How to Hire an AI Automation Freelancer on Upwork or Fiverr

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for How to Hire an AI Automation Freelancer on Upwork or Fiverr.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for How to Hire an AI Automation Freelancer on Upwork or Fiverr.

Describe one workflow and its finished result in a job post on Upwork, or in a message to shortlisted sellers on Fiverr. Shortlist three people with similar automation work, pay the strongest one for a small test task, then hire on a fixed price with milestones. Add the platform's client fees to your budget before you compare quotes.

The platforms handle payment and disputes, not quality. Neither checks that an automation is built sensibly, documented, or running in accounts you own. Those things are yours to specify in the post and the contract, and they're where most small-business automation hires go wrong. If you're still deciding whether a freelancer is the right kind of help at all, start with choosing between a consultant, an agency and a freelancer.

Follow me on Instagram@sagnikteaches

Upwork or Fiverr: matching the marketplace to the job

The two work differently. On Upwork you usually post a job and freelancers send proposals, though it also sells pre-packaged projects that work more like Fiverr's. On Fiverr you mostly browse packaged services called Gigs, message sellers, and ask for a custom offer when your job doesn't fit a package. The practical differences, as each platform's help pages described them in September 2026:

Connect on LinkedInSagnik Bhattacharya
QuestionUpworkFiverr
How you find peoplePost a job and receive proposals, or invite freelancers you've foundBrowse Gigs, message sellers, request a custom offer
Client feesBasic plan: marketplace fee of up to 7.99% on every payment, plus a one-off contract initiation fee of $0.99 to $14.99 per new contract. Business Plus: 10%Service fee of 5.5% on each payment, plus $3.50 on orders under $200. Extras and tips carry their own fee
Paying in stagesFixed-price milestones, funded before work starts, or hourly contracts billed weeklyMilestones on orders over $100: up to 6 in a custom offer, charged as you approve each one
Your review windowHourly: dispute logged hours between Monday noon and Friday 23:59 UTC after each billing week3 days to accept, request a revision or extend the review, or the order completes automatically; 8 days per milestone
Track-record signalsJob Success Score; Rising Talent, Top Rated, Top Rated Plus and Expert-Vetted badgesSeller levels up to Top Rated (individually reviewed); Fiverr Pro for hand-vetted freelancers
Working off the platformWork and payments must stay on Upwork for two years unless you pay a conversion feePayments must go through Fiverr; taking business off-platform risks suspension

A workable rule: Fiverr suits a job you can describe as a package, such as "connect my booking form to a spreadsheet and send a confirmation", where a seller already offers something close. Upwork suits a scoped project with some discovery, several apps and a likely follow-on relationship. Check the fees yourself before you commit, because both platforms revise them; the pages to read are Upwork's Client Marketplace Fee article and Fiverr's paying for orders article.

Subscribe on YouTube@codingliquids

What platform fees add to a small automation budget

Fees look small as percentages and noticeably bigger on small orders. Two quick sums:

  • An $810 project on Upwork's Basic plan: up to 7.99% is $64.72, plus a contract initiation fee of up to $14.99, so budget up to about $890.
  • The same $810 on Fiverr, paid in milestones of $120, $490 and $200: 5.5% of each payment adds $44.55 in total, so budget about $855. Checkout shows whether the $3.50 small-order fee also applies to the $120 payment, which would make it about $858.
  • A $120 test task on its own: on Fiverr, $6.60 plus $3.50 is $10.10 (about 8.4%). On Upwork as a separate contract, up to $9.59 plus up to $14.99 is up to $24.58, which is a fifth of the task.

That last line has a practical consequence. Upwork charges the contract initiation fee once per contract and not again when you add milestones to an existing fixed-price contract, so it's usually cheaper to run the test as the first milestone of a contract and add the rest only if you continue.

For a sense of rates, the platforms' own guides are the only published figures worth quoting. Upwork's 2026 hourly-rates guide lists AI developers at $30 to $50 an hour, AI engineers at $35 to $60 and marketing automation consultants at $40 to $90, and describes these as directional benchmarks. Fiverr's cost guide for AI automation experts, published in April 2026, puts simple two-app integrations using pre-built connectors at $40 to $200 and multi-platform work with custom logic at $200 to $800. Treat both as starting points for comparing quotes, not as prices anyone owes you.

A job post built around one workflow

A vague post attracts vague proposals. Here's a before-and-after from a holiday-let manager looking after 14 properties. The first version:

"Looking for an AI automation expert to automate my holiday rental business with ChatGPT, Zapier and Make. Long-term work for the right person. Budget $200."

It drew dozens of replies within a day, most promising a "complete AI transformation", several quoting prices without a single question, and none mentioning bookings or cleaning. The rewrite names one workflow, the volume, the tools, what finished looks like and four questions every applicant must answer:

Title: Automate booking-to-cleaning handoff for 14 holiday lets
(Make or Zapier, plus one AI drafting step)

About us: we manage 14 holiday-let properties. About 60 bookings a
month arrive through our channel manager.

The job: when a booking is confirmed, automatically
  1. add a cleaning job to our shared Google Sheet (property,
     checkout date and time, number of guests, notes)
  2. message the assigned cleaner with those details
  3. draft a welcome message with AI for me to approve before sending

Tools: [channel manager], Google Workspace, WhatsApp Business app

Done means:
  - 10 test bookings (we supply fake ones) produce correct sheet rows
    and messages
  - a booking with no checkout time is flagged to me, never guessed
  - everything runs in accounts we own
  - a written guide and a screen recording show how to pause, edit
    and restart it

Budget: fixed price, three milestones. Please quote.
Timeline: live within four weeks.

Answer in your own words:
  1. Which platform would you build this on, and what would it cost
     us per month at 60 bookings?
  2. What would you do when a booking arrives without a checkout time?
  3. Which step would you NOT hand to AI, and why?
  4. Describe one similar automation you built that is still running.

On Fiverr the same text works as the brief you send to three sellers whose Gigs are closest, asking each for a custom offer with milestones. If you want help writing acceptance tests like the "done means" list, see how to scope an AI project with deliverables and acceptance criteria.

Screening questions that expose copy-paste proposals

The four questions do the sorting for you. Someone who has built this kind of thing answers question 2 in one or two concrete sentences. A copy-paste proposal ignores it or answers with a general promise about "handling edge cases". Compare two illustrative answers to question 3:

  • Weak: "AI can handle every step with the right prompts, so the whole process will be fully automated."
  • Strong: "I'd keep AI out of the sheet row and the cleaner message. Those are straight copies of booking fields, and a model adds nothing but the risk of a wrong date. AI only drafts the welcome text, and you approve it."

With a dozen proposals you can use a chat assistant for a first pass, as long as it scores against your questions rather than against how confident the writing sounds. Replace freelancers' names with P1, P2 and so on before pasting.

Below is my job post, followed by 12 proposals labelled P1 to P12.
Score each proposal 0-2 on each of these:
  A. answers question 2 (missing checkout time) specifically
  B. names a platform AND an estimated monthly running cost
  C. describes one similar automation that is still running
  D. says it will build in our accounts
Quote the sentence that justifies each score. If a question isn't
answered, score 0. Don't reward length, jargon or confident tone.
Return a table sorted by total score.

In an illustrative run, the assistant gave P7 full marks on C for the sentence "I've built hundreds of automations for property businesses". That's a claim, not a description of one automation, so the correct score is 0, and P7 dropped from first to fifth. The fix is to read the quoted sentences, which is why the prompt asks for them. Expect to correct one or two scores in every batch.

Reading a profile: scores, badges and the history that counts

Platform signals help you rule people out more than rule them in. Upwork's help pages say freelancers with a Job Success Score of 90% or higher have consistently satisfied clients, that scores in the 80s show solid work with some room for improvement, and that small differences, such as 91% against 93%, matter less than large ones. Its Expert-Vetted badge goes to freelancers who have passed a pre-screening with Upwork's talent managers. On Fiverr, Top Rated status requires, among other things, Level 2, a 4.7 rating, a 90% response rate and 40 orders, followed by an individual review; Fiverr Pro freelancers are hand-vetted.

A badge tells you someone keeps clients happy. It doesn't tell you they build automations you can maintain. Look for three things in the work history instead:

  1. Jobs with the same tools and a similar shape. A trigger, a few steps and an AI draft is a different skill from building a custom chatbot.
  2. Reviews that mention handover. Phrases like "clear documentation", "showed us how to change it" or "still running months later" are worth more than "fast and friendly".
  3. Repeat clients. A client who hired the same person three times found the first build maintainable.

Then ask the top three for a short screen recording walking through one past automation, with the client's data hidden. Five minutes of video shows how someone names steps, handles errors and explains their work, which is exactly what you'll need later.

Run the paid test as the first milestone

A test task of two or three hours tells you more than any interview. Keep it small, paid and real enough to reveal habits. For the holiday-let manager, the test was: using three fake bookings, one of them with no checkout time, build step 1 only (booking to sheet row) in a Make account the manager created, and write half a page on how it works. At the $30 to $50 an hour that Upwork's guide lists for AI developers, three hours is $90 to $150 before fees; the manager offered $120.

Judge the result on four points: did the missing checkout time get flagged rather than guessed; is it built in your account; can you follow the half-page explanation without calling them; and did they ask sensible questions before starting. A freelancer who asks "what should happen if the same booking is edited twice?" before writing anything is showing you how they'll behave on the real job.

Access, ownership and milestones in the contract

Most later trouble traces back to three decisions made, or skipped, at this point.

Access

Create the automation accounts (Make or Zapier, and any AI provider account) in the business's name, then invite the freelancer as a member. Never share your main email password or a two-factor code. For AI steps, use an API key from your own account with a monthly spending limit set, so a looping scenario can't run up a bill. Build and test with fake bookings; connect the live channel manager only in the final milestone.

Ownership

Write into the contract that the workflows, prompts, documentation and any code are yours on final payment, and that everything is built in your accounts. Don't assume the platform's standard terms settle this for your situation; read them and add a line if needed. The tutorial on who owns the AI workflows a consultant builds explains why this matters more than it seems.

Milestones

Tie each milestone to something you can check. The manager's contract, filled in:

MilestoneDeliversAccepted whenAmount
1. TestBooking-to-sheet step on fake data, half-page explanationMissing checkout time flagged; built in our Make account$120
2. Full buildAll three steps connected to the live channel manager10 test bookings produce correct rows, cleaner messages and welcome drafts$490
3. Monitored running and handoverTwo weeks live with fixes, written guide, screen recording, error alerts to the managerNo unflagged errors in two weeks; manager can pause and restart it alone$200

Mind the review clocks. On Fiverr you have eight days after each milestone delivery to accept it or ask for a revision, or the order stops and is marked complete, so don't let a delivery sit during a busy changeover weekend. On an Upwork hourly contract, unchecked hours are paid out once the Friday dispute window closes.

A holiday-let manager's hire, start to finish

Here are the numbers behind the example. Before automation, each booking took about 10 minutes of the manager's time: adding the cleaning job, messaging the cleaner and writing a welcome message. At 60 bookings a month that's 600 minutes, or 10 hours. Afterwards, approving the welcome draft and glancing at the sheet takes about 2 minutes per booking, or 2 hours a month. The saving is about 8 hours a month.

The freelancer, found on Upwork, estimated 18 hours of work at $45 an hour and quoted a fixed $810 across the three milestones above. With fees, the manager budgeted up to about $890. Running costs came out small but not zero:

  • Make: the design checked the channel manager every 15 minutes. Make uses a credit for every scheduled check even when nothing arrives, which is about 2,880 credits a month on its own, well above the free plan's 1,000. So the paid plan, from about $9 a month for 5,000 credits, was needed.
  • AI drafting: 60 welcome drafts a month at roughly 1,000 input and 300 output tokens each, on a low-cost model such as gpt-6-luna at $0.10 per million input tokens and $0.50 per million output, costs about two cents a month.

So roughly $10 a month to run, against 8 hours a month saved. If the manager's time is worth $25 an hour, that's $200 a month of time, and the $890 set-up pays back in about four and a half months. The monthly costs are the part people forget; the tutorial on what you pay after an automation goes live covers the rest of them.

Two things went wrong, both caught. In milestone 2 the freelancer built the scenario in their own Make account "to save time" and planned to transfer it later. The manager spotted it at acceptance and didn't release the payment until the scenario was exported and rebuilt in the business's account, which took the freelancer about an hour. Then, in the second week of monitored running, a booking arrived from a new channel with the checkout time in a different field. The flag fired, the manager added the time by hand, and the freelancer mapped the new field within the milestone. That's exactly what a monitored milestone is for.

Signals to walk away before you've paid much

Most bad hires announce themselves early. Walk away, or at least slow down, if a freelancer:

  • asks to continue on WhatsApp or email before an order exists, or suggests paying outside the platform. Fiverr prohibits both, and Upwork requires a conversion fee of 13.5% of the freelancer's estimated annual earnings to take a relationship off-platform within two years;
  • asks for your main account passwords or two-factor codes instead of a member invitation;
  • plans to build in their own accounts and "hand it over later";
  • promises a fully automated process with zero errors, or an AI agent that runs the whole business;
  • quotes far below the platforms' own ranges for the scope, such as $50 for three connected apps with AI and error handling;
  • won't do a paid test, or skips your screening questions.

None of these proves bad intent. Together, they predict the automation nobody can maintain six months later.

Accepting the work and leaving honest feedback

Before releasing the last payment, run your acceptance tests yourself, not in the freelancer's presence: ten fake bookings, the missing-time case, a cancelled booking and an edited one. Pause and restart the automation using only the written guide. Check that error alerts reach you, not the freelancer. Make sure you hold the export of every scenario or Zap, the prompts in a document, and a list of every account and API key involved. The AI consultant handover checklist is a good list to work through even for a small freelance job.

Then leave feedback that helps the next client: what was built, whether it was documented, and whether it worked in week two as well as on day one. If the freelancer did well, keep them on the platform for small changes. A short fixed-price order for each change keeps the history, the payment protection and your records in one place. If you'd like an automation specialist on hand more regularly, hiring a certified Zapier expert is another route worth comparing.

Hiring automation freelancers on marketplaces: follow-up questions

Should I choose an hourly or fixed-price contract for an automation job?

Fixed price with milestones suits most small-business automations, because you can describe the finished result and pay only when each stage works. Hourly suits open-ended work such as untangling an old set of Zaps, where nobody knows the scope yet. If you go hourly on Upwork, set a weekly hour limit and check the timesheet every week, since the window to dispute hours closes on Friday.

Can I hire the same freelancer directly after the first project?

Not freely on Upwork: its terms require work and payments found through the platform to stay there for two years unless you pay a conversion fee, which is 13.5% of the freelancer's estimated annual earnings. Fiverr requires payments to go through the platform and restricts taking the relationship off it. For small ongoing jobs, staying on the platform is usually cheaper than converting.

Do I need an NDA with a marketplace freelancer?

If the freelancer will see customer data, contracts or pricing, a short confidentiality agreement is sensible, and many freelancers will sign one. Better still, build and test with fake or anonymised data, and give access to live systems only for the final milestone. Check what the platform's own terms already say about confidentiality before adding your own document.

How long should a small automation project take?

A single workflow connecting two or three apps with one AI step is often a matter of days to a few weeks, depending mostly on how quickly you supply access, test data and answers. Fiverr's own cost guide describes simple two-app integrations as three to seven days of delivery and multi-platform work with custom logic as one to three weeks.

Further reads

Sources: Upwork Help (Client Marketplace Fee, Contract Initiation Fee, Business Plus fees, Conversion Fee, Job Success Score, disputing hours); Upwork Resources, Upwork hourly rates (2026); Fiverr help pages (paying for orders, milestones, accepting a delivery, Top Rated freelancers, off-platform policy, Fiverr Pro); Fiverr cost guide for AI automation experts (April 2026); Make and OpenAI pricing pages.

Want your automation brief checked before you post it?

On a 1:1 call we'll turn your workflow into a tight brief with acceptance tests, decide which platform it should run on, and list the access and ownership terms to agree before anyone starts.

Book a 1:1 call with me