Churn Definition For Ai Data Agents

Churn has at least six defensible definitions inside one company, and AI agents without a glossary entry for each will pick the most convenient one and produce wrong numbers. The fix is explicit churn definitions per team, each with an owned SQL template.

Ask five people in a company to define churn and you will get five answers. Revenue churn, logo churn, gross churn, net churn, voluntary churn, customer churn. A data agent that conflates them will produce answers that contradict the dashboards and undermine trust. This guide covers how to structure churn definitions so agents get them right. Related: business definitions for AI agents and AI for data infrastructure.

The Six Churn Definitions

• •Revenue churn — lost ARR from existing customers over a period
• •Logo churn — lost customers, regardless of contract size
• •Gross revenue churn — lost ARR before adding upsell
• •Net revenue churn — lost ARR minus upsell from retained customers
• •Voluntary churn — customers who cancelled on their own
• •Involuntary churn — failed payments or forced cancellations

Why Each Definition Matters

Different teams care about different definitions. Finance cares about revenue churn because it feeds forecasts. Sales cares about logo churn to measure account success. Product cares about voluntary churn to measure feature quality. A good data agent serves each team the right definition based on who is asking and what they are asking.

The glossary entry for churn has to encode all six definitions with explicit SQL templates. When a user asks about churn, the agent checks scope (who is asking) and the question (what variant they mean) and picks the matching template. If it is ambiguous, the agent asks.

The Cohort Question

Cohorting is a second axis. Churn can be computed as a period rate (5 percent of customers churned last month) or as a cohort curve (of customers who signed in Q1 2025, X percent have churned as of today). Both are valid. The glossary entry must specify which cohort basis each template uses.

A user who asks what is our churn rate almost always means the period rate. A user who asks about cohort retention means the curve. An agent must distinguish by question phrasing or ask for clarification.

Lookback Windows

Churn also depends on lookback. Annual churn and monthly churn are not just different units — they reflect different behavior. Monthly churn multiplied by 12 does not equal annual churn because customers can churn and return within a year. Each definition must specify its lookback window, and the agent must not convert between them.

Grace Periods and Reactivation

Some companies count a customer as churned only after 30 days past their cancellation date (grace period). Others count on day one. Reactivated customers may or may not reset the churn counter. These are policy choices and the glossary must document them.

Handling Disagreement

When finance and product disagree about churn, the agent should not pick a side. It should surface both numbers with their definitions and let the user pick. Surfacing disagreement is more valuable than hiding it — users learn the definitions matter and start asking more precisely.

Common Mistakes

The biggest mistake is a single churn definition in the glossary. The second is not scoping definitions to teams. The third is not documenting lookback windows. The fourth is silently converting monthly to annual. The fifth is treating churn as a single number instead of a family of related metrics.

Data Workers builds the glossary agent so every churn variant has an owned SQL template, a scope, and a lookback window. The agent picks the right one per question and asks when ambiguous. To see it running against your data, book a demo.

Making Churn Visible

Once the glossary has all the churn variants, the next step is to make them visible in the product. When a user asks about churn, the agent should not just pick one and answer — it should offer the variants explicitly: revenue churn this month was X, logo churn was Y, net revenue churn including upsell was Z. Showing all three with their definitions teaches users the distinction.

Over time users start asking for specific variants by name. That is the goal: an educated user base that knows there is no single churn number and asks the right question. The glossary is not just a backend artifact; it is a teaching tool embedded in the product.

Data Workers surfaces glossary variants automatically in the agent response so users see the options. Within weeks, question quality improves and ambiguous requests drop. The investment pays off in user literacy as much as in accuracy.

Churn Disputes and How to Settle Them

When finance and product disagree about the churn number, the agent can settle the dispute by running both definitions and showing the gap. Finance says revenue churn is 5 percent; product says logo churn is 7 percent. Both are right for their scope, and the gap is explained by differences in contract size between churned customers.

Surfacing the gap turns an argument into a conversation about strategy. Which customers churned — were they high-value or low-value. Why — onboarding, product fit, pricing. The data becomes a starting point for action, not a source of internal conflict. This is what a mature glossary produces over time.

Data Workers makes this pattern the default. When a user asks about churn without a clear scope, the agent runs all relevant definitions and shows them side by side. The user picks the one they care about and the rest become context. Ambiguity becomes a feature, not a bug.

The long-term benefit of structured churn definitions extends beyond agent accuracy. Product teams start aligning on which churn variant drives their OKRs. Finance stops debating numbers in board prep because the glossary is the single source of truth. Customer success teams use the same definitions as renewal forecasts, which reduces surprise at quarter end. The glossary becomes organizational infrastructure, not just agent infrastructure, and the investment pays dividends in every meeting where churn comes up.

Churn is not one definition — it is a family. Put every variant in the glossary with explicit SQL and scope, and your agents stop contradicting the dashboards.