AIHow to estimate and cut your LLM API costs
How to estimate and cut LLM API costs: token pricing, a worked monthly estimate, and a table of savings from caching, routing, batching and trimming.
Bilal Khursheed8 min read
AI service
Lower LLM bills without worse answers
Model routing, caching and prompt work that cut spend without hurting answers.
The problem
Where teams get stuck.
AI features are cheap to prototype and surprising to run at scale. Token spend rises with every user, every long prompt and every retry.
The usual cause is not one expensive call. It is the largest model used for every task, the same context sent again and again, and no view of which feature is spending what.
Cutting cost blindly damages quality. You need to know what each change does to the answers as well as the bill.
Our approach
What makes this work.
- 01
Measure the spend
We instrument your calls so you can see cost per feature, per request and per user.
- 02
Find the waste
We review prompts, context size, model choice, retries and retrieval to find where tokens are spent for no benefit.
- 03
Apply the changes
Smaller models for simple tasks, prompt and response caching, shorter prompts, tighter retrieval and batch processing where a delay is acceptable.
- 04
Check quality
Each change is run against an evaluation set, so a saving that harms answers is caught before release.
Delivery process
How we deliver it, step by step.
Every engagement follows the same seven stages, from requirements to post-launch support.
- 01
Discovery & requirements
Workshops with your stakeholders to define goals, users, scope and constraints, captured in clear written requirements.
- Goals and success measures
- Requirements and user stories
- Scope, estimate and roadmap
- 02
UX & UI design
User flows, wireframes and interface design, reviewed with you until the product feels right before it is built.
- User flows and wireframes
- Interface design
- Clickable prototype
- 03
Architecture & planning
Data model, system design, integrations and security are settled, and the work is broken into sprints.
- Architecture and data model
- Technical decisions on record
- Sprint plan
- 04
Development
Short iterations with code review on every change and a demo of working software each week.
- Weekly demo
- Shared board and written updates
- Reviewed, version-controlled code
- 05
QA & testing
Manual and automated testing across functionality, devices, APIs and performance, followed by user acceptance testing.
- Test plan and test cases
- Automated regression suite
- UAT support and sign-off
- 06
Deployment & launch
Release through a CI/CD pipeline onto monitored cloud infrastructure, with a go-live checklist and rollback plan.
- CI/CD pipeline
- Monitoring and alerting
- Go-live checklist
- 07
Hypercare & support
Close monitoring and priority fixes straight after launch, then ongoing support, maintenance and improvement.
- Hypercare period
- Support under agreed response times
- Improvement backlog
Deliverables
What you receive.
- Cost tracking per feature, request and user
- A findings report ranked by likely saving
- Model routing so each task uses the smallest model that does it well
- Prompt caching and response caching
- Shorter prompts and tighter retrieval context
- Batch processing for work that does not need an instant answer
- Budgets, limits and alerts on spend
- An evaluation set that confirms quality after each change
Technology
What we build it with.
- OpenAI
- Anthropic
- Open-source models
- Python
- Node.js
- Redis
- PostgreSQL
Related work
Where we have done this before.
AI / LegalTech
Client name withheld
Contract analysis with a multi-agent RAG pipeline
Turns 500 to 1,000 page legal contracts into a queryable knowledge base that answers with cited sources.
- Python
- FastAPI
- OpenAI
- Gemini
AI / Sales
Lead Rescue
AI call analysis with speech recognition and lead scoring
Transcribes customer calls, scores lead intent with GPT and NLP, and alerts sales teams to hot opportunities in real time.
- Node.js
- GPT
- LLaMA
- Automatic Speech Recognition
From the blog
Related guides.
What we have written about llm cost optimisation: cost, timelines and the trade-offs behind each decision.
- AI
FAQ
LLM Cost Optimisation: common questions.
How much can we save?
We will not quote a figure before looking at your usage. The saving depends on how the feature was built. The first step is measurement, and you see the findings before any change is made.
Does a cheaper model mean worse answers?
Not for every task. Classification, extraction and routing often work well on smaller models. The evaluation set shows where a smaller model holds up and where it does not.
What is prompt caching?
Providers can store the unchanging part of a prompt, such as instructions or a long document, and charge less when it is reused. It suits features that send the same context many times.
When should we start optimising?
After the feature works and you can see real usage. Optimising too early wastes effort on calls that may change. Basic cost tracking, though, is worth having from the first release.
Should we host our own model to save money?
Sometimes. Self-hosting can pay off at high, steady volume or when data must stay in your environment. It also brings infrastructure and maintenance work. We compare both options with your real numbers.
Discuss your LLM Cost Optimisation project with us.
Share a short brief and we will reply with next steps, usually within one working day.