How to forecast AI spend

What does it take to forecast AI spend?

A forecast is only as credible as its baseline. A number pulled from last month's invoice tells the CFO what you spent, not what you will spend or what you should spend.

Spendassay is the independent audit that supplies the missing inputs. It ties every dollar to the seat or token it bought, prices the waste, and sets a recoverable target so your projection has a floor and a ceiling.

The audit is read-only and metadata-only. It surfaces and prices the waste; your own admins take the action through the step-by-step instructions Spendassay hands them.

How do you build the baseline run-rate?

Start with the spend side. The Ramp connector reads corporate-card charges; CSV import does the same with zero live integrations, so you can forecast before any API access is granted.

Add the usage side to make the number explainable. GitHub covers Copilot seats, Cursor covers its own, and the OpenAI and Anthropic connectors read API token consumption. Okta and Google Workspace surface SSO and OAuth grants, which expose shadow AI that never hit a formal invoice.

Reconciliation is what turns two feeds into a baseline. When a dollar of card spend maps to an active seat or a metered token, the run-rate you forecast from is measured, not assumed.

How do you project future spend?

Split the projection by cost shape. Seat subscriptions are largely fixed until renewal, so they extend at the contracted per-seat rate times your current roster. API token spend is variable, so it follows the recent consumption trend.

Renewals are the inflection points a naive run-rate misses. Spendassay's renewal-opportunity findings list each upcoming contract, its current price, and a cited benchmark, so you can model the renewed rate instead of assuming a flat line.

Every projected number inherits a proof level: Counted beats Compared beats Estimated. A forecast built on counted token history is firmer than one built on an estimated projection, and the tier tells the CFO which is which.

How do you factor recoverable waste into the forecast?

A run-rate projection assumes today's waste continues. That is the wrong number to commit to, because a chunk of it is recoverable inside a quarter.

Spendassay prices each waste type as a dollar-valued finding, and every finding exposes its formula so the number survives CFO scrutiny. Findings never double-count, and reporting stays team-level rather than ranking individuals.

The audit sets the recoverable target at 10 to 15% of seat and subscription spend within 30 days. Subtract that target from the projected run-rate and you have two lines to forecast against: the do-nothing path and the acted-on path.

How do you keep the forecast accurate over time?

The forecast is a living number, not a one-time slide. New hires add seats and a product launch spikes token spend, so the run-rate you projected last quarter is already stale.

Recovery closes the loop. Admins run the step-by-step instructions to reclaim seats and use the negotiation packages at renewal; the banked Recovered statement records the dollars that actually landed, so forecast versus actual is one comparison, not a debate.

Each sync recomputes the audit. As measured data accumulates, modeled estimates get replaced with measured ones, and the forecast tightens without a manual rebuild.

Step by step

1. 1

   Connect a spend source

   Connect Ramp or upload a CSV of corporate-card and invoice charges. This establishes the dollar run-rate, and CSV import works with zero live integrations so you can start forecasting immediately.

2. 2

   Connect a usage source

   Connect GitHub, Cursor, OpenAI, Anthropic, Okta, or Google Workspace. This ties the dollars to seats, tokens, and SSO grants, so Spendassay can reconcile every charge to what it actually bought.

3. 3

   Review the priced findings

   Open the findings for unused seats, underused seats, shadow AI, tool overlap, renewal opportunities, and quality-trend regressions. Each carries a proof level and exposes its formula, so you know how firm each number is.

4. 4

   Set the recoverable target

   The audit flags 10 to 15% of seat and subscription spend as recoverable within 30 days and lists upcoming renewals with benchmarked prices. Subtract that target from the projected run-rate to produce a defensible forecast.

5. 5

   Act via set of step-by-step instructions and track actuals

   Hand the step-by-step instructions and negotiation packages to your admins to execute; Spendassay itself makes no changes to vendor systems. The banked Recovered statement then tracks realized savings against the forecast each sync.

Common questions