What you are actually paying for
Quotes for AI automation vary by a factor of ten for what sounds like the same brief, and the reason is almost never the AI. The model call is the cheap part. What you are paying for is the work around it: understanding the process well enough to automate it, connecting the systems that hold the data, and handling the cases where the process does not go to plan.
It helps to split the cost in two and keep them separate in your head. The build is a one-off: scoping, integration, testing, deployment. The running cost is monthly and mostly consumption-based: AI model usage, any platform fees, hosting. A cheap build with an expensive per-run cost can easily be worse value than the reverse, and most quotes only show you one of them.
What it costs to build
These are the bands we quote in. They track scope rather than hours, and each one is a fixed price agreed before work starts.
One workflow, two or three systems, no bespoke UI.
- Form and inbox enquiries captured, de-duplicated and pushed into a CRM
- Invoices or forms read with AI and posted to your accounting system
- A scheduled report that pulls from two tools and writes the summary
Typical timeline: 1–2 weeks
A multi-step process across several systems, with a dashboard and human approval steps.
- End-to-end lead handling: capture, enrich, qualify, assign, follow up, report
- A customer-facing AI agent on your website or WhatsApp, connected to your CRM
- Document processing at volume with exception handling and an approval queue
Typical timeline: 4–6 weeks
Multiple agents and automations across departments, on infrastructure built to be relied on.
- Inbound voice, web chat and messaging agents sharing one knowledge base
- Automation across sales, delivery and finance with a single reporting layer
- Systems handling regulated data, with audit trails and access controls
Typical timeline: Phased, scoped per project
What moves a project from one band to the next is rarely ambition — it is the number of systems involved and how well they are documented. A process touching two tools with good APIs stays cheap no matter how clever the AI step is. A process touching a twelve-year-old system with no API is expensive even if the logic is trivial. We check this before quoting, which is why the scoping call exists.
What it costs to run
These are billed by the providers directly to you, on accounts in your name. The figures below are estimates to make the total legible — your actual bill depends on volume, and we work through it with you before launch.
| Cost | Typical range | What drives it |
|---|---|---|
| AI model usage | $20 – $400 / month | Billed per token. A text automation handling a few hundred items a month sits at the bottom; a busy customer-facing agent sits near the top. |
| Voice minutes & numbers | $0.05 – $0.15 / minute | Only for voice agents. Combines telephony and speech processing; a hundred five-minute calls a month lands around $25–$75. |
| Automation platform | $0 – $100 / month | Zero when the automation runs as code on your own server. Platform fees apply if you prefer a hosted tool like n8n Cloud or Make. |
| Cloud hosting | $10 – $100 / month | A small automation costs little to run. Cost rises with data volume and uptime requirements, not with the number of automations. |
| Maintenance | $0 – $500 / month | Optional. Automations break when the tools they touch change their APIs — a retainer covers monitoring and fixing that. |
For most businesses the honest total lands between $50 and $200 a month for a text-based automation, and $150 to $500 once voice is involved, because voice minutes are the one line that scales with real-world usage rather than with data.
Working out the payback
The only number that matters is whether it pays for itself, and you can work that out on the back of an envelope before you talk to anybody:
(hours the task takes each month × loaded hourly cost × 12)
vs
(build cost + 12 months of running cost)
A worked example. A team spends five hours a week on enquiry handling and data entry. At £25 an hour loaded, that is roughly £6,500 a year. A $4,000 build with $100 a month of running costs comes to about $5,200 in year one — so it pays back within the first year and the saving compounds after that.
Two adjustments worth making. First, automation rarely removes 100% of a task; plan on 70–85% and keep a person on the exceptions. Second, count the second-order effects if they are real for you — answering enquiries in one minute rather than one day usually moves conversion more than the labour saving is worth, but only count it if you will actually measure it.
When not to build custom
Custom automation is the wrong answer more often than agencies admit. Skip it when the process changes every few weeks and has not settled; when the volume is low enough that the time saved is under an hour a week; or when a $30-a-month product already does the job and you simply have not set it up.
Start with the single most repetitive task your team does, time it honestly for a month, and automate only that. One automation in the lowest band gives you a real payback figure from your own business, and that number — not a guide like this one — tells you whether the next one is worth commissioning.
Common questions
How much does AI automation cost in 2026?
For a single workflow automation, expect $1,000 to $3,000 to build. A full automated process across several systems with a dashboard runs $4,000 to $8,000, and automation across a whole operation starts around $9,000. On top of the build, running costs are typically $50 to $500 a month, depending on volume.
Is AI automation billed monthly or as a one-off?
Usually both, and it helps to separate them. The build is a one-off fixed price. The running costs — AI model usage, any platform fees and hosting — are monthly and billed by the providers directly to you, so they scale with how much you actually use.
How do I work out whether an automation pays for itself?
Multiply the hours the task takes each month by the loaded hourly cost of the person doing it, then compare with the build cost plus twelve months of running costs. Anything that saves five hours a week at $25 an hour saves about $5,400 a year, so a $4,000 build pays back inside a year.
Why is custom automation more expensive than a no-code tool?
A no-code subscription is cheaper until the workflow has exceptions, needs a system without a ready-made connector, or runs at a volume where per-task pricing bites. Custom costs more up front and then costs almost nothing per run. If a $30-a-month tool genuinely fits your process, use it.
What makes an automation project more expensive?
The number of systems it touches and how well they are documented, whether a person has to approve steps, how many exceptions the process has, and whether the data is regulated. A clean three-step process with good APIs is cheap; a ten-step process with a legacy system and compliance requirements is not.
Do AI automation costs go down over time?
Model prices per token have fallen steadily, so the same automation generally costs less to run each year. Build costs have not fallen the same way, because the work is understanding your process rather than writing the AI call.
What is the cheapest way to start with AI automation?
Pick the single most repetitive task your team does, time how long it takes in a month, and automate only that. A first automation in the $1,000 to $3,000 band gives you a real payback figure, and that number tells you whether the next one is worth doing.
Get a number for your process
A guide can only give you bands. Tell us what the task is and how often it happens, and we will tell you which band it falls in and what it would save — on a free call, with no obligation to build anything.
Related: AI automation services · AI agents & chatbots · AI voice agents · all packages and pricing