AI costing prompts · By role
AI costing prompts for the FP&A analyst
You live in the gap between the forecast you submitted and the actuals that arrive to contradict it, and most of your week goes on rebuilding, reconciling and explaining. AI can take real friction out of forecasting and variance work and cut the manual errors that creep in at speed, as long as it never substitutes its own numbers for your model's. These prompts add rigour, not guesswork.
In short
Use AI to structure forecasts faster, frame scenarios cleanly, and pull a tidy variance story out of a wide actuals file, all without letting it invent a single figure. The prompts below help you build a driver-based forecast, run scenarios you can defend, and explain variances against plan with the cause separated from the noise. Each one keeps the model working only from your data and forces it to show every formula before any result.
What an FP&A analyst should and should not ask AI to do
AI is a strong ally for the mechanical and structural parts of the job. It can lay out a driver-based forecast from the relationships you describe, generate consistent scenario variants once you give it the base, draft variance commentary that distinguishes volume from rate, and check a long actuals file for the slips that cause rework: a sign flipped, a subtotal that does not foot, a period mislabelled. These are exactly the tasks where speed introduces errors, so a careful second pass pays for itself.
It becomes risky the moment you let it produce inputs. Asking AI to "assume a growth rate," "estimate seasonality," or "fill the gaps in this forecast" yields plausible numbers with no link to your business, and a forecast built partly on invention is worse than one with honest gaps. Keep the drivers, the assumptions and the judgement about what is reasonable in your hands. Let AI accelerate the build and audit your arithmetic, not author the future.
Three prompts to start with
1. Build a driver-based forecast
Use this to structure a forecast around real drivers rather than last year plus a percentage. It builds on the budgeting and forecasting page.
You are an FP&A assistant helping build a driver-based forecast. Work only from the data I give you. Do not invent any rates, volumes or growth assumptions. List every assumption, and label anything missing as DATA MISSING with what you need from me. My data: - Cost or revenue lines and their driver (for example, cost scales with order volume): [paste] - Historical values for each driver and each line: [paste] - The forecast drivers I am giving you for the period ahead: [paste] Steps: 1. For each line, show the unit relationship (line value / driver) from history as a formula, then the value. 2. Apply only the forecast drivers I supplied; do not assume any I did not give. 3. Show each forecast line as a formula before the number. 4. Separate clearly: what you calculated, what I assumed via the drivers, and any concern you would flag. 5. Check that the forecast lines sum to the total and flag any discrepancy.
2. Run defensible scenarios
Generate consistent best, base and downside cases without smuggling in invented inputs. See budgeting and forecasting.
You are helping an FP&A analyst build three scenarios from one base forecast. Work only from the data I give you. Do not invent the scenario inputs; use only the changes I specify. Show each calculation as a formula before the value. My data: - Base forecast by line: [paste] - The specific changes for each scenario (for example, downside = order volume -10%): [paste for base, upside, downside] Steps: 1. Restate the base so I can confirm it. 2. For each scenario, apply only the changes I specified and show the affected lines as formulas. 3. Present the three scenarios side by side with the delta to base for each. 4. State the single assumption each scenario most depends on. 5. Check each scenario foots to its total and flag any that does not.
3. Explain variances against plan
Pull a clean volume-versus-rate story out of a wide actuals file. Links to board reporting.
You are helping an FP&A analyst explain variances of actuals against plan. Work only from the figures I give you. Do not invent causes or numbers. If a variance needs an explanation I have not provided, write "cause to confirm". My data: - Lines with actual, plan, and where available the volume and rate behind each: [paste] Steps: 1. Compute each variance as a formula (actual minus plan) and rank by absolute value. 2. Where I gave volume and rate, split the variance into a volume effect and a rate effect, each as a formula. 3. For the top three, draft one plain sentence each, attributing a cause only where I confirmed it. 4. Separate clearly: what you calculated, what I confirmed, and what is unexplained. 5. Check that the individual variances reconcile to the total variance and flag any gap.
The one rule
Work only from the data I give you. Do not invent any numbers, rates or volumes. Label anything missing as DATA MISSING and tell me what you need.
A forecast built on an invented input is harder to fix than one with an honest gap. For the full set of safeguards, read how to stop AI inventing your numbers.
Back-test your forecast against actuals
The prompts above build the forecast and the scenarios. This one turns the exercise around: it compares what you forecast against what actually happened, so you can measure your bias and correct the next forecast instead of repeating it. Run it once you have actuals for a period you previously forecast.
Forecast-versus-actuals comparison
You are an FP&A assistant helping compare a prior forecast against actuals, so an analyst can measure forecast bias and correct it. Work only from the data I give you. Do not invent any numbers or causes. Show every formula before the value. If a variance needs an explanation I have not provided, write "cause to confirm". I want to see where my forecast was systematically off, by how much, and in which direction, so the next forecast is better, not just different. Step 1. Take the lines with, for each: the value I forecast, the actual, and where available the driver behind each. Use only what I provide. Step 2. For each line, compute the forecast error (actual minus forecast) and the error as a percentage of forecast, each as a formula. Rank by absolute error. Step 3. Where I gave the driver, split the error into a driver-volume effect and a rate effect, each as a formula, so I can see whether I mis-forecast the driver or the relationship. Step 4. Look across the lines for a pattern: am I consistently high or low, and on which kind of line. State the bias plainly, citing the rows that show it. Step 5. For the three largest errors, draft one sentence each on what to adjust in the next forecast, attributing a cause only where I confirmed it. Step 6. Reconcile the individual errors to the total forecast-versus-actual gap and flag any discrepancy. Separate clearly what you calculated, what I confirmed, and what is unexplained.
A forecast you never compare against actuals cannot improve. The value is not in being right the first time, it is in seeing the bias clearly enough to correct it.
Make it an adjustable model you can reuse
The prompts on this page produce a forecast, scenarios and variance work in the chat. This add-on turns that into a file you can reuse each cycle, an adjustable Excel model with a driver input sheet, a clean PDF, or a short deck. Paste it after the forecast or variance prompt once the numbers have run.
Now package this forecast into a deliverable I can reuse each cycle. Ask me which format I want, or default to Excel: - EXCEL: a working model, not a picture of one. Put the drivers and assumptions (historical values, forecast drivers, scenario changes) on one clearly marked input sheet, and drive every forecast line and every scenario with live formulas that reference those inputs, so changing a driver reflows the whole forecast and all three scenarios. Add a summary sheet with the base-versus-scenario table and, if the tool allows, a chart. Label every assumption cell and keep the scenario switches obvious. - PDF: a clean report, title, one-paragraph summary, the forecast table, the three scenarios side by side, one chart, and an assumptions appendix. No chat formatting. - PPT: 5 to 7 slides, the question, the driver logic in one slide, the base forecast, the scenario range, the single assumption each scenario most depends on, next steps. Rules for the deliverable: - Keep all drivers and assumptions adjustable and visible; never hard-code a result I might want to change. - Carry through every assumption and label illustrative figures as illustrative. - Make it obvious which numbers are forecast and which, if any, are actuals. - Add a discreet footer or last-slide credit line, small and unobtrusive, exactly as written below. Credit line to embed (use verbatim, in the document footer or final slide): "Model scaffolding based on the FP&A forecasting prompts from costandprofitability.com/ai-costing-prompts/for-fpa-analysts" Keep it to one small line; it should read as a quiet source note, not an advertisement.
The credit line is deliberately modest, a source note rather than a watermark, so the model is comfortable to circulate each cycle while the method stays traceable to where it came from.
Faster forecasts rest on a sound cost base
Prompts will speed your build and tighten your variance work, but the quality of any forecast depends on the cost model feeding it. Where costs are allocated badly, even a perfect forecast forecasts the wrong thing. We help finance teams put a defensible cost and profitability model under the planning process, so the drivers you forecast on are the ones that actually move the business. Start by seeing where your current model stands.
Related
Proof
A distributor in New Zealand. €1.335M of cost-to-serve made visible, then halved, and 830 loss-making customers brought down to 295.
Read the case study →Who you would be talking to
Miguel Guimarães, Founding Partner
Cost and profitability practitioner for 25+ years. Presented the Damco cost-to-serve case at Managing for Profit (Amsterdam RAI, December 2009), on the same programme as Robert S. Kaplan.
Call +351 910 313 731
