AI Cost Forecasts Need Prices and Discount Rates That Match

By Srithan Gade

A forecast can calculate accurately yet misstate costs when its cash flows and discount rate use incompatible inflation assumptions.

An AI tool can get the arithmetic right while combining incompatible assumptions in a present-value estimate. Future costs that include general inflation will have a misstated present value if discounted at a rate that excludes it. To correct this, discount nominal cash flows at a nominal rate, or convert the flows into purchasing power at a stated base date and use the corresponding real rate.

The following illustrative support-cost example shows how to check that match. It uses no client data and reports no experimental test of an actual AI error.

Which prices does the forecast use?

In “AI Adoption Is Everywhere. Returns Are Not. Why?”, Terence Tse challenges the applicability of traditional technology payback standards to AI investment assessment. Selecting a reasonable horizon is one consideration. Making monetary amounts comparable within that horizon is another.

Nominal costs are the cash payments expected on their scheduled dates. Real costs express those same future costs in purchasing power as of a stated price-base date. This date can be today or another specified date; in the example, it is also the valuation date.

Aswath Damodaran’s NYU valuation notes explain that cash flows and discount rates should be consistent in risk, type, currency, and nominal or real terms. This article focuses on the last requirement. It does not select a discount rate for a company.

Converting a future cost into base-date prices does not simultaneously discount it. The life-cycle costing manual from the National Institute of Standards and Technology treats inflation adjustment and discounting as separate procedures. A cost due one year from now can be expressed in a base date’s purchasing power and still require discounting at a real rate.

When reviewing a forecast, the immediate task is to determine what each figure means. Record the currency, payment date and price basis beside each forecast cost. For a real cost, specify the price-base date. The label “annual support: £10,000” does not reveal whether £10,000 is a scheduled payment, a current-price estimate or an already discounted figure.

How much can one mismatch change the result?

Suppose a fixed quantity of software support for an AI workflow costs £10,000 at today’s prices, with payment exactly one year from today. Assume 4% general inflation with no relative price change. The future payment would then be £10,400. These inputs are hypothetical; taxes, benefits and exchange-rate effects are excluded. For this illustration, treat general inflation as certain and keep the treatment of risk unchanged.

Assume an annual real discount rate of 5%, keeping the other assumptions unchanged. The compound conversion gives a corresponding nominal rate of 9.2%: 1.05 times 1.04 equals 1.092. Simply adding the percentages would give 9%, omitting the compounding effect.

Each present value below equals the future cost divided by one plus the annual discount rate. Costs are reported as positive amounts.

Source: Original hypothetical illustration. Figures are rounded to the nearest penny.

The mismatch overstates present cost by £380.95, or 4% of the correct present cost before rounding. This valuation difference arises from combining incompatible assumptions. It is not a saving and does not change the supplier’s future £10,400 payment. That payment still matters for cash planning.

Both consistent methods value the same cost at the same date. Changing its representation should not change the result. This check tests consistency; it does not validate the forecast or the selected discount rate.

What changes must be kept intact in the forecast?

To avoid stripping out every increase in cost, distinguish between movements in the general price level and relative price changes. HM Treasury’s Green Book explains that a particular good or service can become cheaper or more expensive relative to the general price level. That distinction also applies to business costs. It supplies a definition here, not a public-sector discount rate for a corporate model.

If workflow volume increases and requires more support at the same real price per unit, total real cost increases. Likewise, a supplier price increase above general inflation leaves a real price increase after removing general inflation. Changes in efficiency or support tier could affect how much support the workflow requires.

A contract that fixes a £10,400 payment one year ahead specifies a nominal cash amount, even without an escalation formula. Do not add inflation solely because the payment is in the future. First establish what the quoted figure represents before adjusting it.

Before instructing an assistant to “strip out inflation”, separate general price changes from workflow volume, resource use and contract terms. Keep both quantity and unit-price assumptions visible. Otherwise, an attempt to fix the mismatch could eliminate an actual change in the planned workflow.

Also name the general inflation assumption underlying the conversion. A supplier’s quote and a finance worksheet may use different dates or price bases. If the sources are silent, flag the assumption as unresolved and ask the person responsible for the estimate. Polished wording should not conceal the missing information.

What assumptions should be easy to verify?

Request the assumptions alongside the present value, with links to supporting documents. For an AI-generated draft, the following is a useful checking procedure:

• Trace each cost to a dated quote, contract or planning assumption, retaining the currency and payment schedule.
• Identify whether each amount represents expected future cash or an amount in constant prices, and name the price-base date of the latter.
• Separate general inflation from changes in volume, resource consumption and relative prices, recording the source of each assumption.
• Confirm whether the discount rate is nominal or real. Have a finance specialist justify its appropriateness, including its period and treatment of risk.
• Recalculate using consistent assumptions. When both nominal and real representations are available, reconcile their present values before rounding.

In the example, the summary can state: “Support costs £10,400 in one year. Assuming 4% annual inflation and a 5% annual real discount rate, its present value today is £9,523.81. The equivalent annual nominal discount rate is 9.2%.” Keep this statement with its assumptions, and avoid relabelling the lower figure as “annual cost”.

If a rate has no nominal or real label, request the underlying finance assumption. The assistant should flag the uncertainty instead of choosing whichever version yields a more attractive total.

Verify each figure with a calculator or spreadsheet, and check whether the assistant’s explanation preserves its meaning. Perform the same checks on costs and benefits before combining them. This example values one cost only. Assessing a project’s net present value or financial return requires all relevant cash flows and appropriate risk assumptions.

Who should approve the financial calculations?

An AI briefing should make its financial assumptions transparent. Include the price basis, valuation date and discount-rate label in the reporting form so subsequent summaries retain that information. When an assumption changes, review the cash flows, rate conversion and explanation together, and revise the affected items. The person approving the forecast must check the calculation and the interpretation of the result. Precise totals become useful when readers understand the future payments they represent and the assumptions used to value them.

Acknowledgements

The author created this draft, and ChatGPT assisted with grammar, technical source checks, reference formatting and document preparation.

About the Author

Srithan Gade is a senior at Lenape High School in New Jersey and co-founder of VenoxAI, a technology services company serving small businesses and nonprofits. His interests include business applications of AI and interpreting evidence. His VenoxAI affiliation is a commercial interest relevant to this article.