Blog · 2026-11-18
What an AI automation project actually costs
The question behind most first calls is the one nobody opens with: what is this going to cost me? Here is the honest structure of it, so you can ballpark your own project before you talk to anyone.
The three costs, separated
Every automation project has a build cost, a running cost, and a maintenance cost. Vendors like to blur them. Keep them apart and the decision gets easy.
The build is the one-time project: mapping the process, building the workflow or app, testing it against real cases, and documenting it. It scales with how many systems are involved and how messy the rules are, not with how impressive the result looks. A single workflow that moves data between two tools is days of work. A custom app with sign-in, a database, and reporting is weeks.
The running cost is what the system consumes every month after I hand it over. For the systems I build, the pattern is consistent:
- An automation platform plan, usually $20-60 a month in your own account
- Hosting and a database for a custom app, usually $5-25 a month
- AI usage for agent work, usually $10-50 a month on your own API key, with a hard spending cap
A typical finished system lands between $25 and $100 a month to run. You pay those bills directly because everything runs in your accounts, which also means nothing dies if you and I part ways.
Maintenance is the wildcard. Software vendors change their systems, and a connection that worked in March can need a patch in June. Budget a few hours a quarter, not a retainer. Every system I ship logs its runs and alerts on failures, so problems announce themselves instead of hiding for a month.
What decides whether it pays
The arithmetic is short: hours saved per month, times what those hours cost you, against the build price plus running costs. A routine that eats two hours a week of a $40-an-hour employee's time costs you about $350 a month. Automating it usually pays for itself well inside the first year, then keeps paying.
The projects that do not pencil out share a profile: rare tasks, fuzzy rules, or a process the owner cannot describe without saying "it depends" five times. Part of my job on an audit call is to tell you when your project is one of those. An automation that needs constant supervision is a second employee.
Why the range is wide
Two businesses can ask for "the same" automation and get quotes an order of magnitude apart. The spread comes from questions like these:
- How many systems does the data touch, and do they have real ways to connect?
- Does a person need to review anything before it goes out the door?
- What happens when the input is wrong: skip it, flag it, or fix it?
- Does it need its own screen, or can it live inside tools you already use?
The last one matters most. The moment a project needs its own interface, you are buying a custom app rather than a workflow, and the build grows accordingly. Plenty of projects that start as "we need an app" ship as a workflow plus a shared spreadsheet, at a fraction of the cost.
How to get a real number
Write down the routine you want gone: what triggers it, which tools it touches, what a finished instance looks like. That one paragraph is enough for a usable estimate. I do this on a free audit call for businesses around Torrance and the South Bay and beyond, and you leave with a ranked list of what is worth automating, whether or not we work together.