The ACCELERAID Platform
Prediction Engine
Propensity, churn, CLV and next-best-action from transaction signals. Explainable ML at enterprise scale — aligned with MaRisk model-documentation requirements.
How it works
From data to score.
Five steps, in real time or in batch.
01
Read the data
Transaction history, spending patterns and frequency and recency metrics come from the Customer Data Platform, including consent status.
02
Calculate scores
Models calculate scores per customer, for example for activity, churn probability and product affinity.
03
Update
Scores are recalculated with every data refresh: weekly, daily or in real time, depending on the package tier.
04
Explain
Every prediction shows the features that drove it (feature importance).
05
Use
Scores are available for segments, triggers and next best action. They support the decision; your team approves the action.
Key Capabilities
Prediction Engine: what it delivers
Capability 1
Propensity Scoring
Churn, purchase and product affinity outputs, trained on your customer data. The models are recalculated with every data refresh and so adapt to changing customer behaviour.
Purchase propensity per product category
Churn probability with early warning signals
Product affinity scoring across the full portfolio
Capability 2
Next Best Action
Real-time decisioning that recommends the next best product, offer, channel or service action. The recommendation applies per customer, not per segment.
Context-aware recommendations per customer
Optimal channel & timing selection
Integrated with the CLM/CVM orchestration engine
Capability 3
Explainable ML
Model documentation, feature-level explainability and model registry – aligned with MaRisk model-documentation requirements. Every score has a traceable explanation — making model outputs transparent for operational adoption and regulatory review in regulated financial environments.
Feature importance for every prediction
Model registry with full version history
Governance documentation on demand
Models
Pre-built scores and segments — 3, 10 or 20 per package tier
From churn prediction to customer lifetime value. All models are designed for transaction data, and every score comes with an explanation.
Models in detail
What each model calculates.
All models work exclusively as decision support. Campaign activation and business decisions remain under the customer’s control.
Behavioural Scoring Models
Applies a quantile transformation to user-level transaction data to normalise behavioural distributions and derive a standardised behavioural score — ensuring comparability across users with different transaction volumes and spending patterns.
- Data
- Transaction history · aggregated spending patterns · frequency & recency metrics
- Retraining
- Automatic recalculation with every data refresh cycle (typically daily)
Activity Scoring Model
Estimates customer engagement probability based on historical transaction behaviour. Scores are recalculated automatically with each data refresh cycle.
- Data
- Transaction recency · frequency · aggregated spend
- Monitoring
- Regular score distribution & stability checks
Predictive Churn Model
Estimates the probability of future customer inactivity or churn based on historical behavioural and transaction patterns. Identifies early behavioural signals of declining engagement.
- Data
- Recency & frequency · spending trends · engagement indicators · card usage metrics
- Retraining
- Periodic — typically weekly or in sync with the customer’s data refresh
Merchant Category Correlation Model
Identifies behavioural affinities between merchant categories to support targeting and cross-category insights. Enables “customers who spend here also spend there” logic for campaigns.
- Data
- Categorised transaction data (MCC) · frequency & co-occurrence metrics
- Retraining
- Automatic recalculation with updated transaction data
Variety Score
Measures diversity of spending behaviour across categories. Provides a single metric reflecting how broadly a customer engages across merchant categories — used for segmentation and targeting.
- Data
- Aggregated transaction category metrics
- Retraining
- Automatic recalculation with every data refresh
- Monitoring
- Distribution consistency checks
Recommendation / Next-Best-Action
Identifies relevant product or campaign opportunities based on transaction patterns. Outputs are recalculated dynamically — core logic is periodically reviewed and validated.
- Data
- Transaction history · category spend · card usage indicators
- Monitoring
- Ongoing monitoring of campaign response trends
Control
What the models may and may not do.
All ACCELERAID models operate transparently, within client-controlled boundaries and with no autonomous financial decision-making.
Client Data Control
All models operate exclusively on data provided and controlled by the Client. No data is shared across clients. No external model training on client data.
Decision-Support Only
No automated credit approval, pricing decision or financial risk decision is performed by any model. All outputs serve exclusively as decision support — final actions remain under Client control.
Client Regulatory Responsibility
The client remains responsible for regulatory compliance regarding the use of model outputs in their jurisdiction. ACCELERAID provides governance documentation on request.
Performance Monitoring
Each model is regularly monitored for score distribution stability, statistical consistency and predictive accuracy. Drift detection triggers review cycles.
Proprietary AI Framework
Built on a TensorFlow-based framework enabling development of additional client-specific models. Custom models are implemented separately under Professional Services scope.
Model governance contact
Questions about model design, governance documentation or regulatory compliance support:
Use Cases
What companies use the Prediction Engine for
Card acquisition
Identify high-propensity prospects and optimise application journeys. AI-based lead prioritisation and personalised checkout flows for card programmes.
Mortgage advisory
Assess mortgage propensity from transaction patterns and life-event signals. Prioritise the right customers at the right time for advisors. All scores serve marketing, sales and service – no creditworthiness or lending decisions.
Churn prevention
Early-warning signals trigger retention journeys before customers leave — across cards, accounts and insurance products.
Cross-Sell
Identify suitable customers in the right lifecycle phase for additional products. Signal-based decisioning replaces batch segment campaigns.
Upsell
Upgrade timing models for premium cards, higher account tiers and insurance products — based on spending patterns and engagement signals.
Segmentation
Dynamic, behaviour-based segmentation that updates in real time — replacing static demographic groups with actionable, data-driven clusters.
Read on
