One-off build versus running cost
| Build (one-off) | Run (ongoing) | |
|---|---|---|
| Includes | Discovery, data work, development, integration, testing | Model or API usage, hosting, monitoring, maintenance |
| Driven by | Complexity and integrations | Usage volume and model choice |
| How to control | Scope a pilot first | Caching, smaller models, usage limits |
What drives AI cost
- 01Use-case complexity
- FAQ bot versus multi-step agent.
- 02Data readiness
- Cleaning, structuring and securing data.
- 03Integration
- Connecting to CRM, ERP and internal tools.
- 04Model choice
- Premium models cost more per use.
- 05Safety and review
- Guardrails and human-in-the-loop.
- 06Evaluation
- Test sets and monitoring.
A cost-conscious AI path
- 1
Workshop
Pick the highest-value use case.
- 2
Proof of value
Small build measured on real tasks.
- 3
Decide
Compare value to run cost.
- 4
Production
Build with controls.
- 5
Optimise
Reduce cost per outcome.
Ways to keep AI affordable
Frequently asked questions
How much does AI development cost?
It depends on use case, data, integration and usage. We recommend a time-boxed proof of value before committing to a production build.
What are the ongoing costs of AI?
Model or API usage, hosting, monitoring and periodic improvement.
Is open-source AI cheaper?
It can reduce usage fees but adds hosting and operations work; the right choice depends on volume and privacy needs.
Can we pilot with a small budget?
Yes. A narrow proof of value is the standard way to start.