"How much does automation cost" is a fair question with an unsatisfying answer: it depends on scope, and scope varies more than almost any other kind of software work. A one-line answer would be dishonest. A useful answer breaks the cost into parts you can estimate against your own situation.
There are three costs, not one: the build, the running cost, and maintenance. Most quotes cover the first and stay quiet about the other two.
1. Build cost, by project size
The single biggest driver is how many systems the automation touches and how much of the process resists a fixed rule. Rough ranges for a first engagement:
- Configuration change in software you already own — $0 to $1,500. Sometimes the answer is a setting, a native automation, or a template in a tool you already pay for. We will tell you when this is the case, and it is more often than vendors admit.
- Single connected workflow — $4,000 to $12,000. One process, two or three systems, error handling, logging, and an exception path. Lead intake to CRM with follow-up, invoice capture and routing, an onboarding sequence.
- Multi-system automation with logic — $12,000 to $35,000. Several processes that share data, conditional routing, reconciliation between systems, a light interface for the exceptions. This is where most operational programmes sit.
- Custom application with automation inside it — $35,000 and up. When the process is a genuine differentiator and no product fits, the automation is a feature of a build rather than the whole job.
These are ranges for scoped, fixed-scope work with a written deliverable — not open-ended hourly arrangements, which we do not run as a default.
2. Running cost
Once it is live, an automation costs money to operate. Three components:
- Platform subscriptions. If it runs on a dedicated automation tool, that is typically $20 to $200 a month depending on volume and features. If it runs on infrastructure you already have, this can be near zero.
- Usage. Per-transaction or per-run fees on some platforms, and API or model usage where the workflow calls an AI service. For most small-business volumes this is tens to low hundreds of dollars a month. High-volume document processing can be more.
- Third-party API tiers. Some systems charge for API access above a threshold. Worth checking before you build a dependency on one.
A realistic figure for a single workflow at small-business volume is $30 to $250 a month, all in.
3. Maintenance — the cost people forget
Automations run unattended for the overwhelming majority of their life, and things change around them. A vendor deprecates an API endpoint. A form gets a new field. Volume grows past what the original design assumed. An edge case appears that nobody anticipated.
Left alone, an automation has a quiet expiry date. Budget for one of two things: internal ownership — someone whose job includes watching vendor changelogs and fixing breakages — or a managed engagement that does it, typically scoped to the size of your system footprint rather than sold as a fixed tier.
What moves the number
Within any size band, four things push cost up:
- Number of systems. Each integration point is design, testing, and a failure mode. Two systems is straightforward; five is a different project.
- Data readiness. If the information a process needs lives in someone's head or an email thread, the first piece of work is building the record the automation will read. Legitimate work, but it belongs in the estimate.
- Review requirements. A workflow that can run fully automatically is cheaper than one where a person approves each consequential step. High-consequence processes need the approval step, and that is correct — it just costs more to build.
- Exception rate. The happy path is cheap. If the process deviates often, the cost sits in handling the deviations, and honest scoping means measuring how often "it depends" actually comes up.
Working out whether it pays back
The calculation is simple and worth doing before you commission anything. Take the hours the process consumes per month, multiply by a loaded hourly cost for the people doing it, and annualise. Compare that to the build cost plus a year of running and maintenance.
A process consuming 20 hours a month at a $40 loaded rate is $9,600 a year. A $10,000 build that removes most of it pays back inside about fourteen months and keeps paying after. A process consuming two hours a month does not clear the same bar, and a better checklist is the right answer.
Where it is not worth it
Be willing to hear no. Automation does not pay back on low-volume processes, on anything where an error is expensive and cannot be reviewed, on a process that is about to be replaced anyway, or where a $20-a-month software setting does the same job. A good scoping conversation ends in "not this one" more often than the industry likes to admit.
If you have a process in mind, the honest cost answer comes from scoping it — the systems, the data, the exceptions. Describe what it involves and we will give you a range and tell you whether the payback is there. More on what AI adds to this on the AI automation page.