market, and stop rebuilding a churn model from scratch in every country.
The deployment
Vodafone built AI Booster, an internal model-productionization platform built on Vertex AI and connected to Neuron, its group data warehouse. The founding use case is cancellation prediction. The Big Data & AI team builds a churn model for one market from usage, billing, contract, and interaction data, pushes it to production through standardized pipelines, then reuses the same template in another country without starting over. The resulting scores feed retention campaigns and offer personalization. The platform runs thousands of models per day across more than eight countries, and moves a model from trial to production in four weeks instead of five months. One point of honesty: the public figures cover the speed and scale of production, not churn points avoided, which Vodafone does not disclose.
Results Proof B
Documented by official Google Cloud sources and an integrator (Datatonic) with named Vodafone spokespeople - solid but operational figures (MLOps speed/scale): no churn impact in percentage points is public, and the sources are on the vendor side.
How it works
Documented architectureThe stack in detail
- plateforme Google Cloud Vertex AI Foundation of the internal AI Booster platform: training, pipelines, and serving of the per-market churn models.
- outil XGBoost Classification algorithm used as a baseline for the cancellation models.
- infra Google BigQuery / BigQuery ML Warehouse of the group data ocean (Neuron) that unifies usage, billing, contracts, and tickets.
- infra Google Cloud Build CI/CD of the standardized pipelines LAB to STAGING to PROD, replicated from country to country.
- outil AI Booster Vodafone's internal MLOps platform built on Vertex AI: thousands of models per day across 8+ countries, PoC to production in 4 weeks.
- integrateur Datatonic Integrator that built the AI Booster platform with the Vodafone teams.
How it runs, concretely
For ops teams-
1Unify the customer data Data team
Usage, billing, contracts, and support tickets gathered in one place, with at least twelve months of depth. Without this foundation, no reliable model.
-
2Train and score AI / data science
A model learns to spot departure signals and assigns each customer a churn probability. An XGBoost is often enough as a starting point.
-
3Hand off to local teams Local marketing / CRM
The scores go into the CRM. Country marketing decides what to do with at-risk customers: an offer, a call, a commercial gesture.
-
4Always keep a control group Data team + marketing
You do not contact 100% of at-risk customers. Part of them serves as a control, otherwise it is impossible to know whether the retention comes from the action or from chance.
-
5Monitor drift and retrain AI / MLOps
Customer behavior changes; a model left as is degrades silently. The platform tracks this drift and triggers retraining.
The historical cancellation label. If it is poorly defined (when do we consider that a customer has churned?), the whole model is biased, whatever its stated accuracy.
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
- Unified customer history (usage, billing, contracts, tickets, interactions)
- Reliable cancellation label, 12+ months of depth
- GDPR governance of the profiling
Org prerequisites
- Data science team + marketing activation loop
- Ability to run controlled tests (control group)
Possible stack
- Vertex AI or equivalent (Databricks, SageMaker, Snowflake+ML)
- XGBoost as baseline
- Activation via CDP/CRM (Salesforce Marketing Cloud, Braze)
The plan, step by step
- Step 1Define the cancellation label (when has a customer churned) and unify the customer data over 12+ months.Deliverable: Validated label definition + unified customer table (usage, billing, contracts, tickets).
- Step 2Train a baseline model (XGBoost) and backtest it on history.Deliverable: Model with performance metrics on past data.
- Step 3Push the scores into the CRM and launch a pilot campaign on the high scores, with an uncontacted control group.Deliverable: Live retention campaign + measurement design with control.
- Step 4Measure the retention uplift and industrialize the pipeline (retraining, drift detection).Deliverable: Uplift reading vs control + documented automated pipeline.
- Step 5Replicate the template on a second segment or market without starting over.Deliverable: Reusable pipeline template, second model in production.
First step: Scope a pilot on a high-value segment: define the churn label, train a baseline model on the existing warehouse, measure the uplift of a targeted retention campaign on high scores vs a control group.
The same pattern, proven in other industries
Albertsons
AI demand forecasting for fresh and automated shelf replenishment
Carrefour Taiwan
customer lifetime value and churn models to focus marketing effort on the customers worth keeping
Every Man Jack
predictive replenishment timing (predicted next order date)
Sources
- S1 Vodafone's ML platform, built on Vertex AI (AI Booster) Interested party archive pending
- S2 How Vodafone uses CI/CD to speed up ML pipelines (customer churn prediction) Interested party archive pending
- S3 AI Booster: How Vodafone is Supercharging AI & ML at Scale - Datatonic Interested party archive pending
- S4 Vodafone and Google Deepen Strategic Partnership with Ten Year, Billion+ Dollar Deal Primary 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.