What an automation or AI chatbot costs per month after it is built

The short answer

The quote you accept covers the build. After handover, a working automation or chatbot keeps costing money every month, in up to seven lines: the platform, hosting, AI usage, paid connected services, the machine or server it runs on, fixes, and your own time watching it. Most buyers see only the first.

None of these has to be large. What causes trouble is finding them on the first invoice, or discovering one when it stops the job. So ask for a monthly running estimate at your volume, line by line, billed to your accounts, in the same conversation as the build quote. The workflow automation brief and the chatbot brief both have room for it.

Why the running cost is hard to see before you buy

A demo runs a few times. A month of real use runs hundreds or thousands of times. Costs that are invisible in a demo scale with those runs: per-task platform fees, AI tokens, calls to other services.

The other reason is that some running costs are not money on an invoice. A usage limit that stops your job halfway, or a laptop that sleeps through the night’s schedule, costs you in missed work. I run a set of scheduled jobs myself, and the running costs that surprised me were of that kind. In money, my own set of personal automations costs me about $150 a month; that is my setup, not a typical price.

The seven lines of a monthly running cost

1. The automation platform. Hosted platforms charge a subscription tied to how many tasks or operations run. Zapier is the easiest to hand to a non-technical team and the most expensive per task at volume; Make is cheaper per operation; self-hosted n8n costs nothing per run. The trade-offs are laid out in n8n, Make, Zapier or code.

2. Hosting or a server. Self-hosting moves the cost from per-run fees to a server, and to someone keeping it updated. A chatbot or app needs hosting too. Some things cost almost nothing: this site is a static site on a hosting plan with a free tier, and the domain renews once a year.

3. AI usage. Every AI step is billed by the token: text the model reads and text it writes, with writing priced higher on the price list I checked. It grows with volume, document size and answer length, and it should run on your own provider key; the API keys explainer covers why.

4. Paid connected services. Email senders, messaging channels, data sources, anything with its own plan. Each also has limits, which leads to the next point.

5. Limits, which cost you in stopped work. Every service has a ceiling: a monthly spend limit, a number of requests per hour, a number of posts per day. Hitting one does not raise the bill; it stops the job.

6. The machine it runs on. A job on a hosted platform runs on someone else’s machine. A job on your office computer or a laptop runs only when that machine is awake.

7. Fixes and your time. Tokens expire, apps change, limits move. Budget a free fix period and then a rate, as set out in putting maintenance in the contract. Add a few minutes a week for you to read alerts.

What keeps costing after handover

The build quote covers one line. The other six arrive every month, or stop the job.

Paid once

The build

Billed every month

Platform→Hosting or server→AI usage→Paid services→Fixes and your time

Costs you in stopped work

A limit is reached→The machine is asleep→Named and alerted before it happens ✓

Three running costs from my own jobs

A spend limit stopped a nightly job. An overnight job of mine that writes social posts with an AI model hit the monthly spending limit on its account. It produced four posts, then ended with an error. A second job that would have drawn on the same account stayed switched off, so that two jobs would not split one limit.

A request limit shaped the schedule. My publishing job posts through a platform’s API. One ten-image post takes about 80 calls; a single video takes a few. The platform allowed 50 posts in 24 hours, which was enough, but when several slots were bunched into the same hours the calls hit the platform’s request limit. The fix cost nothing but a decision: spread the slots across the day and evening.

A sleeping laptop delayed the schedule. The same jobs ran on a laptop that drops into a low-power standby when the screen turns off. One evening slot ran 38 minutes late, a late-night slot did not run at all, and a morning slot ran three hours late (the full field note). Keeping the machine awake on mains power fixed it. The running cost of “it runs on my laptop” is that the laptop has to stay on.

None of these would appear in a build quote. All of them are questions a freelancer can answer before you order.

The running-cost sheet to send

Paste this into your first message and ask the seller to fill the right-hand column at your volume:

Line	What to ask	Seller’s estimate
Platform	Which plan, on my account, at [N] runs a day?
Hosting / server	Where does it run, on whose account, and who keeps it up?
AI usage	Cost of one run or one conversation in your tests, times my volume
Paid services	Which other services need a paid plan?
Limits	Which limit will we hit first, and what happens then?
Machine	Does anything depend on a computer being awake?
Fixes	Free period after delivery, then what rate?

Who can help keep the running cost down

Some sellers make the running cost part of their pitch. On the workflow automation page, Get Efficient builds custom n8n workflows on a platform you own, self-hosted or n8n Cloud, and Ramon C argues for Make when you do not want to run a server. Both are useful conversations to have before you choose where the automation lives.

For a chatbot, the acceptance list on the chatbot job page asks you to know the cost per 100 conversations. That number, times your monthly volume, is most of the running cost.

Where it goes wrong anyway

• Volume grows. The estimate was right for 50 runs a day, and now there are 500. Ask what the bill looks like at ten times your volume.
• A free tier ends. Some services are free up to a point. Ask which ones, and what the next plan costs.
• The seller’s account was the cheap part. If anything runs on the seller’s account, their costs and limits are hidden from you until it stops. What moves the price of a freelance automation makes the same point about quotes.

Wrap-up

Treat the monthly running cost as part of the quote. Ask for all seven lines at your volume, on your accounts, with the AI usage measured in testing and the nearest limit named. Then the first invoice after handover holds no surprises, and neither does the first night a limit is reached.