Blog · AI and self-service analytics
Compare supplied financial scenarios with actual results using explicit assumptions and spreadsheet relationships. Reconcile revenue, costs and operating profit before asking AI to explain the differences.
AI financial modeling works best when finance owns the assumptions and calculation rules, while the external AI model explains outputs that tools have already computed and checked. Start with the scenario workbook you already use and compare its projections with actual results. A plausible explanation cannot substitute for a financial relationship that adds up.
Choose a concrete question: why did operating profit miss the base scenario, and which assumptions explain the gap? Keep the scope narrow enough that the supporting ledger and driver records can answer it.
The broader financial modeling guide covers how financial relationships fit together. Here, use a simple operating projection for a US distributor selling one standardized product. There is no inventory timing difference, and all variable product cost belongs to the units sold. Fixed operating expenses exclude interest and taxes; depreciation, if any, is included in that expense line.
Revenue = Units sold × Realized price per unit
Cost of goods sold = Units sold × Variable cost per unit
Gross profit = Revenue − Cost of goods sold
Operating profit = Gross profit − Fixed operating expenses
These are the relationships being tested, not a complete financial statement forecast. The SEC's financial statement guide distinguishes revenue, cost of sales, operating expenses and cash flows. Do not describe an operating-profit projection as a cash forecast, or assume it includes financing and taxes.
Keep the forecast version that finance approved before the month began. Preserve each scenario's assumptions rather than overwriting the base case when results arrive. Record the period, currency, business scope, preparer and approval status alongside it.
Bring actual sales units and revenue from the sales detail, actual product costs from the matching cost records, and operating expenses from the ledger. Reconcile those totals to the monthly financial close before using them to judge forecast performance. A changing account mapping can manufacture a variance without any underlying business change.
The budget and the latest forecast may serve different purposes. Link both to their own versions; budget versus forecast explains that distinction without mixing the reporting baselines.
The table is a completed monthly comparison. Put the labels in column A, the base case in B, the downside in C and actual results in D. The downside changes demand, price and unit cost while leaving fixed expenses unchanged. The units-sold row is a count; every other row is in USD.
| Driver or result | Base (USD) | Downside (USD) | Actual (USD) |
|---|---|---|---|
| Units sold | 10,000 | 9,000 | 9,500 |
| Realized price per unit (USD) | 50 | 48 | 49 |
| Variable cost per unit (USD) | 30 | 31 | 31 |
| Fixed operating expenses (USD) | 120,000 | 120,000 | 118,000 |
| Revenue (USD) | 500,000 | 432,000 | 465,500 |
| Cost of goods sold (USD) | 300,000 | 279,000 | 294,500 |
| Gross profit (USD) | 200,000 | 153,000 | 171,000 |
| Operating profit (USD) | 80,000 | 33,000 | 53,000 |
With the heading in row 1 and units in row 2, enter these base-column formulas in B6 through B9, respectively:
=B2*B3
=B2*B4
=B6-B7
=B8-B5
Copy the relationships across to C and D. In a review workbook, also compare calculated actual revenue and costs with the independently reconciled ledger totals. Deriving an actual unit price from actual revenue and then multiplying back is a consistency check; it does not independently prove that the original revenue was correct.
The downside produces $33,000 of operating profit. Actual operating profit is $53,000, which is $20,000 above the downside but $27,000 below the base. Both comparisons are correct; name the baseline whenever you report a variance.
Revenue missed the base by $34,500, but that alone does not explain the profit miss. Lower volume also reduces variable costs. Work the bridge in a stated order: change volume first at the base price and cost, then apply actual price and unit cost to actual volume, then change fixed expenses.
| Step | Calculation | Operating profit effect (USD) |
|---|---|---|
| Base operating profit | From base column | 80,000 |
| Volume | (9,500 − 10,000) × (50 − 30) | -10,000 |
| Realized price | 9,500 × (49 − 50) | -9,500 |
| Variable unit cost | −9,500 × (31 − 30) | -9,500 |
| Fixed expenses | −(118,000 − 120,000) | 2,000 |
| Actual operating profit | 80,000 − 10,000 − 9,500 − 9,500 + 2,000 | 53,000 |
Use these formulas against the first table for volume, price, unit cost and fixed expense effects, followed by the bridge check:
=(D2-B2)*(B3-B4)
=D2*(D3-B3)
=-D2*(D4-B4)
=-(D5-B5)
=B9+(D2-B2)*(B3-B4)+D2*(D3-B3)-D2*(D4-B4)-(D5-B5)-D9
The bridge check must equal zero. Its effects sum to negative $27,000, matching 53,000-80,000. Another allocation order can assign interactions differently; use one documented order consistently instead of comparing incompatible bridges.
For gross-profit definitions and the treatment of product costs, use gross profit margin. Keep that accounting choice fixed across all three columns.
Ask which assumptions contributed to the operating-profit gap and which ledger rows support each contribution. The answer can say that lower realized price reduced profit by $9,500 under the stated bridge convention. It cannot establish whether discounts, mix or a contract change caused that lower price without supporting records.
Ask the commercial owner to validate price assumptions, procurement to review unit costs and finance to confirm expense classification. NIST's AI Risk Management Framework places trustworthiness considerations across the use and evaluation of AI systems. For this financial review, a practical application is to retain both the numerical evidence and the human decision about any revised assumption.
Statistical prediction is a separate question from testing a supplied scenario. See predictive analytics for that distinction and AI analytics for the role of checked calculations in business questions.
Bring a sample of your own scenario workbook and actual-results export; Covirage maps the columns, and its tools compute and check the comparison figures. The external AI model explains the results with every figure traced to its rows. Create a free Covirage account, set up your business and bring your own files.
See FP&A reporting for the supporting approach to reviewing financial performance.
A useful projection still needs business assumptions, defined financial relationships and an accountable owner. For this workflow, supply an existing workbook and use checked calculations to compare scenarios. Treat any suggested assumption as a proposal that finance must review.
Bring the scenario workbook, a matching actual-results export and the supporting drivers. For a simple operating projection, those drivers include sales units, realized price, variable cost per unit and fixed operating expenses, with period and currency clearly identified.
Recalculate outputs from the stated assumptions, reconcile actual results to the ledger and explain the difference between forecast and actual through a bridge. Check signs, periods, units and expense classifications before accepting an explanation or changing the next forecast.