Success metrics (KPIs, guardrails, and exit criteria)¶
This page explains Success metrics (KPIs, guardrails, and exit criteria) and how it fits into the RecSys suite.
Who this is for¶
- Stakeholders defining “what success means” for a RecSys pilot
- Product + analytics teams setting measurement and guardrails
What you will get¶
- A practical KPI + guardrail template you can reuse
- A minimal set of exit criteria that prevents “shipping on broken data”
- Links to the exact evaluation workflow pages
The template (recommended)¶
For each surface you pilot:
- Pick one primary KPI that represents business value.
- Pick 2–4 guardrails that must not regress.
- Define success criteria and rollback criteria up front.
Example KPI choices:
- ecommerce: conversion rate, revenue per session, add-to-cart rate
- content: completion rate, time spent, return rate
Common guardrails:
- latency and error rate
- empty-recs rate
- join-rate / instrumentation integrity (for measurement)
Exit criteria (minimum to call the pilot “credible”)¶
- You can validate logs against schemas (no “unknown fields” surprises).
- Join integrity is sane (broken joins invalidate metrics).
- You can produce at least one report that compares baseline vs candidate.
- You have practiced rollback once (config/rules and/or manifest pointer).
How we measure (suite-level workflow)¶
Start here:
- How to run evaluation and decide ship/hold/rollback: Run eval and ship
- Ship/hold/rollback decision playbook: Decision playbook
Then use the recsys-eval workflow pages:
Read next¶
- Evidence (what outputs look like): Evidence
- ROI and risk model: ROI and risk model
- Evaluation, pricing, and licensing (buyer guide): Buyer guide