Stop asking AI to write things. Start asking it to decide things.

1. The business context. Who you sell to, at what price, with what constraints. Without it, the model answers for a generic company that does not exist.
2. The actual data. Attached, pasted or reachable through a connector. This is the part that turns an essay into a finding.
3. The decision. Not "analyse this" but "tell me which of these three to stop doing, and what it costs me if you're wrong".
4. The format. The shape the answer has to arrive in, including what to do when the data won't support a conclusion.

Analyse my pricing and suggest improvements.

We sell [service] to [customer type], currently at [price], with delivery costs of roughly [amount] per engagement and [number] delivered in the last 12 months. Attached is our closed-won and closed-lost data for that period, including the discount applied to each deal.

Tell me whether our list price is too low, about right, or too high, and give me the single price change you would make. Support it with: the win rate at each price band we actually sold at, what discounting did to win rate, and the revenue impact of your recommendation at our current volume.

If the data cannot support a confident recommendation, say so and tell me what we would need to record for the next 90 days to answer it properly.

Attached is our lead and deal data for [period]. For each lead source, calculate: number of leads, number of closed-won deals, total and median revenue, and the average days from first touch to close.

Rank the sources by revenue produced rather than lead volume, and show me where those two rankings disagree. Tell me which source I should move budget away from, and which I should move it toward, with the expected effect at current spend.

Use the rebuilt pricing prompt above, with your own closed-won and closed-lost data attached.

Attached are the accounts we lost in [period], with their usage, support history and billing records.

Work backwards from each cancellation and identify the earliest observable signal that the account was at risk, with the date it appeared and how many days it preceded the cancellation. Then tell me which single signal has the best combination of early warning and reliability, and what an alert built on it would have caught.

Attached is our current customer list with usage and revenue data, plus the history of accounts that have previously expanded.

Identify the accounts that most resemble our past expansions in the 60 days before they expanded. Rank them, show the specific pattern each one matches, and tell me what to say to the top three.

Attached is our card statement and vendor list for the last 12 months.

Group every recurring charge by what job it does for us. Flag: tools whose job overlaps another tool, charges that increased more than 20% year over year, anything billed per seat where our headcount has changed, and anything we are paying annually for a job we stopped doing.

Give me a cancel list ranked by annual saving, and separately note anything where cancelling carries a migration cost I should weigh first.

Here is a description of how [process] runs today, step by step, with who does each step and how long it takes: [description].

Tell me where the work waits rather than where it takes effort, and estimate the total elapsed time versus the total hands-on time. Identify the single step that, if it went to zero, would most reduce elapsed time, and tell me what would become the constraint afterwards.

We need [capability] for a team of [size] doing [volume] per month, with these non-negotiables: [list]. Our budget is [amount].

Compare the realistic options on total cost at our volume, not headline price, including implementation effort and what it costs us to leave in two years. State your recommendation, the strongest argument against it, and what would have to be true for the runner-up to win.

Here is our website copy: [paste]. Here are three competitors' pages: [paste].

Tell me what claim each competitor is making that we are not, what we claim that nobody else does, and which of our claims a buyer could not verify. Then write the one sentence that positions us against the strongest of them, using only claims our own evidence supports.

Attached is our closed-won data with company size, sector, source and deal value, plus our closed-lost data for the same period.

Describe the customer we actually win, as opposed to the one we say we target. Give me the three attributes with the strongest relationship to winning, the three most correlated with losing, and a disqualification question we could ask on the first call.

Attached are [number] sales call transcripts, labelled won or lost.

Extract every objection, group them into themes, and rank by frequency in lost deals. For each theme give me the count, two verbatim quotes, and whether our current website answers it anywhere.

Here is the job spec for a [role] we are about to hire: [paste].

Turn it into a scorecard: the five outcomes this person must deliver in their first year, each with a measure, ranked by how much of the role's value sits in it. Then give me two interview questions per outcome that would surface real evidence rather than a rehearsed answer, and the answer pattern that should worry me.

Here are the notes or transcript from [meeting]: [paste].

Extract only: decisions actually made, with who owns each one and by when; decisions that were discussed but left open, and what is blocking each; and commitments made to people outside the room. Ignore everything else. If ownership or a date was never stated, write "unassigned" rather than inferring it.

Here is everything on my list this week: [paste]. Our goal this quarter is [goal], and the number that measures it is [metric].

Sort these into: moves the metric this week, moves it later, keeps the business running, and does neither. For anything in the last group, tell me what happens if I simply don't do it. Then tell me the one thing on this list I should do first tomorrow morning.

The five questions from the one-page test, answered against your own numbers before you spend anything on automation.

The long consultant-style diagnostic from the connector setup, run against live CRM, Slack and email data.