AI Costing Prompts · By task
Rank customers by profit and build the whale curve
Paste your net profit by customer into any capable AI model and get a ranked profitability table, a cumulative running total, and a plain-language read of where your business actually makes and loses money. This prompt does the arithmetic and the interpretation. It does not invent a single figure.
In short
Give the model net profit for each customer. It ranks them from most to least profitable, sums the total, builds a cumulative running total in ranked order, and finds the peak of that curve. The peak is almost always higher than your reported total, because the loss-making customers at the tail pull the kept profit back down. The prompt names which customers destroy value and quantifies how much profit you would recover by fixing them.
What the prompt is doing
This is the whale curve, the single most persuasive picture in customer profitability analysis. When you rank customers by net profit and plot the cumulative total, the line climbs steeply at first because a handful of customers carry the business. It then flattens across the marginal accounts that barely cover their cost to serve. Finally it dives, because the customers at the bottom are not low-margin, they are loss-making, and each one subtracts from the total you have already built.
The reason the peak matters is that it is the profit you would keep if you simply stopped losing money on the tail. The gap between the peak and your reported total is not theoretical. It is real profit being consumed by accounts you are actively serving. Reading the whale curve correctly turns a vague sense that "some customers are unprofitable" into a specific number and a specific list.
The prompt
You are a profitability analyst. Work only from the data I give you. Do not invent any numbers, customers, or facts. If something is unclear or missing, flag it as an assumption rather than filling it in. Here is net profit by customer (EUR): C1 96,000 C2 71,000 C3 52,000 C4 33,000 C5 18,000 C6 9,000 C7 2,000 C8 -7,000 C9 -21,000 C10 -38,000 Do the following: 1. Rank the customers from most to least profitable. 2. Calculate the total net profit across all customers. 3. Build a cumulative running total in ranked order. 4. Identify the peak of the cumulative curve and state what percentage of total net profit the top customers represent at that peak. 5. Explain in plain language what the whale curve shows here and which customers are destroying value. 6. List any assumptions you had to make. Output a table with columns: Rank, Customer, Net profit, Cumulative.
A worked example
| Rank | Customer | Net profit (EUR) | Cumulative (EUR) |
|---|---|---|---|
| 1 | C1 | 96,000 | 96,000 |
| 2 | C2 | 71,000 | 167,000 |
| 3 | C3 | 52,000 | 219,000 |
| 4 | C4 | 33,000 | 252,000 |
| 5 | C5 | 18,000 | 270,000 |
| 6 | C6 | 9,000 | 279,000 |
| 7 | C7 | 2,000 | 281,000 |
| 8 | C8 | -7,000 | 274,000 |
| 9 | C9 | -21,000 | 253,000 |
| 10 | C10 | -38,000 | 215,000 |
Total net profit = 215,000 EUR. The cumulative curve peaks at 281,000 EUR after C7, the last profitable customer.
The seven profitable customers generate 281,000, which is 131% of total net profit. The top two alone (C1 plus C2) contribute 167,000, or 78% of the total. The bottom three (C8, C9 and C10) destroy 66,000 between them, dragging the kept profit from 281,000 down to 215,000.
In plain language: the curve climbs steeply because a few customers do the heavy lifting, flattens across the marginal accounts that barely contribute, then dives as the value destroyers subtract from the total. Bringing the three loss-makers to break-even would lift net profit from 215,000 to 281,000, a 31% increase with no new customers.
Assumption flagged: net profit is assumed to be fully loaded and comparable across customers. If some accounts carry unallocated cost to serve, the ranking could shift before any action is taken.
What it costs you to run
The sample input above is roughly 250 tokens. With ten customers and a short instruction set, a single run of this prompt is inexpensive on any current model, well under a cent on most. Even scaling the input to a few hundred customers keeps the cost trivial relative to the profit the analysis surfaces.
Token figures are approximate and vary by model and tokeniser.
The guardrail that matters
Work only from the data I give you. Do not invent any numbers, customers, or facts. If something is unclear or missing, flag it as an assumption rather than filling it in.
That single line is what keeps the output trustworthy. Without it, a model asked to "analyse customer profitability" will happily fabricate plausible figures. With it, the model stays inside your data and tells you when it cannot. See how to stop AI inventing numbers for why this matters in costing work.
When you need the real model
This prompt works on net profit you already have. The harder question is whether that net profit is right in the first place, which means tracing cost to serve down to each customer with a defensible method rather than an averaged allocation. That is the work behind a real profitability model.