AI cost optimisation · Fixed price · One week
Know what every AI answer costs, then pay less for it
AI features are billed by the token, and the bill grows with success. Our one-week AI cost audit attributes spend to features and customers, establishes a quality baseline, and delivers a ranked plan to cut cost without lowering answer quality. Fixed price, with an implementation quote at the end.
Six savings levers
Most AI bills contain the same sources of waste. The audit measures which ones apply to you and what each is worth.
Model routing
Classification, extraction and routing rarely need the largest model. Sending each request to the smallest model that meets the quality bar is usually the biggest single saving.
Prompt and context trimming
Long system prompts, repeated instructions and whole documents pasted into every request. Sending only what the model needs often cuts input tokens sharply.
Caching
Major providers charge far less for repeated prompt content that is cached. Most applications leave this money on the table.
Retrieval tuning
RAG systems that retrieve twenty chunks when three would do, inflating every request and lowering answer quality at the same time.
Agent discipline
Agents that loop, retry and re-plan without limits. Caps on steps, tokens and tool calls stop one bug from costing a month's budget.
Batching and scheduling
Work that does not need an instant answer, moved to discounted batch processing and off-peak schedules.
What the week includes
Measurement first, so every recommendation carries a dollar figure and a quality check.
- Token-level cost attribution by feature, model, customer and request type, from your real usage logs
- Cost per answer, per document and per active user, in plain numbers
- A quality baseline: an evaluation set built from real examples, so savings can be verified before release
- A ranked savings plan across routing, caching, prompt trimming, retrieval tuning, agent limits and batching, each with a dollar estimate
- A cost-per-outcome dashboard and budget alerts in your own accounts
- A 12-month spend forecast your finance team can use
- An implementation quote, which you can accept, give to your own engineers, or ignore
How it runs
One week from access to the savings plan.
Days 1 to 2: measure
Usage logs, provider invoices and request traces. We attribute every dollar to a feature and a customer.
AttributionDays 3 to 4: baseline and model
An evaluation set from real examples, then each savings lever tested against it to estimate the saving and the quality effect.
EvaluationDay 5: plan and brief
The ranked savings plan, the dashboard, the forecast and a 60-minute briefing with your engineering and finance leads.
BriefingHow we work
FinOps discipline applied to AI spend.
Evaluation before optimisation
No change is recommended without a measured effect on quality. The evaluation set stays with you for future changes.
Provider-neutral
We work with OpenAI, Anthropic, Bedrock, Azure OpenAI and Vertex AI, and keep your system able to switch. We take no commissions.
Your accounts
Dashboards and alerts are built in your own cloud and observability accounts, not ours.
Finance can read it
The forecast and the cost-per-outcome figures are written for the budget conversation, not only for engineers.
Tools and platforms we use most
Pricing
One fixed price for the audit. Implementation is quoted separately.
| Option | Price |
|---|---|
| AI cost auditOne week. Attribution, quality baseline, ranked savings plan, dashboard, forecast and briefing. | $4,900 |
| ImplementationPutting the savings in place: routing, caching, prompt changes, retrieval tuning, agent limits, monitoring. | fixed quote |
| Building cost-efficient LLM apps trainingOne day for your engineers, so the next feature is built efficiently from the start. | $790 per seat |
| Ongoing AI cost partnerMonthly review of spend, new-model evaluation and quarterly re-tuning. | from $3,500/mo |
All prices in AUD, excluding GST. Model and hosting costs remain billed by your provider. Registered startups under two years old receive 15% off their first engagement. If the likely saving is less than the audit costs, we will say so in the first conversation.
What this looks like in practice
A representative engagement of this kind.
Questions
How much do we need to be spending for this to be worth it?
The audit pays for itself quickly once model spend is in the low thousands of dollars a month or AI is part of your product's unit economics. Below that, the free first conversation will usually give you what you need.
Will cutting costs make the answers worse?
Not if it is done carefully. We build an evaluation set from real examples first and test every change against it, so you see the quality effect before anything ships.
Which providers do you work with?
All the major ones: OpenAI, Anthropic, Amazon Bedrock, Azure OpenAI and Google Vertex AI. The levers are the same; the implementation differs.
Do you need access to our code?
Read access to usage logs, invoices and prompts is enough for the audit. For implementation, we work in your repositories or hand the plan to your engineers.
What is the typical saving?
Across the engagements this offer is built on, 30 to 60% of model spend, with routing and caching the largest contributors. Your figure depends on how the system was built, which is why the audit measures before it recommends.
Is this the same as FinOps?
It applies FinOps principles, attribution, forecasting and accountability, to AI spend, which existing cloud FinOps tools largely cannot see because it is billed per token through API providers.
Often paired with
AI security review
A fixed-price, two-week review of how your business uses AI: shadow AI, data flows, vendor terms, prompt-injection testing, policy and Privacy Act readiness.
Read moreLLM agents · RAG · Document AIAI and automation
Assistants, agents and document pipelines built on your own data, measured for accuracy and handed over running in your cloud.
Read moreAWS · Terraform · FinOpsCloud and DevOps
AWS architecture, migration and cost reduction, with infrastructure as code, CI/CD and monitoring set up once and explained properly.
Read moreFind out what your AI costs per answer
A 30-minute conversation to scope the audit. Fixed price, one week, and a savings plan with a dollar figure on every line.