Design Churn Prediction - System Design | WinJob

Churn prediction for B2B subscription revenue

Design a B2B churn prediction platform for Salesforce, HubSpot, and Amplitude-class retention teams. This section covers churn prediction for b2b subscription revenue during the understanding phase.

Operational invariants

Every published score binds model_version, feature_cutoff_ts, and score_snapshot_id so CRM disputes replay the exact inputs. Batch publication never flips CRM aliases until row-count and PSI gates pass.

Failure modes to mention

Failure guard 1.1: Label leakage when including post-cancel support tickets in training features
Failure guard 1.2: Scoring enterprise parents without rolling up child workspace usage
Failure guard 1.3: Treating trial expirations as paid churn in labels

Open with revenue-at-risk math: 200K accounts × ARR × churn rate beats naming XGBoost first.

Java

javaOne Dark Pro

1 public final class ChurnCtx1 {
2   private final String accountId;
3   private final double churnProbability;
4   private final String modelVersion;
5  
6   public ChurnCtx1(String accountId, double churnProbability, String modelVersion) {
7     this.accountId = accountId;
8     this.churnProbability = churnProbability;
9     this.modelVersion = modelVersion;
10   }
11  
12   public boolean requiresCsmReview(double threshold) {
13     return churnProbability >= threshold;
14   }
15 }

Python

pythonOne Dark Pro

1 from dataclasses import dataclass
2  
3 @dataclass(frozen=True)
4 class ChurnCtx1:
5     account_id: str
6     churn_probability: float
7     model_version: str
8  
9     def requires_csm_review(self, threshold: float) -> bool:
10         return self.churn_probability >= threshold

TypeScript

typescriptOne Dark Pro

1 export interface ChurnCtx1 {
2   accountId: string;
3   churnProbability: number;
4   modelVersion: string;
5 }
6 export function requiresCsmReview(s: ChurnCtx1, threshold: number): boolean {
7   return s.churnProbability >= threshold;
8 }

Why interviewers care

Churn Prediction interviews reward crisp scope, explicit trade-offs, and failure stories—not generic microservice diagrams.

Interview checkpoint

Name one failure story for Churn prediction for B2B subscription revenue that proves you understand real outages, not happy-path diagrams.

Key Highlights

•200K accounts
•CSM Console → Churn API → Batch Scorer
•Enterprise SaaS churn is a **retention operations** problem: models rank account

1	public final class ChurnCtx1 {
2	private final String accountId;
3	private final double churnProbability;
4	private final String modelVersion;
5
6	public ChurnCtx1(String accountId, double churnProbability, String modelVersion) {
7	this.accountId = accountId;
8	this.churnProbability = churnProbability;
9	this.modelVersion = modelVersion;
10	}
11
12	public boolean requiresCsmReview(double threshold) {
13	return churnProbability >= threshold;
14	}
15	}

1	from dataclasses import dataclass
2
3	@dataclass(frozen=True)
4	class ChurnCtx1:
5	account_id: str
6	churn_probability: float
7	model_version: str
8
9	def requires_csm_review(self, threshold: float) -> bool:
10	return self.churn_probability >= threshold

1	export interface ChurnCtx1 {
2	accountId: string;
3	churnProbability: number;
4	modelVersion: string;
5	}
6	export function requiresCsmReview(s: ChurnCtx1, threshold: number): boolean {
7	return s.churnProbability >= threshold;
8	}