"AI automation" is used to describe everything from a Zapier workflow to a research project. That vagueness is convenient for vendors and unhelpful for anyone trying to decide whether to spend money on it.
Here is the plain version, with no assumed knowledge. If you already know the concept and want prices, they are on our AI automation page.
Start with ordinary automation
Ordinary automation follows rules you write down. When a form is submitted, create a record. When an invoice is 30 days old, send a reminder. If the amount is over $5,000, route it to a manager.
This has existed for decades and it is excellent — reliable, cheap to run, easy to audit. If your process can be fully described as rules, you may not need AI at all, and a good vendor will tell you that.
What AI adds
AI handles the steps where you cannot write the rule down. Three kinds in particular:
Reading things that vary
An invoice from one supplier looks nothing like an invoice from another. A rule-based system needs a template per layout. An AI step reads a document it has never seen and pulls out the supplier, the total and the line items. This is the single most useful thing AI automation does for most businesses.
Making judgement calls
Which of nine departments does this email belong to? Is this complaint urgent? Which of these forty applicants match the brief? You could approximate these with rules, but the rule set grows until nobody can maintain it.
Producing text
Drafting a reply, summarising a long thread, generating a first-pass proposal from your own past documents. Usually as a draft a person approves, not as something sent unreviewed.
What it is not
- It is not a robot doing someone's job. It handles specific repetitive steps within a process, not a role.
- It is not always right. Well-built systems escalate when unsure instead of guessing. Anyone promising perfect accuracy is overselling.
- It is not a replacement for fixing a broken process. Automating a bad process gives you a faster bad process.
- It is not one product you buy. It is a set of techniques applied to your specific workflows.
What it typically automates
- Reading invoices, receipts, applications and forms into your systems
- Routing enquiries to the right person or department
- Moving data between systems that do not talk to each other
- Chasing people for documents, approvals and timesheets
- Answering repetitive customer questions
- Answering the phone and booking appointments
- Generating recurring reports
- Flagging exceptions a person should look at
Notice how ordinary that list is. The valuable applications are usually boring, high-volume and specific — not the impressive demos.
What it costs
Two costs, and vendors sometimes only mention the first.
Build: one-time. Ours is $499 for one process end to end and $1,499 for a department. US agencies typically quote $5,000–$25,000 for comparable scope, and enterprise consultancies considerably more.
Running: ongoing, and it scales with volume. AI model usage plus any platform subscription — typically $20–$300 a month for a small business. This should be billed to your own accounts so you can see it. Full breakdown in our cost guide.
How to tell whether it is worth it for you
Pick the task your team complains about most. Estimate hours a week, people doing it, and their loaded hourly cost — roughly 1.3× the wage. Multiply out to a year.
If that number comfortably exceeds the build cost plus a year of running costs, it is probably worth doing. If it does not, find a bigger task. The ROI calculator does this properly and will tell you when the answer is no.
Frequently asked questions
Do I need to be technical to use this?
No. You need to be able to describe how the work happens now. The technical part is the vendor's job, and a good one will document what they build in plain English so your team can maintain it.
Will it replace my staff?
In practice it takes the parts of jobs nobody wanted. The businesses that get most from it are usually the ones about to hire an admin person who would rather not, or teams so buried in data entry that the real work is slipping.
How long does it take?
About two weeks for a single process, three to five for a department. Most of the first week is mapping the process properly — which is also where the useful discoveries happen.
What if my process changes?
Well-built workflows are made to be edited. Documented properly, your team can handle small changes; structural ones need the builder. Ask any vendor how they document before you sign.
Not sure whether your process is a good candidate? The free 30-minute audit maps where your hours go and ranks what is worth automating — including the honest answer if nothing is. See how it works or message us on WhatsApp.
