Define a North Star and Success Metrics for a New Feature with Guardrails

Get a North Star metric, up to three primary metrics each paired with a guardrail, exact formulas from your existing data, targets, and ship or roll-back rules.

You getNorth Star, table and leading indicators

Prompt

Prompt
You are a senior product manager who sets success metrics for new features at software companies.

Your job: define a North Star metric and a small set of success metrics for one feature, with each metric paired with a guardrail.

Before you answer, ask me these 4 questions in one message, then wait for my reply:
1. What is the feature, and who is it for?
2. What user or business problem should it solve?
3. What data do we already track (events, tables, tools such as Amplitude, Mixpanel or SQL)?
4. When will we judge success (for example 30, 60 or 90 days after launch)?

Rules:
- The North Star must measure value users get, not raw activity such as page views or signups.
- Give no more than 3 primary metrics. Each one gets a guardrail that would reveal harm (churn, support tickets, latency, cannibalization of another feature).
- For every metric, name the exact event or field from my data and the formula. If the data does not exist, say so and give the cheapest way to start tracking it.
- Every target is a number with a time window and says where the baseline comes from.
- Flag any metric that could be gamed and explain how.

Output format:
1. North Star: one sentence, formula, why it reflects user value
2. Table: Primary metric | Formula | Data source | Target | Guardrail | Guardrail threshold
3. Leading indicators: 2-3 signals to watch in weeks 1-2
4. Decision rules: conditions to ship, iterate or roll back

End with one line listing your assumptions and data gaps.

Example output

At a glanceSaved Replies: primary metrics and 60-day targetsEach primary metric is paired with a guardrail threshold
ItemTarget (60 days)Guardrail
Weekly adoption35% of active agentsCSAT drop no more than 2 points (4.4/5)
Median first response time18 min, down from 26 minReopen rate at or below 9%
Replies per resolved ticket2.4, down from 3.0Escalations no rise above 6%

Example for: "Saved Replies" in Deskline, a fictional help desk tool for small support teams; data in Mixpanel plus a Postgres tickets table; judged at 60 days.

1. North Star

Tickets resolved per agent-hour on accounts using Saved Replies.

  • Formula: count(ticket_resolved) / sum(agent_active_minutes) * 60, filtered to workspaces with at least one saved_reply_inserted event in the period.
  • Why: the feature exists to let agents close more tickets in less time. Resolved tickets per hour captures that value; reply count alone would not.

2. Primary Metrics and Guardrails

Primary metricFormulaData sourceTarget (60 days)GuardrailGuardrail threshold
Weekly adoptionAgents with ≥3 saved_reply_inserted in a week ÷ active agentsMixpanel35% of active agents (baseline 0%, new feature)Customer satisfaction (CSAT) on tickets using a saved replyNo drop of more than 2 points vs. the 30-day pre-launch CSAT of 4.4/5
Median first response timeMedian of first_reply_at - created_atPostgres tickets18 min, down from 26 min (last 30 days)Reopen rateStays at or below 9% (current 8.1%)
Replies per resolved ticketcount(reply_sent) / count(ticket_resolved)Mixpanel + tickets2.4, down from 3.0Escalations to tier 2No rise above 6% of tickets (current 5.2%)

Gaming risk: adoption can be inflated if agents insert a saved reply and then delete it. Count only inserts where the reply is actually sent (saved_reply_inserted followed by reply_sent within 10 minutes).

3. Leading Indicators (Weeks 1-2)

  • Saved replies created per workspace: aim for 5+ in week 1.
  • Share of inserted replies edited before sending: 30-60% is healthy; below 10% suggests copy-paste replies that may hurt CSAT.
  • Search-to-insert rate in the replies picker: below 40% means agents can't find what they saved.

4. Decision Rules

  • Ship to all plans: adoption ≥ 35% and first response time ≤ 20 min, with all guardrails within thresholds.
  • Iterate: adoption 15-35%, or one primary metric misses target while guardrails hold. Review picker search and onboarding.
  • Roll back or gate: CSAT drops more than 2 points or reopen rate exceeds 9% for two straight weeks.

Assumptions and gaps: agent_active_minutes is not tracked today; use session length from the desktop app as a proxy until a heartbeat event ships (about 1 day of engineering).

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • Pairing every primary metric with a guardrail and threshold stops teams from chasing a number while hurting quality or retention.
  • Requiring an exact event or field and a formula turns vague goals into metrics you can query on launch day.
  • Decision rules set ship, iterate and roll-back conditions in advance, so results can't be reinterpreted after the fact.

Use this in ChatGPT, Claude or Gemini before a feature kickoff or launch review. Have the feature description, the problem it solves, and a list of the events or tables you already track.

Common edits

  • Swap the 60-day window for a 2-week experiment and ask for A/B test sample size notes.
  • Ask for SQL or Mixpanel query drafts for each metric using your event names.

Prompt details

By
Prompts-GPT editors
Published

Did this prompt work?

Keep exploring

More in Product Management