What AI automation costs.
From audit to production, pricing explained for agentic AI and automations. Where it's one-off, where ongoing, and where free. Development, production, model usage, and the oversight agentic systems need — each priced separately, so you can see what you're paying for.
What each stage costs, what moves the number, where the risk sits, and who carries it. You can stop after any stage and keep what you paid for.
AI automation pricing by stage
Is there anything here worth pursuing?
Stage
Free AI audit
Indicative cost
Free
Duration
45 minutes
- A call to understand how the work runs today
- A written note on where automation could apply
- A rough sense of the value at stake
- A clear yes or no on whether to go further
No cost, no commitment, and no obligation to continue.
We can tell within an hour whether a problem is worth solving. Charging for that conversation would only slow both of us down.
What an agentic AI build includes
An agentic system is not one piece of software. These are the parts, and you would need all of them whether we build it or you do.
- Automation platform — n8n, or equivalent orchestration
- Model access — the LLM itself, and the choice of which
- Solution architecture — how the parts connect and where state lives
- Application code — the logic that is not covered by the platform
- Database and storage — for state, history and audit trail
- Hosting — your cloud, ours, or your own hardware
- Evaluation harness — how you know the agent is behaving
- Monitoring and logging — what happened, when, and why
Where the money goes
Some of these you pay for directly. Some are inside the build price. Here is the split.
The components are commodity. Selecting them, connecting them, and making them work together as one system is the engineering, and that is what the build price covers. Keeping it working after launch is what a retainer covers.
What sits behind the number
Four things that determine what you pay, and what you get for it.
Two projects that sound identical in a meeting can differ several times over in price. These are the six things that account for most of it.
Systems to integrate
Each additional system means new auth, new data shapes, new failure modes. The breadth of what we connect is set out in our capabilities.
Autonomy level
A scripted workflow and a system that decides for itself are not the same job. Agents need evaluation harnesses, guardrails, tool permissions and a human review path, and that work is where the difference in price sits.
Data readiness
Clean, accessible data shortens everything. Messy data is usually the real project.
Compliance requirements
Audit trails, data residency and access control add engineering, not paperwork, and the bar differs by sector.
Environments to support
Dev, staging and production triples the surface we build and test against.
Who operates it after launch
Building for handover differs from building for a system we will run ourselves.
Why the ongoing cost is the one that surprises people
These are published industry findings, not claims about our engagements. They are why we price the operate stage as seriously as the build.
McKinsey estimates roughly $0.5 million in recurring annual cost for an off-the-shelf coding assistant, alongside roughly $0.5 million to $2.0 million in one-time integration cost.
McKinsey · 2023
McKinsey reports a rule of thumb of about $3 in change-management spend for every $1 spent developing a model.
McKinsey · 2024
Gartner projects that 55% of AI-optimized IaaS spending will support inference workloads in 2026, with inference operating continuously in production applications.
Gartner · 2025
Cost over time
Build cost is front-loaded. Operating cost is gradual.
Most implementation cost sits in the build. After launch, operating spend accumulates gradually as the system is monitored, maintained, and improved.
Cited for context. These are industry-wide figures and are not a projection of any outcome on your engagement.
Questions about pricing
Yes. It is a discovery call and a short written summary. It is not a project plan. The detailed assessment is the paid discovery engagement.
Engagements start at $10,000. A discovery runs from $10,000 and a build is priced from what discovery finds. The range is wide because the work is, and anyone quoting a figure before mapping your systems is guessing.
Because it is the work. Mapping workflows, auditing data and systems, and costing an opportunity backlog is a real deliverable you own regardless of whether you build with us.
There is no single rate. Agentic systems are usually more expensive than scripted workflows because they also require evaluation harnesses, guardrails, tool permissions and a human escalation path. We price the build after discovery; engagements start at $10,000. Where a scripted workflow solves the problem, we will say so and build that instead.
No. Model API usage, cloud infrastructure and third-party licenses are billed at cost. Wherever possible they run in your own accounts, so you see the meter directly and can turn it off without asking us. Where they run in ours, the underlying invoices come attached to yours. We make money building and running systems, not reselling tokens.
Most of our work is with organizations in the US, EU and UK. We invoice from our UAE entity and operate remote-first across those time zones.