Peloton
genAI-personalized training plans and recommendations from usage data
Peloton IQ, launched in October 2025 on Amazon Bedrock (GPT and Llama 4 Scout), generates personalized weekly training plans and recommendations: millions of insights a week for more than 6 million members and nearly 50 million workouts a month.
Key points
- GenAI-personalized weekly training plans and insights (Peloton IQ).
- Runs on Amazon Bedrock, with OpenAI's GPT and Meta's Llama 4 Scout.
- Millions of insights a week for more than 6 million members.
- Evidence B, confirmed status in production since October 2025.
Objective
Reignite member engagement by turning training history into personalized weekly plans and recommendations, to support usage frequency and subscription retention.
The deployment
Peloton IQ is the AI and computer-vision layer launched in October 2025 across all Peloton connected equipment and the mobile app. For each member, it generates an editable weekly training plan, adjusted to goals (strength, weight, longevity), level, and stated preferences. With each workout, the recommendations refine, with performance targets computed from the member's history and third-party wearable data (Garmin Connect, Fitbit, Apple Health). The insights turn activities, on and off the platform, into performance, recovery, and consistency trends. The system runs on Amazon Bedrock, with a dual workflow: cached batch pre-generation for immediate display, and real-time generation triggered by the member's actions. Peloton uses OpenAI's GPT models for complex reasoning and Meta's Llama 4 Scout for real-time insights, with fine-tuning on SageMaker.
Results Proof B
Quantified platform case study (AWS) on the architecture and volume (millions of weekly insights, 6 million members, 50 million monthly workouts), corroborated by Peloton's investor release announcing Peloton IQ. Usage volumes documented, but no isolated retention KPI, hence B.
How it works
Documented architectureThe stack in detail
- plateforme Amazon Bedrock AWS's managed genAI platform on which Peloton IQ runs, with a dual workflow: cached batch pre-generation and real-time generation.
- llm OpenAI GPT OpenAI's GPT models used for complex reasoning (building the training plans).
- llm Meta Llama 4 Scout Meta's model used for real-time insights triggered by the member's actions.
- plateforme Amazon SageMaker AWS service used for fine-tuning the models on Peloton's data.
- outil Integrations wearables (Garmin Connect, Fitbit, Apple Health) Third-party data sources that enrich the member's context beyond the workouts on Peloton equipment.
How it runs, concretely
For ops teams-
1Member context collection data / AI team
Training history, connected-device biometrics, preferences, and third-party wearable data are aggregated.
-
2Batch pre-generation AI (Bedrock, batch)
Insights and performance targets are pre-computed and cached for instant display.
-
3Real-time generation AI (Llama 4 Scout / GPT)
On the member's action, an insight or plan adjustment is generated on the fly by the LLMs.
-
4Delivery and editing member
The weekly plan displays on the screen and the app; the member can edit it.
Training history and third-party wearable data. If the member does not connect their devices or trains little, the context thins out and the insights lose relevance.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- training or usage history per user
- connection to wearables or biometric sensors
- stated preferences and goals
Org prerequisites
- ML team to maintain prompts and fine-tuning
- legal basis for processing health data
- capacity for real-time display on the app and devices
Possible stack
- managed LLM platform (Bedrock, Vertex AI, Azure OpenAI)
- batch + real-time pipeline
- wearable integration
The plan, step by step
- Step 1Structure the training or usage history per user (workouts, goals, stated preferences) in a queryable profile.Deliverable: Unified usage profile per user
- Step 2Connect a managed LLM to a first insight case, for example the weekly summary, in batch pre-generation.Deliverable: Batch insights in internal beta
- Step 3Cache and display the insights in the app, measuring open and engagement on a beta cohort.Deliverable: Feature in user beta with engagement metrics
- Step 4Add real-time generation triggered by the user's action and editable plans.Deliverable: Personalized plans in production on the app
- Step 5Connect third-party wearables or sensors to enrich the context and refine the performance targets.Deliverable: Enriched pipeline + frequency and retention assessment
First step: Structure the training history per user and connect a managed LLM to a first insight case (for example a weekly summary), before adding real-time generation.
The same pattern, proven in other industries
Boots
machine learning scoring of offer-usage propensity, applied to mail and the app
Meta
AI feed and video recommendation (large-scale models / LLMs)
Michaels
genAI personalization of marketing messaging (copy) at scale, tuned to the brand voice
Sources
- S1 Peloton IQ: How Peloton Generates Millions of Personalized Fitness Insights Weekly Using Amazon Bedrock Interested party archive pending
- S2 Peloton Enters New Era with AI-Powered Peloton IQ and New Product Portfolio Primary archive pending
- S3 From Personalized Insights to Actions: Powering Peloton IQ for Cross Training 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.