Understand your cloud and AI spending

Your spending has increased. Should you be concerned? Find what changed: more customers, a new service or operations repeated by mistake. An increase can support useful activity; the amount alone does not tell you whether it is a problem.

Cloud spending here means technical services billed to your company, such as hosting or databases. AI spending includes paid calls to a model, for example. Start with one provider and one specific question.

Fictional example: a team pays €1,200 in August, then €1,500 in September for the same services. The difference is €300. The billing breakdown shows €200 more for hosting and €100 more for AI calls.

The team launched a feature in September. That is a lead, not yet a proven explanation. It checks the launch dates, call counts and errors before deciding to reduce spending. The figures in this guide are illustrative and do not represent customer results.

Choose two completed months and keep the same accounts, services and currency. Check that the data is up to date. If you added a provider to the second report, separate its cost before concluding that existing services have become more expensive.

• Use consistent treatment of taxes, discounts and refunds.
• Look for an annual payment or a prepaid commitment: it can explain a higher bill without increased consumption.
• Record missing data. An unknown amount is not zero.

In our example, the same services appear in both reports and both months are complete. If August were incomplete, the €300 difference would not reliably measure the increase.

Start with the provider whose amount changed the most. Then inspect the available breakdown: account, service, project or AI model. Compare the timing of the increase with changes the team actually made.

• More activity: are there more customers or completed tasks?
• A more expensive choice: did you change a model, service or configuration?
• Repeated work: do errors cause several attempts for one task?

Billing details help locate the spending. To know whether calls succeeded or were repeated, also consult your application’s logs and measurements.

Fictional example, continued: the team confirms that the extra €100 in AI spending comes from repeated calls after errors. It decides to fix that behavior, gives one person responsibility and sets a review date.

After the change, the team compares spending and successfully completed tasks. It also checks errors and response time. A lower cost is useful only if the expected service still works.

If the cause is unknown, an investigation can be the action. Write down what needs checking, who will do it and by when. Do not claim savings before observing them.

Keep a short shared note after the meeting. Someone else should be able to understand the situation and continue the work from it.

• What changed: the service, period and amount.
• What is confirmed and what still needs checking.
• The agreed action, its owner and its review date.
• The result observed at the next review.

The online version links to the detailed method and Fivemetrics resources. You can use this guide with your usual invoices or reports.