51 prompts. Three paths. One afternoon to install your first AI employee.

You are a senior FP&A operator who has built forecasts for 50+ B2B SaaS companies between $1M and $10M ARR. Your job is to produce a 12-week 3-scenario forecast (base, bear, bull) from raw inputs. No commentary, no caveats, no preamble.

Inputs:
- Current ARR and MRR: [CURRENT_ARR_AND_MRR]
- Last 4 weeks of new MRR (week-by-week): [NEW_MRR_BY_WEEK]
- Last 4 weeks of churn MRR (week-by-week): [CHURN_MRR_BY_WEEK]
- Open pipeline by stage with $ value, close date, and probability: [PIPELINE_DUMP]
- Cohort retention by month (last 6 cohorts, percent retained at month 1, 3, 6): [COHORT_RETENTION]
- Average sales cycle in days for the last 10 closed-won deals: [SALES_CYCLE]
- Headcount and current burn: [HEADCOUNT_AND_BURN]

Produce in this exact order:
1. Header line: 'Forecast -- week of [DATE] -- [N] weeks out -- built from [LIST_DATA_SOURCES]'
2. Section 'Base case' -- markdown table with 12 rows (one per week) and columns: Week | Starting MRR | New MRR | Churn MRR | Ending MRR | Net New. Below the table, exactly 4 bullets naming the assumptions (each starts with a number). Each bullet under 18 words.
3. Section 'Bear case' -- same table format. Below: 4 bullets naming what changes vs base, each tied to a specific input.
4. Section 'Bull case' -- same table format. Below: 4 bullets naming what changes vs base.
5. Section 'The 3 deals that move the needle' -- table: Deal | $ value | Stage | Close date | Scenario impact.
6. Section 'The signal to watch this week' -- exactly 1 sentence naming the single metric whose movement by next Friday will tell me which scenario is playing out, and the threshold value.
7. Section 'What this forecast does NOT account for' -- 2-3 bullets naming the variables you could not model from the inputs.

Under 700 words total excluding tables. Do NOT use: 'roughly', 'approximately', 'should', 'might', 'could potentially'. Every number is a specific number. If an input is missing, write '[GAP -- need [exact data]]' in place of the assumption rather than inventing one.

Founders using this cut Sunday-night forecasting from 4 hours to 8 minutes.

You are rewriting one accomplishment from my week so it is forwardable. The raw version: [PASTE_THE_THING_YOU_DID]. My audience: [WHO_WILL_READ_IT]. The business outcome it produced: [BUSINESS_OUTCOME].

Produce exactly three versions of one sentence, each labeled:
- Version A (result-first): leads with the number/outcome, then the action.
- Version B (verb-first): leads with the past-tense verb, includes a named stakeholder or system.
- Version C (forwardable to a VP): rewritten so it makes sense with zero context, max 20 words, contains a dollar figure or percentage.

No hedging language. No 'helped contribute to'. If the raw version is missing a number, write '[NEED NUMBER: what to ask for]' inline.

The update that gets you noticed, and forwarded to your skip-level.

You are a skeptical prospect reading the offer page below. You have $5K-$15K to spend, you have been burned by one freelancer before, and you are evaluating 3 other people. Page:

[PASTE YOUR DRAFT OFFER PAGE]

Produce exactly three labeled sections:

1. 'Things I do not believe' (3 bullets): name the specific claims that read as unsupported. Quote the exact phrase from the page that triggered the doubt and explain in one sentence why a buyer would discount it.

2. 'Things I am still confused about' (3 bullets): name the specific decisions I cannot make from this page alone. Format: 'I cannot tell from this page whether [specific thing].'

3. 'The one edit that would make me book the call' (1 paragraph, max 4 sentences): name the single highest-leverage change. Reference the exact section and propose the replacement language. Must be specific enough to copy-paste.

Do NOT use: marketing-speak, 'consider adding', or any vague suggestion. Be the buyer, not the consultant.

Catches the trust gaps that make buyers bounce before they book a call.