A controlled experiment comparing two variants of a message, offer, or experience to determine which performs better against a defined metric. In real-time decisioning, A/B tests are run continuously across customer segments to optimise next best actions.
AI systems that autonomously perceive context, make decisions, and execute actions without requiring step-by-step human instruction. Agentic AI in enterprise settings orchestrates multi-step workflows — detecting a customer signal, deciding the optimal action, selecting the channel, and executing delivery — all within milliseconds.
A defined interface that allows software systems to communicate and share data with each other. In real-time platforms, APIs enable low-latency event ingestion, decision retrieval, and action execution across core banking, CRM, channel, and data systems.
The process of assigning credit for a conversion or outcome to specific marketing touchpoints or interactions. Multi-touch attribution models weight each touchpoint's contribution rather than crediting only the first or last interaction.
The practice of dividing a customer base into distinct groups based on shared characteristics — demographic, behavioural, transactional, or predictive — so that communications and offers can be tailored to each group's specific context and needs.
An immutable, timestamped record of every decision, action, and data access within a system. Audit trails are essential for regulated industries — they enable organisations to demonstrate compliance, reproduce decision logic, and respond to regulatory enquiries with full transparency.
A data processing method where records are collected over a period and processed together as a group, typically on a scheduled basis (nightly, hourly). Contrasted with stream processing, batch systems introduce latency that makes real-time engagement impossible — a signal detected at 11pm may not trigger action until the next morning.
The analysis of actions taken by customers — clicks, transactions, app usage, browsing paths — to understand intent, predict future behaviour, and personalise engagement. Behavioural signals are the foundation of real-time decisioning systems.
A software component that executes predefined business logic — eligibility criteria, suppression rules, regulatory constraints — independently of application code. In real-time platforms, a rules engine enforces compliance guardrails on every AI-generated recommendation before it is delivered.
A packaged software system that creates a persistent, unified customer database accessible to other marketing and operational systems. A CDP assembles data from multiple sources, resolves identity across channels, and makes profiles available for activation — including real-time decisioning engines.
A predictive model that assigns a probability score to each customer indicating their likelihood of lapsing, cancelling, or switching to a competitor within a defined time window. High-accuracy churn models enable pre-emptive retention actions rather than reactive win-back campaigns.
The systems and processes used to collect, store, and enforce customer permissions for data use and communications. In regulated industries, every outbound communication must be checked against live consent records before delivery — a function that must operate at the same speed as the decisioning engine itself.
The practice of offering a customer a complementary or related product to one they already hold. Effective cross-sell relies on real-time context — a salary credit event is a significantly stronger cross-sell moment than a generic monthly campaign, and timing the offer within the same session multiplies conversion rates.
The complete sequence of interactions a customer has with an organisation — from initial awareness through purchase, onboarding, usage, and renewal or churn. Journey mapping identifies the moments where intervention has the highest impact, which real-time systems can then target automatically.
The total net revenue an organisation can expect to earn from a customer over the entire relationship. CLV models inform which customer segments justify higher engagement investment and guide next best action recommendations toward maximising long-term value rather than short-term conversion.
The framework of policies, roles, and processes that manage data availability, usability, integrity, and security across an organisation. Strong data governance is a prerequisite for compliant AI deployment — it ensures that models are trained on appropriate data and that outputs are auditable.
The requirement that data be stored and processed within a specific geographic jurisdiction. Regulated sectors — banking, healthcare, government — frequently mandate in-country residency. Real-time decisioning platforms must support on-premise or regional cloud deployment to satisfy these requirements without sacrificing latency.
A discipline that applies AI, analytics, and contextual data to improve, automate, and scale organisational decision-making. Decision intelligence systems combine predictive models, business rules, and optimisation logic to determine the best action for each individual customer at each moment in time.
The elapsed time between receiving a customer event and returning a decision. Enterprise-grade real-time systems target sub-100ms decision latency — fast enough to inject a personalised offer into an active mobile session before the user navigates away. Higher latency means the customer moment is missed.
A software design pattern in which services communicate by producing and consuming events rather than making direct synchronous calls. EDA enables real-time decisioning by allowing any system event — a transaction, login, claim submission — to immediately trigger a downstream decision and action chain.
A continuous, ordered sequence of timestamped events flowing from source systems (core banking, CRM, app, IoT) into a processing pipeline. Event streams are the input substrate for real-time decisioning — every customer behaviour generates an event that can trigger an immediate action.
AI systems designed to produce outputs that humans can understand and verify. In regulated industries, explainability is a legal and ethical requirement — every automated decision affecting a customer (credit, insurance, benefit eligibility) must be accompanied by a human-readable reason code that can be presented to the customer or regulator.
The process of transforming raw data into structured inputs (features) that improve machine learning model performance. In real-time systems, features must be computed at inference time from live event data — requiring a feature store that serves pre-computed and on-the-fly values with sub-millisecond retrieval.
Data collected directly from customers through an organisation's own channels and touchpoints — transactions, app usage, web behaviour, call centre interactions, and declared preferences. First-party data is the most accurate, compliant, and commercially valuable data source for personalisation.
A method of tracking customer progression through a defined sequence of steps — awareness, consideration, application, onboarding, activation — and identifying where drop-off occurs. Funnel analysis pinpoints the intervention points where real-time re-engagement would recover the most revenue.
The European Union's primary data protection regulation, establishing rights for individuals over their personal data and obligations for organisations that process it. GDPR requires lawful basis for processing, data minimisation, purpose limitation, and the right to explanation for automated decisions — all directly relevant to AI-driven personalisation systems.
A customer classification used in banking and wealth management for individuals with significant investable assets (typically USD 1M+). HNI customers expect bespoke, proactive service — real-time signals from portfolio movements, life events, or market shifts should trigger immediate, contextual engagement from their relationship manager.
The delivery of highly tailored content, offers, and experiences to individual customers based on real-time behavioural data, predictive models, and contextual signals — going beyond segment-level personalisation to treat each customer as a segment of one. Requires real-time data infrastructure and sub-second decision execution.
The process of matching and merging multiple identifiers — device IDs, email addresses, phone numbers, account numbers — to a single persistent customer profile. Accurate identity resolution is the foundation of cross-channel personalisation; without it, the same customer appears as multiple unknown users across systems.
Customer interaction that occurs during an active session — a mobile app visit, online banking session, or call centre interaction. In-session engagement is the highest-value moment for real-time decisioning: the customer is present, attentive, and the decision latency required is measured in milliseconds, not hours.
A data point indicating that a customer is likely to take a specific action — browsing loan calculator pages (purchase intent), reducing account balance (churn signal), checking data usage limits (upgrade intent). Capturing and acting on intent signals in real time is the primary value proposition of real-time decisioning platforms.
The automated coordination of customer touchpoints across channels and time, ensuring that each interaction is contextually relevant and sequenced appropriately. Journey orchestration moves beyond scheduled campaigns to dynamically route customers through paths determined by their real-time behaviour and AI-driven decisions.
A quantifiable metric used to evaluate progress towards a defined business objective. In real-time engagement, common KPIs include next best action conversion rate, churn rate reduction, in-session offer acceptance, response latency, and decision auditability compliance rate.
In insurance, the probability that a customer will allow their policy to expire without renewal. Lapse risk models score policyholders based on behavioural signals — missed premium payments, reduced engagement, life stage changes — enabling proactive retention outreach before the policy lapses.
The time delay between a stimulus and a response in a system. In customer engagement, latency has two critical dimensions: decisioning latency (time from event to decision, ideally <100ms) and channel latency (time from decision to message delivery). Both must be minimised to capture real-time customer moments.
A deep learning model trained on large corpora of text, capable of generating, summarising, and transforming natural language. In enterprise decisioning, LLMs are used for generating personalised communication content, summarising customer context for agents, and translating decision outputs into customer-facing explanations.
A subset of artificial intelligence in which models learn patterns from data to make predictions or decisions, improving their accuracy over time without being explicitly reprogrammed. ML models in real-time decisioning platforms are continuously retrained on outcome data — each customer interaction becomes a training signal.
The category of software tools and platforms used to plan, execute, measure, and optimise marketing activities. The MarTech stack typically spans data management (CDP, DMP), decisioning, channel execution (email, SMS, push, in-app), analytics, and compliance — all of which must work in concert to deliver real-time personalisation.
The processes, controls, and documentation standards for managing the full lifecycle of AI/ML models — development, validation, deployment, monitoring, and retirement. In regulated industries, model governance frameworks ensure that models are fit for purpose, free from discriminatory bias, and subject to ongoing performance oversight.
An AI-driven decisioning framework that determines the single optimal action to take for a specific customer at a specific moment — whether that is a retention offer, a product recommendation, a service alert, or no action at all. NBA considers the customer's context, predicted intent, business objectives, and compliance constraints simultaneously.
A variant of Next Best Action focused specifically on identifying the product, service, or promotion most likely to be accepted by a given customer at a given moment. NBO models balance commercial objectives (revenue, margin) with customer relevance (propensity to convert, satisfaction impact).
A customer experience strategy that delivers seamless, context-aware interaction across all channels — mobile app, web, SMS, email, branch, call centre, agent — with full continuity of context. An omnichannel decisioning system suppresses duplicate messages, respects channel preference, and updates the customer's state in real time across all touchpoints.
The automated coordination of multiple systems, processes, or actions to achieve a defined outcome. In real-time decisioning, orchestration refers to the end-to-end coordination of: event ingestion → contextualisation → AI decision → compliance check → channel selection → message delivery → outcome tracking.
The tailoring of customer experiences, messages, and offers to the individual based on their data, behaviour, and preferences. Personalisation exists on a spectrum from segment-level (same message for a cohort) to hyper-personalised (unique message for each individual, in real time, based on their current session context).
The use of statistical models and machine learning to forecast future customer behaviours or outcomes from historical data. Predictive analytics informs next best action recommendations, churn scores, propensity models, and lifetime value calculations — the inputs that drive real-time decisioning.
A machine learning model that predicts the probability of a specific customer behaviour — product purchase, churn, claim submission, upgrade — given their current profile and context. Propensity scores are the primary inputs to next best action and offer ranking systems.
A data structure that holds events or messages in order until they are processed. In event-driven architectures, message queues (such as Apache Kafka topics) act as the backbone of real-time pipelines — decoupling producers of events from consumers and ensuring no event is lost during high-volume processing spikes.
The automated process of detecting a customer event, evaluating it against predictive models and business rules, and delivering a personalised action — all within milliseconds of the triggering event. Real-time decisioning collapses the gap between signal and response from days to under a second, converting customer moments into revenue.
The percentage of customers who remain active within a defined period. Improving retention rate by even a small percentage has a disproportionate impact on revenue — retaining an existing customer is 5–7× cheaper than acquiring a new one. Real-time churn intervention significantly outperforms batch retention campaigns.
A customer segmentation framework that scores customers on three dimensions: how recently they transacted (Recency), how often they transact (Frequency), and how much they spend (Monetary value). RFM scoring identifies high-value customers for prioritised engagement and lapsing customers for pre-emptive retention.
The four-stage decisioning loop that describes how Appice processes every customer signal. Sense — ingests and enriches signals from any data source in real time (Appice Integrate). Decide — evaluates rules, segments, and propensity scores to determine the next best action (Appice Intent). Act — fires the action across the right channel at the right moment (Appice Interact). Learn — records outcomes and feeds them back to improve future decisions (Appice Insights). Every real-time customer interaction runs through this loop, typically in under 100ms.
The division of a customer base into groups with shared characteristics for targeted communication and decision-making. Modern segmentation has evolved from static demographic groups to dynamic, AI-driven micro-segments that are updated in real time as customer behaviour changes.
Any data point that indicates a meaningful change in a customer's context, intent, or state — a transaction, a login, an app navigation event, a change in usage pattern. Signals are the inputs to real-time decisioning systems; the ability to detect the right signals and act on them immediately is what separates real-time platforms from batch systems.
Continuous computation on data as it arrives, rather than waiting to accumulate batches. Stream processing frameworks (Apache Kafka, Apache Flink) are the technical foundation of real-time decisioning — they enable sub-second event detection, feature computation, and action triggering at enterprise scale.
In wealth management and financial services, the regulatory requirement to verify that a product or recommendation is appropriate for a specific client given their financial situation, risk tolerance, and objectives. Real-time decisioning systems must embed suitability checks into every recommendation before it is presented to a client or advisor.
A defined event or condition that initiates an automated action or workflow. In real-time engagement, triggers replace scheduled sends — instead of sending a renewal reminder on day 30, the trigger fires the moment a customer's usage pattern indicates lapse risk. Trigger-based communication is significantly more relevant and effective than calendar-based campaigns.
Data sourced from external providers that was collected from individuals who have no direct relationship with the purchasing organisation. Third-party data is declining in availability and reliability due to browser cookie deprecation, increasing privacy regulation, and consumer consent withdrawal — making first-party data enrichment strategies critical.
A predictive model that estimates the incremental impact of a treatment (offer, communication, intervention) on a customer's behaviour — measuring not just whether a customer will respond, but whether the treatment caused the response. Uplift modelling prevents wasted spend on customers who would have converted anyway.
The practice of encouraging an existing customer to upgrade to a higher-value product, larger plan, or premium service. Real-time upsell is triggered by contextual signals — a subscriber approaching their data limit is a significantly higher-intent upsell moment than a generic monthly campaign.
The speed at which data is generated and must be processed. In fraud detection and real-time decisioning, velocity rules detect anomalous transaction frequency (e.g., five card transactions in two minutes across different locations) and trigger immediate intervention within the active session.
An HTTP callback mechanism that sends automated notifications from one system to another when a defined event occurs. Webhooks enable real-time integration between platforms — a core banking system can notify a decisioning platform of a transaction the moment it is processed, triggering an immediate customer action.
The use of software to execute a sequence of tasks — data collection, decision-making, action delivery, outcome logging — without manual intervention. In agentic AI systems, workflow automation operates across systems in real time, coordinating multiple APIs and services to deliver a complete customer response within a single transaction.
See Explainable AI. XAI techniques include SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and attention visualisation — tools that translate complex model outputs into human-readable decision rationales, satisfying regulatory requirements in banking, insurance, and healthcare.
The use of dynamic pricing, capacity management, and personalised offer construction to maximise revenue from a customer base. In insurance, yield optimisation at renewal combines real-time risk signals with competitive pricing models to present each customer with the optimal price point — maximising renewal probability while protecting margin.
Data that a customer intentionally and proactively shares with an organisation — declared preferences, stated communication choices, explicit feedback — in exchange for a more personalised experience. Zero-party data is the highest-trust data type: it carries explicit consent and directly expresses customer intent, making it the most reliable input for personalisation models.