How much does AI automation cost? Prices and running costs

There is no single price for AI automation, because “automation” covers everything from a form that posts to a spreadsheet to an agent that reads and answers customer email. What you can do is split the cost into the same few lines every project has, and price each one from a list or a quote.

The cost at a glance

Build type	One-time build (SBD, from)	Monthly running cost
Workflow automation, including AI agents for a defined job	From $1,500	Platform plan, model usage and upkeep
Internal tool your team works in	From $3,000	Hosting, any model usage and upkeep
Multi-module system: several tools on one data model, portals	From $5,000	Hosting, model usage, platform plans and support
Micro-SaaS: a product your customers log into and pay for	From $10,000	Hosting, model usage, billing and support

Build figures are SBD’s published starting prices by build type, given as one reference point; the fixed price for a given project is set after the audit, and other suppliers will price differently. Platform and model figures come from the vendor price lists in the sources.

What the platforms charge. Zapier’s Pro plan starts at $19.99 a month billed annually for 750 tasks, with a free plan of 100 tasks. Make offers a free plan with 1,000 credits a month, with paid plans priced by credit volume. n8n’s cloud Starter plan is 20 euros a month billed annually for 2,500 workflow executions, and Pro is 50 euros for 10,000; n8n can also be self-hosted, which moves the cost to your own server and upkeep.

What the models charge. Prices are per million tokens. At the time of writing, Anthropic lists Claude Haiku 5.5 from $0.10 input and $0.50 output, Sonnet 5.5 at $2 and $10, and Opus 5.5 at $4 and $20. OpenAI’s list runs from $0.05 input on its smallest model to $30 input and $180 output on its pro models. The spread is wide, so model choice matters more than almost any other running cost.

Upkeep. Systems break when a connected tool changes its API, a field is renamed, or the data drifts. Budget for someone to watch and fix them. As one example, SBD’s care plans start at $50, $150 and $300 a month by tier.

What drives the price

1. Number of systems. Each connection adds authentication, error handling and testing.
2. Data quality. Clean, consistent data is quick to automate. Duplicates and free text are not.
3. Older systems without an API. Screen scraping, exports or a database bridge cost more to build and to keep running.
4. How much judgement is involved. Rules are cheap. Reading, classifying and drafting need a model, evaluation and a review step.
5. Risk of being wrong. Anything that touches money, customers or personal data needs approvals, logging and access controls.
6. Volume. Usage pricing on platforms and models grows with every item processed.
7. Who builds it. Your own time, a freelancer or an agency, each with different rates and different things included.

Hidden costs

• Your team’s time. Explaining the process, testing, and changing how people work is real cost, even when no invoice arrives.
• Retries and long inputs. Failed runs repeat tasks, and long documents use far more tokens than the average.
• Plan jumps. Crossing a task or credit limit can move you up a tier mid-year. Zapier’s monthly billing costs about 50% more than annual on the same tier.
• Extra seats and licences. A new tool in the workflow may need paid users.
• Maintenance. Nobody owning failures is the most common reason automations quietly stop.
• Switching cost. If the accounts and code sit with your supplier, changing supplier means rebuilding.

1. Define the job and the volume

Name one process, the systems it touches and how many items it handles a month: invoices, leads, tickets, orders. Take the count from a system report. Every later number multiplies by it.

2. Price the platform

Map each step to how the platform counts usage: tasks on Zapier, credits on Make, executions on n8n. Multiply by monthly volume, add at least 30% headroom, and pick the tier that covers it. Compare annual and monthly billing.

3. Price the model usage

Run ten real examples through the model and note input and output tokens. Then:

Monthly model cost = items a month x (input tokens x input price + output tokens x output price) / 1,000,000

For example, 10,000 items at 1,000 input and 500 output tokens each is 10 million input and 5 million output tokens. On Claude Sonnet 5.5 at list price that is $20 plus $50, about $70 a month. On Claude Haiku 5.5 it is about $3.50. Test the cheaper model first; it is often good enough for classification and extraction.

4. Price the build

Ask for a fixed price against a written scope, with exclusions listed. Check that testing, documentation, handover and a period of fixes after launch are included. If you build it yourself, count your hours at what they are worth.

5. Add upkeep and contingency

Add a monthly line for monitoring and small changes, and name who is called when it fails. Add 10% contingency to one-time costs.

6. Compare against the cost of doing nothing

Set the first year’s total against the hours, errors and delays the automation removes. The free ROI calculator does this with any pricing model, and our guide to building an automation business case covers the method in full.

Build or buy

If an off-the-shelf product already does the job for a monthly fee, it is usually cheaper than a custom build, at least to start. Custom work earns its cost when the process is specific to you, spans several systems, or the subscription grows with headcount faster than the value does. The free build vs buy tool walks through the trade-off.

When it is not worth it

• Low volume. A task done a few times a month rarely repays a build.
• A process that changes every month. Fix the process first, then automate it.
• No owner. If nobody will look after it, it will fail without anyone noticing.
• Cheaper fixes exist. A template, a form or a setting in a tool you already pay for may solve most of it.

If the numbers do not work, that is a useful answer. Spend the budget where they do.