The prompt
Score 10 of 10, Excellent
Use it when You need to know why you win and lose from your deal notes and lost-reason fields.
- {{deal_export}}
- {{period}}
- {{business_context}}
- {{main_question}}
- {{audience}}
Full prompt and why it works
System prompt
You are a sales operations analyst. You count honestly, read the notes behind every dropdown field, and say how confident each finding is.
User message
<deals> {{deal_export}} </deals> <context> Period covered: {{period}} What we sell and to whom: {{business_context}} The question we most want answered: {{main_question}} </context> Analyze the won and lost deals above to find out why we win and why we lose. The result goes to {{audience}}, who will use it to decide what to change in how we sell, so it needs counts they can check and conclusions that say how sure they are. Produce four parts: 1. Overview: the number of deals won and lost, and the win rate overall and for each segment, deal size band, lead source and competitor that the data includes. 2. Why we lose: a table with the columns Reason, Deals, Example deal IDs (up to 3), and Evidence. Read the notes, not only the lost-reason field: a rep who picks "price" from a dropdown often means something else, such as no clear value or no budget owner, and the notes usually show it. Where the notes contradict the field, go with the notes and say so. 3. Why we win: the same table for won deals. 4. Patterns worth acting on: up to five, each with the evidence, your confidence (strong, moderate or weak), and one change to try. Count each deal once per table, under its main reason, and report numbers rather than words like "most". Call any group of fewer than 10 deals too small to support a conclusion, because a pattern from four deals is usually noise. Describe patterns as associations, not causes: the data can show that deals with several contacts close more often, but not that adding contacts causes the win. Leave out contact names and other personal details. Make sure every count can be checked against the export, so no one has to take a number on trust. <example> A pattern written in the form wanted, with the numbers left as slots: "Deals from partner referrals were won [n] of [N] times ([x]%), against [n] of [N] ([y]%) for outbound. Moderate confidence: the partner deals were also smaller. Try: ask the two most active partners for introductions in the mid-size segment." </example> If fields are missing or inconsistent, for example outcomes not recorded or reasons left blank, report how many deals are affected and leave them out of the counts instead of guessing. If the export is too long to count reliably in one pass, say so and mark counts as approximate with "~".
Variables
- {{deal_export}}
- One row per closed deal with ID, outcome, amount, segment, lead source, competitor, lost reason and notes, exported from your CRM. Remove contact details first.
- {{period}}
- The dates the deals closed in.
- {{business_context}}
- What you sell and to whom, in a sentence or two.
- {{main_question}}
- The one thing you most want to know, for example "why we lose mid-size deals", or "none".
- {{audience}}
- Who reads it, for example "the sales leadership team".
Why it works
- Background facts. It reads the deal notes as well as the lost-reason field, because a dropdown that says "price" often hides a different reason.
- What good looks like. Groups under 10 deals are called too small and patterns are framed as associations, so the team does not rebuild its process around noise.
- If unsure. Deals with missing fields are counted and set aside rather than guessed, and long exports get approximate counts marked as such.