Your AI model bill grows with your product. First, find out exactly what you are paying for

A cost audit for companies whose own product runs on AI: model calls, tokens, caching, the right model for each task.

In short

A noticeable share of an AI bill often goes not to useful work but to repetition: the same long context in every request, an expensive model where a simpler one would do, unlimited retries, calls whose result nobody uses. The audit shows what your bill is made of, and gives a savings calculation based on your own data, not a percentage promise.

What we check

  • What makes up the bill: which product features call models, how many tokens each call uses, and how much of that is input context.
  • Model choice: where a cheaper model does the job without losing quality, tested on your examples, not on provider tables.
  • Caching: a repeated system prompt and context that can be cached at the provider (prompt caching), and answers that do not need to be requested again.
  • Context length: conversation history and RAG documents sent to the model in full when it only needs part of them.
  • Retries and errors: unlimited retries, timeouts, calls whose result nobody uses.
  • Agents: loops in which an agent calls the model and tools more times than the answer requires.
  • Tracking: whether spend is visible by feature, customer and model, whether there are limits and alerts, or whether the bill comes as a surprise at the end of the month.
  • A single gateway: whether all calls go through one place where they can be counted, limited and switched between models.

What you get

  • A cost map: the bill broken down by product feature, model and call type. You see exactly what you are paying for.
  • Measures for each item: what to change, how much it saves, calculated on your data, and what it may cost in answer quality or latency, so you can make an informed decision.
  • An order of work: the measures that deliver the most at the lowest risk come first.
  • A quality check: for every model switch or context cut, how to confirm on your examples that the answers did not get worse.

The scope and the price are set after a short call about the product and how it uses models.

After the audit

  • We fix it. We make the changes in the order set by the audit and check the result against the real bill. We stay responsible for what we did: if spend grows again after our changes without a growth in load, we look into it.
  • Your team fixes it. We hand over the report and the measures in a form your team can carry out without us. Once the changes are in place, we run a follow-up check: that spend went down where expected and that answer quality held.

Ongoing cost monitoring

A separate service for products whose AI load grows with them:

  • a monthly report: spend by feature and model, the changes over the month and their causes;
  • alerts when spend moves away from the usual;
  • a review of new models and provider pricing: whether it is worth switching.

Why us

We build products with AI inside them ourselves: RAG systems, chatbots, agents and MCP connections. Controlling model costs is part of every project we do: prompt caching, a single gateway for all calls, the right model for each task. The audit is based on what we do in our own work.

Frequently asked questions

Do you need access to the code?

It is better, because that shows where each call comes from. We can start with the provider bill and the call log, and bring in the code once it is clear where most of the spend is.

The product runs on a ready-made platform, not on our own code. Will the audit help?

Yes, in whatever the platform lets you control: model choice, context length, number of calls. If the platform does not let you control spend, we will say so.

Will answer quality suffer?

Every measure is tested on your examples before it goes in. If a model switch makes answers noticeably worse, we do not recommend it.

How much can we save?

It depends on the product. The answer is the calculation based on your data in the audit report, not a promise made in advance.

Where to start

Send us last month's provider bill and a brief description of which product features use models. That is where the audit starts.

Talk to us

Related: Audits for AI-built systems, AI solutions, LLM cost calculator on slavin.ai

SLAtech LTD