thinkhard.SI

Prompt repository Product

Prioritize a feature backlog with a scoring method

You need a ranked backlog with the reasoning visible to stakeholders.

Registry no.
PRD-008
Collection
Product
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
6
Added

ProductAny chat

The prompt

Score 10 of 10, Excellent

Use it when You need a ranked backlog with the reasoning visible to stakeholders.

  • {{product}}
  • {{goals}}
  • {{constraints}}
  • {{stakeholders}}
  • {{method}}
  • {{backlog}}
Full prompt and why it works
Prompt
Background:
Product: {{product}}
Our goals for this period, most important first: {{goals}}
Team capacity this period, and commitments or deadlines already fixed: {{constraints}}
Who will read this ranking: {{stakeholders}}
Scoring method to use: {{method}}

The backlog (one item per line: what it is, who asked and how many, any estimate of effort or reach we have):
"""
{{backlog}}
"""

Task: Score and rank the backlog above, and show the reasoning so that stakeholders can challenge the inputs rather than argue with the result.

Steps:
1. Clean the list: merge duplicates, and restate any item written as a solution as the problem it solves.
2. Define the scales before scoring. For RICE, use Reach (people affected per period, from our data), Impact (3 massive, 2 high, 1 medium, 0.5 low, 0.25 minimal), Confidence (100, 80 or 50 percent) and Effort (person-weeks), with Score = Reach x Impact x Confidence / Effort. For another method, define its scales the same way. If I wrote "suggest one", pick the simplest method our data supports and say why.
3. Score every item, showing each input.
4. Adjust for what scores miss: commitments already made, dependencies, and the goals above. List each item you moved and why.
5. Group the result into Now, Next and Later, with Now fitting within our capacity.

Rules:
- Use only numbers from the backlog. Where an input is missing, give a rough estimate in square brackets, mark it as an assumption and use the lowest confidence.
- The number of people asking is not the same as impact: say when a request comes mostly from one loud customer or one team.
- This is a proposal for discussion. The product owner makes the final call.

Format: a table with the columns Rank, Item, Problem it solves, one column per input, Score, Notes. Then the adjustments, the Now, Next and Later groups, "Not doing this period" with a one-line reason for each, and a summary for stakeholders of under 150 words.

A good result lets any stakeholder point to the exact input they disagree with.

Example of a Notes entry, for form only: "Effort [3 weeks] is an assumption; confirm with engineering."

If anything is missing that you need to score an item at all, list it under "Needs information" with the question that would unblock it. Do not guess.

Variables

{{product}}
The product or area the backlog is for.
{{goals}}
This period's product goals, most important first, for example "reduce churn in the first 30 days".
{{constraints}}
Team capacity, deadlines, dependencies and anything already promised to customers.
{{stakeholders}}
Who will read the ranking, for example "sales, support and the leadership team".
{{method}}
The scoring method: RICE, ICE, value versus effort, or "suggest one".
{{backlog}}
The items, one per line, with whatever you know: description, who asked, how many, effort or reach estimates.

Why it works

  • What good looks like. Every input is shown, so stakeholders argue about an item's reach or effort rather than about a score they cannot see inside.
  • What to avoid. Missing inputs are marked as assumptions at the lowest confidence, and request counts are not mistaken for impact, the two ways scoring most often misleads.
  • Who it's for. It is framed as a proposal for discussion, with the product owner making the final call, which is how a ranked backlog earns trust outside the product team.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.