Glossary
The language of ARPU growth.
The vocabulary of customer decisioning and ARPU growth, defined precisely enough to be quoted. Each term states what it means, how it is calculated where that applies, and why it moves revenue per customer.
57 terms
Markin vocabulary
The words we use to describe how an autonomous growth-science team works. Signal in, decision out, evidence attached.
SignalA signal is an observed change in customer behaviour, product state, payment health or market context that carries information about future revenue.Revenue opportunityA revenue opportunity is a named, sized and addressable gap between what a customer segment is worth today and what it could be worth.HypothesisA hypothesis is a falsifiable statement linking an action to a revenue outcome for a defined population: if we do X for segment Y, metric Z moves by roughly N because of mechanism M.Candidate actionA candidate action is one specific intervention eligible for one specific customer at one moment: an offer, a message, a price, a plan change, a support intervention or deliberately doing nothing.ExperimentAn experiment is a controlled release of a candidate action against a randomised holdout, sized in advance so the result can distinguish a real effect from noise.Decision layerA decision layer is the system that sits between the data warehouse and the execution tools and decides, per customer, what deserves to happen next.Growth agentA growth agent is an autonomous worker that runs one stage of the growth-science loop without being prompted: reading signals, writing hypotheses, designing experiments, launching them through connected systems or reading out causal results.Opportunity feedAn opportunity feed is a continuously refreshed, ranked list of revenue opportunities detected in a customer base, each with its population, expected value, supporting evidence and suggested next action.Hypothesis provenanceHypothesis provenance is the complete, inspectable chain behind a decision: which signals raised it, which analysis sized it, who or what wrote the hypothesis, which experiment tested it, and what the causal read-out was.Decision volumeDecision volume is the number of distinct, evidenced customer-level decisions a business makes in a period.Autonomous growth scienceAutonomous growth science is the practice of running the full scientific loop over a customer base without a human in every step: observe, hypothesise, decide, launch, measure causally, and update.
Revenue and ARPU metrics
How revenue per customer is measured, and the ratios executives use to judge whether growth is worth its cost.
ARPUARPU, average revenue per user, is total revenue in a period divided by the average number of active users in that period.ARPAARPA, average revenue per account, is revenue divided by the number of accounts rather than individual users.ARPPUARPPU, average revenue per paying user, divides revenue only by users who paid in the period.Lifetime valueLifetime value is the discounted margin a business expects from a customer over the whole relationship.LTV:CAC ratioThe LTV:CAC ratio divides expected customer lifetime value by fully loaded customer acquisition cost.Payback periodPayback period is the time taken for the gross margin generated by a customer to repay the cost of acquiring them.Net revenue retentionNet revenue retention measures revenue from an existing cohort at the end of a period against its revenue at the start, including upgrades, downgrades and churn, but excluding new customers.Gross revenue retentionGross revenue retention measures how much starting cohort revenue survives a period counting only losses: churn and downgrades, never expansion.Expansion revenueExpansion revenue is additional revenue from customers a business already has: upgrades, add-ons, cross-sell, higher usage or a move to a richer plan.Revenue per available customerRevenue per available customer spreads revenue across everyone reachable, including dormant and non-paying users, rather than only the active base.Contribution margin per userContribution margin per user is revenue per user minus the variable costs of serving that user: delivery, payment fees, support, content or bandwidth, and any incentive granted.Monetization rateMonetization rate is the share of active customers who pay anything in a period.
Retention and churn
Churn is the largest single drag on ARPU in a large B2C base. These are the terms that describe it precisely.
Churn preventionChurn prevention is the practice of identifying customers likely to leave and choosing the intervention, if any, that retains the most revenue net of its cost.Churn rateChurn rate is the share of customers, or of revenue, lost in a period.Voluntary vs involuntary churnVoluntary churn is a customer deciding to leave.Churn predictionChurn prediction estimates the probability that a given customer will stop paying within a defined horizon, using behavioural, transactional and service signals.Survival analysisSurvival analysis models time until an event, handling customers who have not churned yet as censored rather than discarding them.Retention curveA retention curve plots the share of a cohort still active against time since acquisition.Cohort analysisCohort analysis groups customers by a shared starting characteristic, usually acquisition month, and follows each group over time.Win-backWin-back is the practice of returning a lapsed or cancelled customer to paying status.Save offerA save offer is an incentive presented to a customer who is about to leave: a discount, a pause, a plan downgrade or a service remedy.DunningDunning is the sequence of retries and communications that recovers a failed payment: retry timing, card-update prompts, alternative payment methods and grace periods.Time to churnTime to churn is the expected remaining tenure of a customer, derived from a survival model rather than a binary risk score.
Decisioning and next best action
Choosing what to do for each customer, and the models and rules that turn a prediction into an action.
AI agent for marketingAn AI agent for marketing is a system given a commercial objective, tools such as a warehouse, models and execution channels, and the autonomy to decide which actions best serve that objective.Next best actionNext best action is the single intervention that maximises expected value for a specific customer at a specific moment, chosen across every available option including doing nothing.Next best offerNext best offer selects the most valuable commercial proposition for a customer: which product, plan, bundle or price to present.Propensity modelA propensity model estimates the probability that a customer takes an action, such as buying, upgrading or cancelling.Uplift modelAn uplift model estimates the change in outcome caused by treating a customer, rather than the outcome itself.Eligibility rulesEligibility rules define which actions a customer may legally, commercially and contractually receive: consent status, market availability, existing plan, discount history, regulatory constraints.ArbitrationArbitration is the step that resolves competing candidate actions for the same customer into one decision, ranking them by expected value under constraints such as contact limits, budget, fairness and business priority.Contact policyContact policy governs how often, through which channels and with what spacing a customer may be contacted.Recommendation engineA recommendation engine ranks items by predicted affinity for a user, usually inside one surface such as a catalogue, feed or storefront.Real-time decisioningReal-time decisioning evaluates a customer's current context and returns an action within the latency budget of the moment, typically tens of milliseconds inside an app or call.Reinforcement learning for marketingReinforcement learning treats customer decisions as a sequence, optimising cumulative long-term reward rather than the next click.
Experimentation and causality
Proving an action caused the lift. Without these terms, reported growth is correlation with a dashboard.
A/B testAn A/B test randomly assigns customers to two or more variants and compares a pre-declared metric between them.Holdout groupA holdout is a randomly selected set of customers deliberately excluded from an action, kept as the counterfactual.IncrementalityIncrementality is the portion of an outcome that would not have happened without the action.Statistical powerStatistical power is the probability that a test detects an effect of a given size when that effect is real.Sequential testingSequential testing allows results to be monitored continuously and stopped early without inflating false positives, using always-valid confidence intervals instead of a single fixed-horizon check.Multi-armed banditA multi-armed bandit allocates traffic adaptively toward variants that are performing well while continuing to sample the others.Novelty effectThe novelty effect is a temporary response lift caused by a change being new rather than better.Guardrail metricA guardrail metric is a measure an experiment must not damage even if the primary metric improves: unsubscribe rate, complaint volume, margin, support contacts, app uninstalls.
Data and activation stack
The systems that hold the data and deliver the action. Markin sits between them as the decision layer.
Customer data platformA customer data platform collects customer data from multiple sources, resolves it to unified profiles and makes segments available to downstream tools.Reverse ETLReverse ETL syncs modelled data from a warehouse back into operational tools such as engagement platforms, CRMs and ad networks.Customer 360Customer 360 is a consolidated view of everything known about a customer across systems: profile, transactions, interactions, service history and consent.Feature storeA feature store computes, versions and serves the model inputs used in training and in production, guaranteeing that both see the same definition of a feature.