How much does business process automation cost for a small business?

Where the money actually goes

Automation pricing looks opaque from the outside because most firms will not publish a number. Here is the structure underneath almost every engagement, whoever you buy it from.

The audit is the part people try to skip, and skipping it is the most common reason an automation project costs more than it should. Building against a process nobody has measured produces an automation that handles the common case and quietly hands the exceptions back to a person — which is how you end up paying for something that saves half of what was promised.

The five things that move the number

If you can answer these, you can place yourself in the range fairly accurately before speaking to anyone.

1. How many systems are involved

This dominates everything else. Two systems is the straightforward case. Each additional system adds authentication, field mapping, its own failure modes and another vendor whose API can change without warning. Going from three systems to six roughly doubles a build — it does not add fifty per cent.

2. Whether the APIs are any good

A modern REST API with webhooks and decent documentation is cheap to work with. A SOAP endpoint, an undocumented internal API, or a system with no API at all forces a workaround — a scheduled file export, a database connection, or browser automation as a last resort. Expect that to add $5,000 to $15,000 depending on which rung of that ladder you land on.

3. How many exceptions the process has

"Except when it is a rush order." "Except for our three biggest accounts." "Except in December." Every exception is a branch that has to be designed, built, tested and documented. On a mature process, exception logic is frequently the majority of the engineering — and it is invisible to anyone estimating from a flowchart.

4. One-way or two-way

Pushing data in one direction is straightforward. Keeping two systems genuinely in sync means conflict rules for when both sides change, loop prevention so each write does not trigger the other, and reconciliation for when they drift. Bidirectional sync is roughly double the work for the same field count.

5. The state of your data

Duplicate customer records, three spellings of the same supplier, inconsistent date formats. If a cleanup is needed before automation can work reliably, it should be quoted separately and visibly rather than absorbed into a vague number. Any firm that does not raise this has not looked at your data.

Ongoing costs after the build

Two separate things, and it is worth keeping them separate in your head.

Platform subscriptions are what you pay the software vendor to run the automation. For a typical small-business stack this is $30 to $150 a month. Make is priced per operation, so it rises with volume. Self-hosted n8n is a fixed infrastructure cost of roughly $10 to $40 a month regardless of volume. AI API usage for something like an email triage agent is often under $50 a month at a few hundred messages a day.

Support is optional and is what you pay a person to keep it alive. It matters more than people expect, because automations do not fail loudly — they fail silently, and you find out from a customer. APIs get versioned, credentials expire, vendors deprecate endpoints. In a stack of a dozen automations, something needs attention every few months.

When it costs less than this

Three situations genuinely come in under the ranges above.

• You build it yourself. If your workflow is a trigger and two actions between two mainstream tools, build it in Zapier or Make in an afternoon. Paying someone $8,000 for that would be poor value and any honest firm will tell you so.
• An off-the-shelf product already does it. Sometimes the answer is a $40-a-month SaaS tool, not a custom build. Invoice capture, e-signature routing and appointment scheduling are all well-served by existing products.
• The process should be deleted, not automated. Reports nobody reads, approvals that never get rejected, duplicate data entry caused by two teams not talking. Automating these makes waste faster and permanent.

When it costs a lot more

Beyond $35,000 you are usually looking at one of: a legacy system with no integration path that needs a bespoke bridge; a regulated environment where audit trails and validation add real engineering; a data migration bundled into the project; or genuine enterprise scale where throughput changes the architecture. None of these are hidden — a competent firm will flag them during scoping, and if the number moves after work has started, that is an estimating failure rather than a fact of the domain.

Working out whether it pays for itself

The arithmetic is simple, and worth doing before you talk to anyone.

Take the hours the process consumes each month. Multiply by a loaded hourly cost for whoever does it — for a US administrator, $35 to $55 an hour is a reasonable range once you include employment costs. That is your monthly saving. Divide the build cost by it to get payback in months.

A process costing 15 hours a month at $45 an hour is $675 a month. A $12,000 build pays back in about eighteen months on labour alone.

Eighteen months sounds slow, and on labour alone it often is. But labour is usually the smaller half. The larger half is the things that are harder to put on an invoice: invoices raised the same day instead of a week later, so cash arrives sooner. Leads answered in sixty seconds instead of four hours. Errors that stop happening. Work that continues when the person who normally does it is on holiday.

If the labour saving alone justifies it, the decision is easy. If it does not, the honest question is whether those second-order effects matter enough in your business — and that is a judgement call, not a calculation.