The short answer
The cost of AI automation depends on the workflow, the tools it connects, the state of your data, where a person has to approve, the channel it runs on, the volume it handles and the data-protection work around it. Building it is one cost. Running it, through AI usage, software and ongoing support, is another.
Ask three suppliers what AI automation costs and you get three numbers that cannot be compared, because each one priced a different system. The price follows the work, so the useful question is what the work involves.
Here are the seven things that move it, and the few that bring it down.
The seven cost drivers
Each of these changes the build, the running cost or both:
- The workflow: how many steps it has, how many decisions sit inside it and how often the exceptions come up.
- The tools it connects: whether each one has an API, a connection that lets your tools pass information to each other without anyone retyping it, and how much access it gives.
- Your data: where it lives and whether it is consistent. Gaps and duplicates in your records mean clean-up before anything can rely on them.
- Approvals and handovers: every point where a person signs off, or the system passes a case to someone, is a path to design and test.
- The channel: an AI receptionist on the phone needs call handling, a transfer to a person and testing on real calls. Email and text are simpler.
- Volume: AI models are paid for by use, so the running cost rises with the number of enquiries, documents or calls the system handles.
- Data protection: personal data means documented data flows, access rules and a data processing agreement, scoped before the build.
Messy data raises the cost most often. When your data is organised, ideally in one place, the AI reads it and gets on with the task.
When it is spread out, the AI has to hunt for it, sort it and ask what is right, wrong, old or new. Every one of those rounds costs time and usage. The number of tools adds cost too, but less than messy data or expectations do.
Expectations and examples
Many owners expect to write one prompt and get a finished result. It never works that way: even the best prompt needs a round of changes.
Context cuts the rounds. The more the AI knows about you, the project and the goal, the better the first result. AI is a layer on top of your data.
Examples matter as much. Without specific good and bad examples, the AI never learns what good looks like to you. Ask for a chart and it keeps coming back with the wrong axis, labels and data until you show it the chart you want.
Building it and running it are two costs
The build is a one-off: mapping the workflow, connecting the tools, testing with real cases and training your team.
Running it is monthly. It covers AI model usage, any software subscriptions the system needs, and someone watching it. Systems meet cases nobody mapped, tools change their connections, and staff leave.
That is why every live system we build runs under Managed AI Operations: monitoring, maintenance, support, usage and cost tracking, and a monthly report.
What brings the cost down
- Start with a knowledge file. Have the AI interview you about your business, how you talk and write and who you sell to, and save the answers as a file every later task draws on.
- Then pick one small workflow, such as an email organiser, or the one that repeats most often, and prove it before adding the next.
- Build on the tools you already pay for. Replacing a CRM to suit an automation is rarely worth it.
- Decide the approval points before the build, so they are designed in once.
- Bring the person who does the work. They know the exceptions, and each exception is a path the system has to handle.
Why we publish no prices
A price on a web page would have to assume the answer to all seven drivers. It would be too high for a simple job and too low for a complex one, and either way it would describe a system you are not buying.
I will not quote before I see your setup, because until then I do not know your tools, what needs connecting or the work involved. When someone tells me they want lead generation, I ask who they are targeting and what a lead costs them now.
Size changes the build too. A system for 20 people costs less than one for 50, which needs more access levels, more fallbacks and a stricter back end. Some need a proper database and hosting, and some have to run on the client's own machines.
We put a number on the problem first. The Free Business Check is a 30-minute call on what the problem costs you now, and it ends in your AI Recommendation Plan. If a build makes sense, the proposal states its price and how the return will be measured.
Frequently asked
Free resources
Ready to remove the work itself? Start with Workflow Automation.