Keep-customer swarm · Subscription-Lifecycle Agent · Churn-prediction-per-subscriber skill · Build pillar · Published July 11, 2026

How to build a per-location signal-aware churn model end-to-end

This guide explains how to architect the churn-prediction-per-subscriber skill on the subscription-lifecycle agent end-to-end at multi-location multi-unit subscription franchise scale: per-portfolio per-location per-subscriber per-canonical-per-subscriber-per-location-signal-ingestion + per-per-subscriber-feature-engineering-spec + per-per-subscriber-multi-model-ensemble-spec + per-per-subscriber-propensity-calibration-spec + per-per-subscriber-explainability-spec + per-per-subscriber-save-flow-handoff + per-per-subscriber-per-location-FBC-feedback-loop + per-per-subscriber-per-location-audit-trail + per-portfolio audit-trail.

Engage the subscription-lifecycle agent See the 32-agent swarm

What you will build

Per-portfolio per-canonical-per-subscriber-per-location-signal-ingestion across per-visit-frequency + per-visit-recency + per-visit-monetary-RFM + per-cancellation-attempt + per-customer-service-call + per-customer-service-chat + per-staff-turnover-at-location + per-location-NPS-shift + per-location-review-velocity + per-payment-failure + per-subscription-pause + per-subscription-upgrade-downgrade + per-app-engagement-event + per-email-engagement-event + per-loyalty-tier-shift + per-per-location-competitor-opening + per-per-location-economic-context + per-foot-traffic-integration-handoff + per-receipt-stream-handoff + per-call-stream-handoff + per-chat-stream-handoff.
Per-canonical-per-subscriber-feature-engineering-spec — per-RFM-feature + per-rolling-window (7-day + 30-day + 90-day + 365-day) + per-per-location-relative + per-cross-location-relative + per-velocity + per-acceleration + per-seasonal + per-categorical-encoding + per-Feast + per-Tecton + per-Hopsworks + per-Vertex-AI-Feature-Store + per-Databricks-Feature-Store + per-feature-drift-monitoring.
Per-canonical-per-subscriber-multi-model-ensemble-spec + per-propensity-calibration-spec — per-XGBoost + per-LightGBM + per-CatBoost + per-random-forest + per-logistic-regression-baseline + per-Cox-proportional-hazards + per-DeepSurv + per-Transformer + per-LSTM + per-multi-LLM-naturalization + per-model-stacking-blending + per-cross-validation + per-hyperparameter-tuning + per-A/B-test + per-MLflow + per-Weights-Biases + per-Comet-ML + per-Platt-scaling + per-isotonic-regression + per-temperature-scaling + per-Bayesian-binning + per-expected-calibration-error + per-reliability-diagram + per-Brier-score + per-recalibration-FBC-pattern-learning.
Per-canonical-per-subscriber-explainability-spec + per-save-flow-handoff — per-SHAP + per-LIME + per-feature-importance + per-counterfactual-explanation + per-attribution-graph + per-causal-DAG + per-multi-LLM-narrative + per-chain-of-thought + per-propensity-tier (very-high-imminent + high-this-week + medium-this-month + low-watch-only + no-risk) + per-handoff-to-save-flow-propensity-scoring + per-handoff-to-per-tier-loyalty-journey-content + per-handoff-to-per-member-next-best-action + per-handoff-to-tier-transition-timing + per-handoff-to-subscriber-lifecycle-cadence + per-handoff-to-multi-stream-severity-routing.
Per-canonical-per-subscriber-per-location-FBC-feedback-loop + per-audit-trail — per-FBC-per-subscriber-per-location-feedback + per-pattern-learning + per-false-positive-pattern-learning + per-false-negative-pattern-learning + per-propensity-recalibration + per-feature-importance-recalibration + per-save-flow-effectiveness-tracking + per-cohort-drift-monitoring + per-per-subscriber-per-location-canonical-audit-record + per-CCPA-CPRA-DSAR + per-GDPR-DPIA + per-CASL-CRTC + per-PIPEDA-OPC + per-EU-AI-Act-Article-13-transparency-disclosure + per-EU-AI-Act-Article-14-human-oversight + per-NIST-AI-RMF-attestation + per-immutable-WORM-storage.

Why per-vendor-Klaviyo-Smart-Sending-account-level-churn breaks at multi-location multi-unit subscription franchise scale

Per-vendor-Klaviyo-Smart-Sending-canonical-account-level-churn ships per-account per-subscriber per-RFM-cohort primitive. Per-vendor-Recharge + Bold-Subscriptions + Skio + Stay-AI + Loop-Subscriptions + Ordergroove + Smartrr + Awtomic + Subbly + Recurly + Chargebee + Zuora + Stripe-Billing-canonical-account-level-churn ship per-vendor per-native account-level churn primitives.

At 1-account-1-location-1-subscription scale per-account per-subscriber per-RFM-cohort primitive is enough. At multi-location multi-unit subscription franchise scale per-per-subscriber-per-location-signal-ingestion + per-per-subscriber-feature-engineering-spec + per-per-subscriber-multi-model-ensemble-spec + per-per-subscriber-propensity-calibration-spec + per-per-subscriber-explainability-spec + per-per-subscriber-save-flow-handoff + per-per-subscriber-per-location-FBC-feedback-loop + per-per-subscriber-per-location-audit-trail.

Per-account-subscriber-RFM-cohort + per-per-subscriber-per-location-signal-ingestion-blind + per-feature-engineering-blind + per-multi-model-ensemble-blind + per-propensity-calibration-blind + per-explainability-blind + per-save-flow-handoff-blind + per-FBC-feedback-loop-blind + per-multi-format-audit-trail-blind + per-EU-AI-Act-Article-13-blind + per-EU-AI-Act-Article-14-blind + per-NIST-AI-RMF-blind.

The operator-side architecture above per-vendor-churn-prediction primitive is canonical-per-subscriber-per-location-signal-ingestion + per-per-subscriber-feature-engineering-spec + per-per-subscriber-multi-model-ensemble-spec + per-per-subscriber-propensity-calibration-spec + per-per-subscriber-explainability-spec + per-per-subscriber-save-flow-handoff + per-per-subscriber-per-location-FBC-feedback-loop + per-per-subscriber-per-location-audit-trail + per-portfolio-audit-trail.

What is in market today

Per-platform per-subscription-management-vendor

Recharge, Bold Subscriptions, Skio, Stay AI, Loop Subscriptions, Ordergroove, Smartrr, Awtomic, Subbly, Recurly, Chargebee, Zuora, Stripe Billing, Maxio (formerly Chargify), Paddle. Per-account per-subscriber per-subscription primitive. Per-canonical-per-subscriber-per-location-signal-ingestion-canonical-per-subscriber-feature-engineering-canonical-per-subscriber-multi-model-ensemble-canonical-per-subscriber-propensity-calibration-canonical-per-subscriber-explainability-canonical-per-subscriber-save-flow-handoff-canonical-per-subscriber-FBC-feedback-canonical-per-subscriber-audit-trail is not the primitive.

Per-platform per-MLOps-feature-store-vendor

Feast, Tecton, Hopsworks, Vertex AI Feature Store, Databricks Feature Store, AWS SageMaker Feature Store, Azure ML Feature Store, Featureform. Per-account per-feature per-entity primitive. Per-canonical-per-subscriber-RFM-feature-canonical-per-subscriber-rolling-window-canonical-per-subscriber-per-location-relative-canonical-per-subscriber-cross-location-relative-canonical-per-subscriber-velocity-canonical-per-subscriber-acceleration-canonical-per-subscriber-seasonal-canonical-per-subscriber-feature-drift-monitoring is not the primitive.

Per-platform per-AutoML-model-platform-vendor

DataRobot, H2O.ai, Dataiku, Google Vertex AI AutoML, Azure ML AutoML, AWS SageMaker Autopilot, MLflow, Weights & Biases, Comet ML, Neptune.ai, ClearML. Per-account per-model per-experiment primitive. Per-canonical-per-subscriber-multi-model-ensemble-canonical-per-XGBoost-LightGBM-CatBoost-random-forest-logistic-regression-Cox-DeepSurv-Transformer-LSTM-canonical-per-model-stacking-blending-canonical-per-model-A/B-test-canonical-per-model-MLflow-Weights-Biases-CometML is not the primitive.

Per-platform per-explainability-AI-vendor

SHAP, LIME, Captum, InterpretML, IBM AI Explainability 360, Fiddler AI, Arthur AI, Truera, Robust Intelligence, What-If Tool (Google), Aequitas, Saidot, Credo AI, Holistic AI. Per-account per-model per-explanation primitive. Per-canonical-per-subscriber-SHAP-canonical-per-subscriber-LIME-canonical-per-subscriber-feature-importance-canonical-per-subscriber-counterfactual-canonical-per-subscriber-attribution-graph-canonical-per-subscriber-causal-DAG-canonical-per-subscriber-multi-LLM-narrative-canonical-per-subscriber-chain-of-thought is not the primitive.

How the architecture is built

Per-portfolio per-canonical-per-subscriber-per-location-signal-ingestion-substrate. Per-17-signal-classes + per-foot-traffic-integration-handoff + per-receipt-stream-handoff + per-call-stream-handoff + per-chat-stream-handoff canonical-signal-ingestion.
Per-portfolio per-canonical-per-subscriber-feature-engineering-spec. Per-RFM + per-rolling-window + per-per-location-relative + per-cross-location-relative + per-velocity + per-acceleration + per-seasonal + per-categorical-encoding + per-feature-store + per-feature-drift-monitoring canonical-feature-engineering.
Per-portfolio per-canonical-per-subscriber-multi-model-ensemble-spec. Per-XGBoost + per-LightGBM + per-CatBoost + per-random-forest + per-logistic-regression-baseline + per-Cox-proportional-hazards + per-DeepSurv + per-Transformer + per-LSTM + per-multi-LLM-naturalization + per-model-stacking + per-cross-validation + per-hyperparameter-tuning + per-A/B-test + per-MLflow-Weights-Biases-CometML canonical-ensemble.
Per-portfolio per-canonical-per-subscriber-propensity-calibration-spec. Per-Platt-scaling + per-isotonic-regression + per-temperature-scaling + per-Bayesian-binning + per-expected-calibration-error + per-reliability-diagram + per-Brier-score + per-recalibration-FBC-pattern-learning canonical-calibration.
Per-portfolio per-canonical-per-subscriber-explainability-spec. Per-SHAP + per-LIME + per-feature-importance + per-counterfactual-explanation + per-attribution-graph + per-causal-DAG + per-multi-LLM-narrative + per-chain-of-thought canonical-explainability.
Per-portfolio per-canonical-per-subscriber-save-flow-handoff. Per-propensity-tier (very-high-imminent + high-this-week + medium-this-month + low-watch-only + no-risk) + per-handoff-to-6-sibling-skills canonical-handoff.
Per-portfolio per-canonical-per-subscriber-per-location-FBC-feedback-loop. Per-FBC-per-subscriber-per-location + per-pattern-learning + per-false-positive-pattern-learning + per-false-negative-pattern-learning + per-propensity-recalibration + per-feature-importance-recalibration + per-save-flow-effectiveness-tracking + per-cohort-drift-monitoring canonical-FBC.
Per-portfolio per-canonical-per-subscriber-per-location-audit-trail + per-portfolio-audit-trail. Per-subscriber-per-location-canonical-audit-record + per-CCPA-CPRA-DSAR + per-GDPR-DPIA + per-CASL-CRTC + per-PIPEDA-OPC + per-EU-AI-Act-Article-13 + per-EU-AI-Act-Article-14 + per-NIST-AI-RMF + per-immutable-WORM canonical-audit.
Per-portfolio per-subscription-lifecycle-agent-canonical-bundle. Per-per-location-churn-prediction + per-subscription-analytics + per-subscriber-lifecycle-cadence + per-per-member-monthly-clv + per-per-member-next-best-action + per-tier-transition-timing + per-save-flow-propensity-scoring + per-lifecycle-flow-architecture + per-predictive-analytics-customer-retention + per-dtc-cancellation-reason-clustering + per-llm-cancellation-reason-clustering canonical-bundle.
Per-portfolio per-canonical-end-to-end-SLA. Per-signal-ingest-to-feature-engineer-to-score-to-calibrate-to-explain-to-handoff-SLA canonical-end-to-end-SLA.
Per-portfolio per-canonical-end-to-end-replay. Per-replay-spec + per-replay-trace + per-replay-decision canonical-replay.
Per-portfolio per-canonical-shadow-deployment. Per-shadow-vs-live-model-comparison + per-shadow-cohort-spec + per-shadow-promotion-rule canonical-shadow.
Per-portfolio per-canonical-EU-AI-Act-NIST-AI-RMF-compliance. Per-Article-13-transparency-disclosure + per-Article-14-human-oversight + per-NIST-AI-RMF-attestation + per-FBC-feedback-as-human-oversight-record canonical-AI-compliance.

Frequently asked questions

What is a per-location signal-aware churn model for multi-unit subscription brands?

A per-location signal-aware churn model runs per-portfolio per-location per-subscriber per-canonical-per-subscriber-per-location-signal-ingestion (per-visit-frequency + per-visit-recency + per-visit-monetary-RFM + per-cancellation-attempt + per-customer-service-call + per-customer-service-chat + per-staff-turnover-at-location + per-location-NPS-shift + per-location-review-velocity + per-payment-failure + per-subscription-pause + per-subscription-upgrade-downgrade + per-app-engagement-event + per-email-engagement-event + per-loyalty-tier-shift + per-per-location-competitor-opening + per-per-location-economic-context per-canonical-signal) + per-canonical-per-subscriber-feature-engineering-spec + per-canonical-per-subscriber-multi-model-ensemble-spec + per-canonical-per-subscriber-propensity-calibration-spec + per-canonical-per-subscriber-explainability-spec + per-canonical-per-subscriber-save-flow-handoff + per-canonical-per-subscriber-per-location-FBC-feedback-loop + per-canonical-per-subscriber-per-location-audit-trail + per-portfolio audit-trail.

Why does per-vendor-Klaviyo-Smart-Sending-canonical-account-level-churn break at multi-location-multi-unit subscription franchise scale?

Per-vendor-Klaviyo-Smart-Sending-canonical-account-level-churn ships per-account per-subscriber per-RFM-cohort primitive. Per-vendor-Recharge-canonical + per-Bold-Subscriptions-canonical + per-Skio-canonical + per-Stay-AI-canonical + per-Loop-Subscriptions-canonical + per-Ordergroove-canonical + per-Smartrr-canonical + per-Awtomic-canonical + per-Subbly-canonical + per-Recurly-canonical + per-Chargebee-canonical + per-Zuora-canonical + per-Stripe-Billing-canonical-account-level-churn ship per-vendor per-native account-level churn primitives. At 1-account-1-location-1-subscription scale per-account per-subscriber per-RFM-cohort primitive is enough. At multi-location multi-unit subscription franchise scale per-canonical-per-subscriber-per-location-signal-ingestion + per-canonical-per-subscriber-feature-engineering-spec + per-canonical-per-subscriber-multi-model-ensemble-spec + per-canonical-per-subscriber-propensity-calibration-spec + per-canonical-per-subscriber-explainability-spec + per-canonical-per-subscriber-save-flow-handoff + per-canonical-per-subscriber-per-location-FBC-feedback-loop + per-canonical-per-subscriber-per-location-audit-trail.

How does per-portfolio per-canonical-per-subscriber-per-location-signal-ingestion + per-feature-engineering work?

Per-portfolio per-canonical-per-subscriber-per-location-signal-ingestion runs per-portfolio per-canonical-per-subscriber-per-location-visit-stream-handoff (sibling at /how-to-build-foot-traffic-integration-for-multi-location-attribution) + per-canonical-per-subscriber-per-location-receipt-stream-handoff + per-canonical-per-subscriber-per-location-call-stream-handoff + per-canonical-per-subscriber-per-location-chat-stream-handoff + per-canonical-per-subscriber-per-location-cancellation-attempt-event + per-canonical-per-subscriber-per-location-payment-failure-event + per-canonical-per-subscriber-per-location-subscription-pause-event + per-canonical-per-subscriber-per-location-subscription-upgrade-downgrade-event + per-canonical-per-subscriber-per-location-staff-turnover-event + per-canonical-per-subscriber-per-location-NPS-shift-event + per-canonical-per-subscriber-per-location-review-velocity-event + per-canonical-per-subscriber-per-location-app-engagement-event + per-canonical-per-subscriber-per-location-email-engagement-event + per-canonical-per-subscriber-per-location-loyalty-tier-shift-event + per-canonical-per-subscriber-per-location-competitor-opening-event + per-canonical-per-subscriber-per-location-economic-context-event. Per-canonical-per-subscriber-feature-engineering-spec runs per-portfolio per-canonical-per-subscriber-RFM-feature-spec + per-canonical-per-subscriber-rolling-window-feature-spec (per-7-day + per-30-day + per-90-day + per-365-day per-canonical-window) + per-canonical-per-subscriber-per-location-relative-feature-spec + per-canonical-per-subscriber-cross-location-relative-feature-spec + per-canonical-per-subscriber-velocity-feature-spec + per-canonical-per-subscriber-acceleration-feature-spec + per-canonical-per-subscriber-seasonal-feature-spec + per-canonical-per-subscriber-categorical-encoding-spec + per-canonical-per-subscriber-feature-store-spec (per-Feast + per-Tecton + per-Hopsworks + per-Vertex-AI-Feature-Store + per-Databricks-Feature-Store per-canonical-feature-store) + per-canonical-per-subscriber-feature-drift-monitoring-spec.

What does per-portfolio per-canonical-per-subscriber-multi-model-ensemble + per-propensity-calibration + per-explainability do?

Per-portfolio per-canonical-per-subscriber-multi-model-ensemble-spec runs per-portfolio per-canonical-per-XGBoost-gradient-boosted-tree-spec + per-canonical-per-LightGBM-spec + per-canonical-per-CatBoost-spec + per-canonical-per-random-forest-spec + per-canonical-per-logistic-regression-baseline-spec + per-canonical-per-survival-analysis-Cox-proportional-hazards-spec + per-canonical-per-survival-analysis-DeepSurv-spec + per-canonical-per-deep-learning-Transformer-spec + per-canonical-per-deep-learning-LSTM-spec + per-canonical-per-multi-LLM-naturalization-spec (per-GPT-4o + per-Claude-Sonnet + per-Gemini-Pro per-canonical-LLM) + per-canonical-per-model-stacking-blending-spec + per-canonical-per-model-cross-validation-spec + per-canonical-per-model-hyperparameter-tuning-spec + per-canonical-per-model-A/B-test-spec + per-canonical-per-model-MLflow-Weights-Biases-CometML-experiment-tracking. Per-canonical-per-subscriber-propensity-calibration-spec runs per-portfolio per-canonical-per-Platt-scaling-spec + per-canonical-per-isotonic-regression-spec + per-canonical-per-temperature-scaling-spec + per-canonical-per-Bayesian-binning-spec + per-canonical-per-propensity-expected-calibration-error-spec + per-canonical-per-propensity-reliability-diagram-spec + per-canonical-per-propensity-Brier-score-spec + per-canonical-per-propensity-recalibration-FBC-pattern-learning. Per-canonical-per-subscriber-explainability-spec runs per-portfolio per-canonical-per-SHAP-Shapley-value-spec + per-canonical-per-LIME-local-interpretable-spec + per-canonical-per-feature-importance-spec + per-canonical-per-counterfactual-explanation-spec + per-canonical-per-attribution-graph-spec + per-canonical-per-causal-DAG-spec + per-canonical-per-multi-LLM-narrative-explanation + per-canonical-per-chain-of-thought-extraction.

What does per-portfolio per-canonical-per-subscriber-save-flow-handoff + per-FBC-feedback do?

Per-portfolio per-canonical-per-subscriber-save-flow-handoff runs per-portfolio per-canonical-per-subscriber-propensity-tier-spec (per-very-high-imminent + per-high-this-week + per-medium-this-month + per-low-watch-only + per-no-risk per-canonical-tier) + per-canonical-per-subscriber-save-flow-routing-spec (per-handoff-to-save-flow-propensity-scoring-skill at /save-flow-propensity-scoring) + per-canonical-per-subscriber-per-location-handoff-to-per-tier-loyalty-journey-content-skill (sibling at /per-tier-loyalty-journey-content) + per-canonical-per-subscriber-per-location-handoff-to-per-member-next-best-action-skill (sibling at /per-member-next-best-action) + per-canonical-per-subscriber-per-location-handoff-to-tier-transition-timing-skill (sibling at /tier-transition-timing) + per-canonical-per-subscriber-per-location-handoff-to-subscriber-lifecycle-cadence-skill (sibling at /subscriber-lifecycle-cadence) + per-canonical-per-subscriber-per-location-handoff-to-multi-stream-severity-routing-skill (sibling build-pillar at /how-to-build-multi-stream-severity-routing-for-anomaly-detection-and-compliance-ops). Per-canonical-per-subscriber-per-location-FBC-feedback-loop runs per-portfolio per-canonical-FBC-per-subscriber-per-location-feedback + per-canonical-FBC-per-subscriber-per-location-pattern-learning + per-canonical-FBC-per-subscriber-per-location-false-positive-pattern-learning + per-canonical-FBC-per-subscriber-per-location-false-negative-pattern-learning + per-canonical-FBC-per-subscriber-per-location-propensity-recalibration + per-canonical-FBC-per-subscriber-per-location-feature-importance-recalibration + per-canonical-FBC-per-subscriber-per-location-save-flow-effectiveness-tracking + per-canonical-FBC-per-subscriber-per-location-cohort-drift-monitoring.

What does per-portfolio per-canonical-per-subscriber-per-location-audit-trail + per-subscription-lifecycle-agent-canonical-bundle do?

Per-portfolio per-canonical-per-subscriber-per-location-audit-trail runs per-portfolio per-canonical-per-subscriber-per-location-canonical-audit-record (per-subscriber-ID + per-location-pointer + per-signal-ingestion-record + per-feature-engineering-record + per-model-ensemble-decision + per-propensity-score + per-propensity-calibration-record + per-SHAP-feature-importance-record + per-LIME-explanation-record + per-counterfactual-explanation-record + per-multi-LLM-narrative + per-save-flow-handoff-decision + per-FBC-feedback-record per-canonical-audit-record) + per-canonical-CCPA-CPRA-DSAR-export + per-canonical-GDPR-DPIA-export + per-canonical-CASL-CRTC-export + per-canonical-PIPEDA-OPC-export + per-canonical-EU-AI-Act-Article-13-transparency-disclosure + per-canonical-EU-AI-Act-Article-14-human-oversight + per-canonical-NIST-AI-RMF-attestation + per-canonical-immutable-WORM-storage. Per-subscription-lifecycle-agent-canonical-bundle integrates the churn-prediction-per-subscriber skill with sibling skills on the same agent: per-canonical-per-location-churn-prediction (parent commercial pillar at /per-location-churn-prediction) + per-canonical-subscription-analytics (sibling commercial pillar at /subscription-analytics) + per-canonical-subscriber-lifecycle-cadence (sibling commercial pillar at /subscriber-lifecycle-cadence) + per-canonical-per-member-monthly-clv (sibling commercial pillar at /per-member-monthly-clv) + per-canonical-per-member-next-best-action (sibling commercial pillar at /per-member-next-best-action) + per-canonical-tier-transition-timing (sibling commercial pillar at /tier-transition-timing) + per-canonical-save-flow-propensity-scoring (sibling commercial pillar at /save-flow-propensity-scoring) + per-canonical-lifecycle-flow-architecture (sibling commercial pillar at /lifecycle-flow-architecture) + per-canonical-predictive-analytics-customer-retention (sibling commercial pillar at /predictive-analytics-customer-retention) + per-canonical-dtc-cancellation-reason-clustering (sibling commercial pillar at /dtc-cancellation-reason-clustering) + per-canonical-llm-cancellation-reason-clustering (sibling build-pillar at /how-to-build-llm-cancellation-reason-clustering).

Engage the subscription-lifecycle agent

Per-portfolio per-location per-subscriber per-canonical-per-subscriber-per-location-signal-ingestion + per-per-subscriber-feature-engineering-spec + per-per-subscriber-multi-model-ensemble-spec + per-per-subscriber-propensity-calibration-spec + per-per-subscriber-explainability-spec + per-per-subscriber-save-flow-handoff + per-per-subscriber-per-location-FBC-feedback-loop + per-per-subscriber-per-location-audit-trail + per-portfolio audit-trail shipped as the orchestration layer above your existing per-subscription-management-vendor + per-MLOps-feature-store-vendor + per-AutoML-model-platform-vendor + per-explainability-AI-vendor primitive.

Book a scope call Browse the 32-agent catalog