Guide
How to Set AI Budgets and Alerts
Spendassay approaches AI budgets as an audit, not a throttle. Connect a spend source and a usage source, and Spendassay prices your waste, sets a recoverable target of 10 to 15% within 30 days, and fires alerts when spend or usage crosses the thresholds you set. Your admins act through step-by-step instructions.
Can Spendassay set a hard budget that caps AI spend?
No. Spendassay's audit is read-only and never writes to, routes, or throttles vendor systems. It sets a dollar-priced recoverable target instead of a hard cap, then tracks recovery against it. Alerts tell you when spend or usage crosses a threshold. Your own admins act through step-by-step instructions or vendor consoles.
A hard budget cap would require writing to your vendors to block seats or cut off API keys. Spendassay does not do that. It is a buyer-side audit that reads metadata only, and it never reads source code, prompts, or message content.
Instead of a cap, Spendassay gives you a target. It measures what you spend across seats, API tokens, and infra, prices the waste it finds, and sets a recoverable number. The north-star target is 10 to 15% of seat and subscription spend surfaced as recoverable within 30 days.
When spend or usage crosses a threshold you set, an alert fires. The alert points at the finding and its step-by-step instructions. The action stays with your admins.
What data does Spendassay need to set a budget target?
Spendassay needs one spend source and one usage source. The spend source is corporate-card data (Ramp) or a CSV export. The usage source is a seat or token connector such as GitHub, Cursor, OpenAI, Anthropic, Okta, or Google Workspace. CSV import works with zero live integrations.
The spend source tells Spendassay what you pay. Connect Ramp for corporate-card feeds, or import a CSV of card and invoice lines. This is where shadow AI first shows up, detected from card spend the finance team did not expect.
The usage source tells Spendassay whether that spend is used. Seat connectors (Copilot, Cursor, ChatGPT, Claude) and token connectors (OpenAI, Anthropic) reveal unused seats, underused seats, and token anomalies. Okta and Google Workspace add SSO and OAuth grants that expose shadow AI.
You do not need every connector on day one. One spend source plus one usage source is enough to produce a first set of dollar-priced findings and a recoverable target.
How do I set spend and usage alerts?
Set alert rules against the metrics Spendassay already tracks: monthly AI spend, unused-seat count, token spend, and shadow-AI detections. Each rule watches a metric and a threshold. When the metric crosses it, Spendassay notifies you and links the underlying finding. Alerts inform your admins; they never act on vendor systems.
Alert rules watch the numbers behind your AI cost report. Common rules include total AI spend above a monthly ceiling, unused seats above a count, token spend above a dollar line, and any new shadow-AI tool detected on a corporate card.
Every alert carries the finding that triggered it, with a proof level (counted, compared, or estimated) and the formula behind the number. You see why the alert fired, not just that it fired.
Because the audit is read-only, an alert is a signal, not an action. It routes your attention to the step-by-step instructions. Your admins decide whether to reclaim seats, consolidate overlapping tools, or open a renewal negotiation.
How does an alert become recovered dollars?
Each finding ships with a set of step-by-step instructions your admins run, or a renewal negotiation package. The instructions list the affected seats and the exact console steps. As seats are reclaimed and the finding closes on a later sync, Spendassay banks the amount into a Recovered statement your CFO can read.
A finding names the waste and prices it, so finance can see exactly which seats are unused or which renewal is overpriced against public list rates. The dollar figure carries its proof level so the number holds up.
The instruction set is the action layer that stays with you. It gives your admins the affected-user list and the step-by-step console actions to remove seats or downgrade a plan. For renewals, Spendassay generates a negotiation package with a cited benchmark and drafted asks.
Recovery is measured, not assumed. When a later sync confirms the seats are gone, the finding closes and the dollars move into the Recovered statement, with the date banked. That statement is the proof your budget target was met.
What thresholds should I set for AI budgets and alerts?
Start with thresholds tied to your findings, not round numbers. Set the spend ceiling near your current monthly AI run-rate so any jump surfaces fast. Alert on unused seats above roughly 5% of paid seats, on any new shadow-AI tool, and on token spend that breaks its trailing trend.
Anchor the spend ceiling to your real run-rate. If you spend $18K a month across AI tools, a ceiling a little above that catches sudden increases without daily noise.
Alert on unused seats as a share of paid seats. Idle seats above about 5% usually clear the cost of the review. Spendassay flags seats dormant past 60 days and underused seats over a trailing 30-day window.
Always alert on new shadow-AI detections, regardless of size. A new AI tool on a corporate card is worth knowing about even at $50 a month, because it signals ungoverned adoption. TODO(spendassay: confirm any default alert threshold values shipped in the product to cite exact numbers).
Step by step
- 1
Connect a spend source
Connect Ramp or import a corporate-card and invoice CSV. This gives Spendassay the dollars you pay across AI seats, API tokens, and infra, and surfaces shadow AI from unexpected card spend.
- 2
Connect a usage source
Add a seat or token connector such as GitHub, Cursor, OpenAI, Anthropic, Okta, or Google Workspace. This tells Spendassay whether the spend is actually used. CSV import works if you have no live integrations.
- 3
Set your recoverable target and alert thresholds
Accept the recoverable target Spendassay prices (aim for 10 to 15% within 30 days) and create alert rules on monthly spend, unused-seat count, token spend, and new shadow-AI detections.
- 4
Review the dollar-priced findings
Open each finding to see the waste, its dollar value, its proof level (counted, compared, or estimated), and the formula behind the number. Confirm which findings are worth acting on.
- 5
Act through the step-by-step instructions
Run the step-by-step instructions: your admins use the affected-seat list and console steps to reclaim seats or consolidate tools, or send the renewal negotiation package. Spendassay tracks recovery and banks confirmed savings into the Recovered statement.
Common questions
Does Spendassay automatically cut off AI spend when a budget is hit?
No. Spendassay's audit is read-only and never writes to, routes, or throttles vendor systems. When an alert threshold is crossed, it notifies you and links the finding and its step-by-step instructions. Your own admins take the action in the vendor console.
Can I set AI budgets without connecting any live integrations?
Yes. CSV import works with zero live connectors. Upload a corporate-card and invoice CSV as your spend source, and a seat or usage export as your usage source. Spendassay produces dollar-priced findings and a recoverable target from those files alone.
What does Spendassay alert on?
Spendassay can alert on total monthly AI spend, unused and underused seats, token spend anomalies, tool overlap, renewal opportunities, and newly detected shadow AI. Each alert carries the underlying finding, its proof level, and the formula behind the number.
How does Spendassay prove a budget target was met?
Recovery is measured, not assumed. When a later sync confirms reclaimed seats or a closed finding, Spendassay banks the amount into a Recovered statement with the date. That statement gives your CFO an audit-grade record of savings against the target.
Run the audit on your own AI spend
Connect read-only sources and get a CFO-credible AI cost report. Free to start, about 10 minutes to connect.