AI Income & Cash Flow

How Much to Charge for AI Automation

Charge $1,500 for one AI workflow. See the $250 hourly rate, the monthly tool bill, and the US tax that still applies.

One person pouring coffee while another rinses a cup in a warehouse kitchen
Key Takeaways
  • Aenfinite publishes a single-workflow sprint at $1,500 fixed: one pipeline, up to four apps, error handling, and a diagram, in 7 to 10 days.
  • XRAY publishes workflow help at $250 an hour, with a four-hour opening block at $1,000, and a dedicated team month at $15,000.
  • n8n Cloud Starter is 20 euros a month billed annually and includes 2,500 executions. A run every five minutes is about 8,600 to 8,900 a month.
  • Claude Haiku 5.5 is $0.10 per million input tokens and $0.50 per million output tokens when the prompt is at most 100,000 tokens.
  • US self-employment tax is 15.3% once net earnings are $400 or more. Publication 54 says the rules are generally the same if you live abroad.
  • For payments made in 2026, a US client files Form 1099-NEC at $2,000 of nonemployee services. Before 2026 the line was $600.

Disclosure: this article contains affiliate links. If you open an account through one of them, Cashflow Abroad may earn a referral commission at no extra cost to you.

Charge $1,500 for one AI automation workflow. That is the fixed price Aenfinite publishes for a Single Workflow Sprint: one production pipeline, error handling, up to four connected apps, and a written diagram, delivered in 7 to 10 days. XRAY publishes the hourly alternative at $250, opening with a four-hour block at $1,000, and a staffed month at $15,000.

The software bill stays in the buyer's accounts, so $1,500 is the build, not a promise to fund their logins. If they would rather pay once and own the job, Michael Heredia lists Workflow Automation at $3,500 for one named repeating task, with model and host charges billed to them and no required monthly fee to him. After you move, US self-employment tax can still reach that fee, the same way it reaches other AI income from a US buyer.

How much should you charge for one workflow?

Use the published sprint when the buyer can name one pipeline, and leave the $15,000 month alone. Aenfinite's pricing page puts n8n workflow work in a $1,500 to $5,000 band, then prices that sprint at $1,500 and an overhaul of four to eight workflows at $3,500 to $6,000 across two to three weeks. $1,500 divided by XRAY's $250 hour is six hours. If the named job will not fit in that window, sell the hourly block instead of the fixed price.

Published offer Price What they say is included
Aenfinite Single Workflow Sprint $1,500 fixed, 7–10 days One pipeline, error handling, up to four apps, documentation
Aenfinite Operations Overhaul $3,500–$6,000, 2–3 weeks Four to eight workflows, a self-hosted n8n install, AI summaries, CRM sync
XRAY expert sessions $250 an hour; first block $1,000 A live build, a recording, and a short recap. Extra time is billed in 15-minute increments
XRAY monthly $15,000 a month A dedicated team from discovery through maintenance
Michael Heredia Workflow Automation $3,500 once One named job in tools the buyer already pays for

Which price matches the job

Quote $1,500 for one pipeline and four apps or fewer. Quote $3,500 to $6,000 when they want several workflows and a server they host. Quote $250 an hour, prepaid as a $1,000 four-hour block, when the process is still fuzzy. Unused hours on that block stay valid for six months, and a session can be cancelled free up to 24 hours ahead. Do not quote $15,000 a month unless you are staffing a team.

Who pays, and what you are selling

The buyer is a US team small enough to ask about a startup discount and busy enough to pay for one repeating job. Make says teams under 20 employees can check a Start-up Plan for 50% off Business. n8n says the same headcount can check a Startup plan for 50% off Business, and it lists Business at 667 euros a month when billed annually. You do not need either discount to send a $1,500 quote. You need a company that already has a form, a sheet, or a CRM, and one job a person still does by hand every week.

Name the job in the proposal. A lead form that should create a record and a draft a human still sends. An invoice list that should produce a chase note. A Friday numbers pull. Those are the kinds of jobs on Michael Heredia's workflow offer. Your tighter version stops at four apps, which is the cap on Aenfinite's $1,500 sprint, and it stops when a person has to approve the message that leaves the building.

Post the offer before you build a second one

Write one sentence a buyer can accept: one pipeline, up to four apps, error alerts, a diagram, half to start and half at sign-off, built in their account. To see whether a US company writes back, post that offer as a free listing on Brixaz and wait. A second workflow is a new quote, which is also how the $3,500 workflow offer is structured: another job is priced when they want it.

Do not price it like a phone line

The same Aenfinite page prices a voice receptionist at $1,500 to set up, then typically $0.08 to $0.12 per conversation minute, billed on the buyer's own voice accounts. That product is sold on talk time. A back-office workflow is sold on the build. If they want the phone answered, the appointment-setter breakdown is the closer match. Folding minutes into this quote mixes two bills and makes the $1,500 look like it includes a call center.

What the client pays each month

Put production in the client's n8n Cloud and on the client's model key. n8n Cloud Starter is 20 euros a month when billed annually. The n8n help center, updated 22 October 2026, says Cloud Starter includes 2,500 executions a month. One execution is one full run, however many steps sit inside it. Your delivery hours are the scarce input. The hosting line is not.

Piece Published price or rate Limit that changes the quote Who pays
n8n Cloud Starter 20 euros a month, billed annually 2,500 executions a month, 5 at once, 5-minute maximum run Client
n8n Cloud trial No credit card for the cloud trial 1,000 executions, 5 at once, 180-second timeout Whoever opens the trial
n8n Cloud Pro 50 euros a month, billed annually Help center lists Pro-1 at 10,000 executions a month Client, if the webhook volume needs it
Claude Haiku 5.5 $0.10 per million input tokens and $0.50 per million output tokens when the prompt is at most 100,000 tokens Over 100,000 input tokens the rates are $0.50 and $2.50 Client's API key
Make Credits; plan dollar cards were not readable on the pricing page Most actions are 1 credit. Routers and error handlers are 0. Free is 1,000 credits a month Client, if you build there instead
Starter math

Say one form run sends 2,000 input tokens and receives 400 output tokens on Haiku 5.5, under the 100,000-token prompt price. Input is 2,000 / 1,000,000 × $0.10 = $0.0002. Output is 400 / 1,000,000 × $0.50 = $0.0002. One run is $0.0004. Forty runs in a month are $0.016. Those 40 runs are 40 of Starter's 2,500 executions. The client's month is 20 euros plus $0.016 of tokens. Your $1,500 is the build, collected 50% up front, and it is not a share of that 20 euros.

A single clay pipe filling a jar while a bundle of pipes waits

Make is the other builder, and its pricing FAQ is enough to scope a job even though the dollar cards did not render on the page. As of August 27, Make bills in credits. Most actions consume one credit. Routers and error handlers consume none. The Free plan is 1,000 credits a month. When credits run out, scenarios stop, webhooks wait only up to the queue on the plan, and a poll later looks back to the last successful run. A webhook avoids spending a unit on every empty check.

Say forty forms arrive in a month

Say a 12-person US software firm, under the 20-employee line both startup plans mention, gets about 40 demo requests a month. Nobody on the team has priced a startup discount yet, and you should not promise they will qualify. The job you sell is one pipeline: the form lands, Haiku 5.5 marks fit or not and drafts a reply, a row is written to the sheet or CRM they already pay for, and a person sends the email. That is four touches if you count the alert channel, which is the sprint's published ceiling.

Step What you set up Why it stays inside the quote
1. Trigger A webhook from their form, in their n8n account n8n's pricing FAQ says a run every five minutes is about 8,600 to 8,900 executions a month, past Starter's 2,500
2. Draft Haiku 5.5 on their key, prompt kept at or under 100,000 tokens That tier is $0.10 and $0.50 per million tokens. The longer-prompt tier is five times the input rate
3. Record One write to the sheet or CRM they already use The model drafts. A person sends. Refunds, pricing, and legal replies stay with the human
4. Alert An error path to the channel they already watch On a paid n8n plan, editor test runs do not count against the monthly quota. Webhook runs do
5. Time cap Confirm a live run finishes Starter cuts a run off at 5 minutes. The cloud trial cuts one off at 180 seconds
Hands coiling a plain cable beside a wall jack in a hallway

Those forty runs use 40 of the 2,500 Starter executions and $0.016 of tokens. That is the published rate times a volume you replace with their real form count before you promise the 20-euro plan is enough. Hundreds of submissions a day belong on Pro, listed at 50 euros a month billed annually, where the help center's Pro-1 cap is 10,000 executions. Haiku 5.5's lower rate stops at 100,000 tokens. Above that, input is $0.50 per million and output is $2.50. A repeated instruction can be read from the prompt cache at $0.01 per million on the shorter tier. Haiku 4.5, on the same list, is $1 and $5 per million input and output tokens.

What breaks the quote

The miss that eats the month is a schedule, not the model. At the execution limit, n8n does not queue the work. The trigger fires, the run fails immediately with the message "Execution limit reached. Consider upgrading your plan," and it fails before any node runs. Those failed runs stay failed. They are not replayed when someone upgrades, and failed executions do not count against the quota, so the log fills with misses that did not even spend a unit. The quota resets at the start of the calendar month, not on the day the card is charged. A one-time reset exists, once per customer, for Trial, Starter, and Pro-1, by emailing help@n8n.io. Build the first job as a webhook so you never need that email.

Hosted n8n stores data in the EU, on servers in Frankfurt. If the form collects names, email addresses, or notes about a deal, say that before you connect the webhook. Self-hosted n8n, including the Community Edition n8n offers on GitHub, keeps the data wherever the buyer places the server. Aenfinite's overhaul is the published offer that includes deploying that self-hosted instance, at $3,500 to $6,000. The $1,500 sprint does not say it includes a VPS. Selling a server inside the sprint price is how a six-hour job becomes a hosting contract you did not price.

The tax on a $1,500 invoice

Publication 54 says a self-employed US citizen follows the same self-employment tax rules whether they live in the United States or abroad, and that self-employment income still counts when the foreign earned income exclusion takes it off the income-tax line. The IRS self-employment tax page sets the rate at 15.3%, which is 12.4% for Social Security and 2.9% for Medicare, figured on Schedule SE. You owe it when net earnings are $400 or more. For 2026, the Social Security slice applies to the first $184,500 of combined wages, tips, and net earnings, the contribution base on the Social Security wage-base table. An extra 0.9% Medicare tax applies once wages, compensation, and self-employment income pass $200,000 if you file single, or $250,000 if you file jointly. Publication 54's own example is a consultant abroad who qualifies for the exclusion and still pays self-employment tax on a $68,000 net profit. How that sits next to income tax you might exclude is the subject of the self-employment tax trap for freelancers abroad.

The US client files Form 1099-NEC, not you. IRS instructions set the threshold for payments made in 2026 at $2,000 for services by someone who is not an employee. Payments made before 2026 used $600. One $1,500 sprint is under the 2026 line. A second paid job in the same calendar year can push the total across it, and the form still asks for the recipient's taxpayer identification number. They also file at any amount if they withheld backup tax. You report the fee on your return either way. The December 2026 form instructions are the revision for 2026 payments filed in early 2027.

Hand it over and get paid

Do the work in their accounts, in this order, and do not open a paid plan on your own card to "get them started."

Order Action Stop if
1 Name the job, the apps (four or fewer), and what a human still approves. Collect 50%. They describe "everything in the back office" instead of one pipeline
2 They create n8n and the model key. You build on that instance. They want the credentials to live in your login
3 Ship the webhook. Run a live test. Confirm it finishes inside 5 minutes, or inside 180 seconds on the trial. The only design that works is a poll every few minutes
4 Hand over the diagram and watch them sign off. Collect the other 50%. A new app appears that would make five integrations
5 Leave the workflow, the keys, and the history in their account. The agreement still says you own the scenario

Take the dollars into a US business account you can run while you live abroad. Mercury is built for that operating account, so the deposit arrives in dollars without a new local bank in the city where you sleep. Aenfinite's page says the client owns the intellectual property at the end and that there is no mandatory platform fee. Match that in writing. A buyer who cannot log in without you did not buy a workflow. They rented you.

Quote $1,500 when the job is one pipeline and four apps or fewer, with half paid before you touch the account. Quote $3,500 to $6,000 when they want four to eight workflows and a server they host. Quote a $1,000 block at $250 an hour when the process is still a conversation. Leave $15,000 a month for a team you are not staffing. Keep the prompt inside Haiku 5.5's 100,000-token tier, trigger on the form instead of a five-minute clock, and set self-employment tax aside before the deposit feels like spending money.

This is general information, not tax, legal, or financial advice. Rates, plan limits, and IRS thresholds change. Check the pages below and a qualified tax professional before you file or sign a client agreement.

Data notes / sources checked

Prices and IRS figures were checked on 17 November 2026.

Frequently asked questions

How much should I charge for one AI automation workflow?

Charge $1,500 when the job is one pipeline, up to four apps, error handling, and documentation. That is the fixed sprint Aenfinite publishes, delivered in 7 to 10 days. Use $250 an hour, with a $1,000 four-hour minimum, when the scope is still unclear.

Does the client pay for n8n and the model, or do I?

The client pays. n8n Cloud Starter is 20 euros a month billed annually for 2,500 executions, and Claude Haiku 5.5 bills their API key at $0.10 per million input tokens and $0.50 per million output tokens on prompts of 100,000 tokens or fewer.

Do I still owe US self-employment tax if I build this abroad?

Yes. Publication 54 says self-employment tax rules for a US citizen are generally the same abroad, and the foreign earned income exclusion does not remove the income from that tax. The rate is 15.3% when net earnings are $400 or more.

When does a US client file a 1099-NEC for this work?

For payments made in 2026, the IRS threshold is $2,000 for services by someone who is not an employee. One $1,500 sprint is under that line. A second paid job in the same year can push the total over it. Before 2026 the threshold was $600.

This guide is general information, not personalized tax, legal, or investment advice. Rules change; verify current thresholds with official sources or a qualified professional before acting.

AI incomecash flow abroadremote business