Book a Demo
Platform · Decide

Decide every intent.
In real time.

Real-time decisioning on every signal — propensity, next-best-action, churn, fraud and eligibility — fired on every card swipe, churn flag or policy-lapse warning. Audit-logged and explainable, inside your perimeter.

Talk to Solutions Explore the model catalogue →
Decisioning Console
LIVE
Models in production
Propensity
XGBoost
NBA / NBO
Bandit
Churn
Survival
Fraud
Iso·Forest
LTV
Pareto/NBD
Eligibility
Rules+ML
Recommend
Two-tower
Risk Score
GBM
Decisioning trace · live
PROPENSITY
Term deposit · score 0.87
12ms
NBA
Offer A > Offer B (Δ +18%)
9ms
CHURN
Risk 0.72 · 30-day window
7ms
FRAUD
Anomaly 0.94 · block
5ms
GUARDRAIL
Consent + region check passed
2ms
8 models live 3.2M decisions / hr P95 14ms
What is Decide

Decisioning that fires
the moment a signal arrives.

Decide is the decisioning layer of the Appice loop. Every signal from Sense is scored, ranked and gated — through models, rules and regulatory guardrails — before anything reaches a customer. Decisions land in milliseconds, with reason codes attached.

Sub-15ms decisions

Models and rules co-execute in-stream — no queue, no batch scoring window, no waiting for a nightly run.

Explainable by design

Every decision carries reason codes, feature contributions and a full audit trail — ready for regulators and risk committees.

Guardrails-first

Consent, residency, eligibility and frequency limits are checked before any decision is acted on.

Decision Model Catalogue

Every AI model Decide runs with

Out-of-the-box model families — pre-trained on industry-specific signals and retrainable on your own data. Bring your own models too: BYO MLflow, SageMaker, Vertex AI or custom binaries.

Propensity Models
Product propensity (cross-sell) Upgrade / upsell scoring Conversion likelihood Channel response propensity Offer acceptance prediction
Next Best Action
NBA ranking engine Multi-armed bandit experiments Reinforcement-learning agents Offer eligibility scoring Channel arbitration
Churn & Retention
Churn risk (30/60/90-day) Account-dormancy prediction Wallet-share decay Renewal / win-back likelihood Subscription decay
Fraud & Risk
Real-time anomaly detection Transaction-pattern scoring Account-takeover signals Velocity / device-graph rules AML alerting hooks
Customer Lifetime Value
Predicted CLV Revenue-trajectory modelling Profitability tiering Cost-to-serve prediction Margin-aware ranking
Segmentation & Scoring
Behavioural segmentation RFM / engagement scoring Look-alike audiences Micro-segment generation Dynamic cohort assignment
Eligibility & Rules
Visual rules engine Regulatory guardrails Consent / opt-in checks Frequency & fatigue caps Region / residency rules
Explainable AI
Per-decision reason codes Feature-contribution traces Model-version pinning Bias & drift monitoring Full audit log export
Bring Your Own Models
MLflow registry AWS SageMaker endpoints GCP Vertex AI Azure ML Custom Python / ONNX binaries
Models architecture

8 model families. 50+ pre-built. Every signal scored.

Predictive or generative, classical ML or LLM — Appice scores them all on the same decision, in milliseconds, every score logged with reason codes. Bring your own models too.

50+
Pre-built models
<10ms
Scoring latency
8
Model families
See the full model architecture →
8 model families
Propensity Scoring
Churn & Retention
Fraud & Anomaly
Credit & Eligibility
Next Best Action
Generative AI
Conversational Agents
Custom & BYOM
How Decide fits in the loop

Sense → Decide → Act → Learn.

Decide sits between Sense and Act — every signal becomes a scored, audited decision before it ever turns into a message, action or block.

01

Signal arrives

A typed event from Sense — transaction, behaviour or anomaly.

02

Models score

Propensity, churn, fraud and CLV models execute in parallel — milliseconds end-to-end.

03

Rules & guardrails

Consent, eligibility, frequency and regional checks gate the decision.

04

Hand-off to Act

Act executes — message, offer, block, escalate — with reason codes attached.

Ready to decide

See Decide working.

Send us a sample event flow — we will show you the models that fire, the reason codes they emit, and the actions they would trigger.

Book a Demo