Start by listing the automation work you'd give them in the first six months and how many hours a week it really is. Then choose employee, part-timer or contractor; write the advert around outcomes; test shortlisted candidates with a paid practical task on your own tools; and give them access through accounts the business owns.
The most common mistake is hiring before the work is defined. An "AI person" with no backlog either drifts or builds something impressive that nobody asked for. The second most common is hiring a builder with nobody to own the processes they automate. If you haven't yet weighed a hire against outside help, comparing a consultant with an in-house hire covers that choice, and whether a small business should hire a prompt engineer explains why the title rarely fits.
Count the backlog before you write the job
A backlog is simply the list of jobs you'd hand the specialist, each with a rough build time, the upkeep it'll need, and the hours it should save. It turns "we need someone for AI" into a job with a size. Here's one for an illustrative group of three campsites with 380 pitches, glamping pods and 14 year-round staff, busy for about 26 weeks a year:
| Automation | Build hours | Upkeep hours a week | Staff hours saved a week (season / off-season) |
|---|---|---|---|
| Booking confirmations and pre-arrival messages with AI-drafted answers to common questions | 40 | 1 | 12 / 3 |
| WhatsApp and email FAQ drafts for reception to approve | 50 | 1.5 | 15 / 4 |
| Review replies drafted for the manager | 15 | 0.5 | 4 / 1 |
| Supplier invoice capture into the accounts | 30 | 0.5 | 5 / 3 |
| Pitch allocation suggestions for awkward bookings | 45 | 1 | 6 / 1 |
| Weekly occupancy and feedback summary for the owners | 20 | 0.5 | 3 / 0 |
| Total | 200 | 5 | 45 / 12 |
The estimates are rough, and that's fine; their job is to size the role. Two hundred build hours spread over six months is about 8 hours a week, plus 5 hours of upkeep once things are live and time for the next ideas. That's a two-day-a-week role, not a full-time job, at least for the first year. A group this size can keep a part-time specialist busy; a single small campsite probably can't, and would be better served by a project and a short retainer, or by training someone already on the team. Before you write the list, it helps to know what's already running; an automation audit of your existing Zaps and scenarios often finds half-built work to finish first.
Employee, part-timer or contractor: the shape of the role
| Shape | Suits | Watch out for |
|---|---|---|
| Full-time employee | A backlog of 30+ hours a week that won't run out, several teams to support | Rarely the right first hire for a small business; hard to justify if the backlog shrinks |
| Part-time employee | Steady work of one to three days a week, a wish to build knowledge in-house | Finding people who want part-time technical roles; cover for holidays |
| Long-term contractor | Uneven or seasonal work, a need to start quickly | Ownership and documentation terms; being one client among many in your peak season |
| Existing staff member, trained | Someone who knows the processes and has hours to spare | Protecting their time from the day job; getting outside review of their first builds |
For contractors, the published marketplace rates give a starting point: Upwork's 2026 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. For employees, benchmark the salary yourself: look at five or six current advertisements for automation or operations-automation roles where you'd recruit, note the ranges, and add your employer costs. Employment status rules differ by country, so if a contractor would work only for you on fixed days, check with an HR or employment adviser whether the arrangement is really self-employment.
Skills that matter for small-business automation work
The job is mostly connecting everyday tools, adding AI steps where they help, and keeping it all running. The skills that matter, roughly in order:
- Process mapping. Turning "how reception handles a late arrival" into steps, exceptions and a decision about what not to automate.
- No-code platforms. Real experience with at least one of Zapier, Make, n8n or Power Automate, including error handling.
- Working with APIs and webhooks. Enough to read documentation and connect an app that has no ready-made connector.
- Prompt design and testing. Writing instructions for AI steps, building a small test set and checking outputs, not just trying a prompt once.
- Data hygiene. Spotting duplicate records, inconsistent fields and missing values before they break an automation.
- Documentation and handover. Writing things down so someone else can run and change them.
- Security basics. Named accounts, least-privilege access, API keys with spending caps, no passwords in messages.
- Explaining things to non-technical staff. The automations only work if reception trusts them.
What you don't need: experience training AI models, a research background or a data science degree. A realistic mistake is hiring exactly that profile. A holiday-let company that hires a machine-learning graduate for automation work may find them building a custom model to predict cleaning times, which is interesting and months long, while the booking confirmations the manager actually wanted stay manual.
An outcome-based advert, before and after
A vague advert attracts generalists and hopefuls. The campsite group's first draft:
"Exciting opportunity for an AI expert to transform our business with the latest technology. Must have experience with ChatGPT, automation and machine learning. Passion for innovation essential."
It attracted applicants who listed every AI tool they'd heard of and none who asked about the campsites. The rewrite describes the work, the tools and what success looks like in six months:
Automation and AI Specialist (part-time, 2 days a week)
About us: three family-run campsites, 380 pitches and glamping pods,
14 year-round staff, busiest April to September.
In your first six months you will:
- automate booking confirmations and pre-arrival messages, with AI
drafting answers to common guest questions for staff to approve
- set up AI-drafted replies to WhatsApp and email FAQs
- automate supplier invoice capture into our accounts software
- document everything so our office team can run it day to day
Our tools: [booking system], Google Workspace, [accounts software],
WhatsApp Business, Make.
You'll need: two or more automations you built that ran for at least
six months; hands-on Make or Zapier experience; comfort reading API
documentation; clear writing.
Nice to have: hospitality or bookings experience; some scripting.
Hours: 2 days a week, fixed days agreed together; hybrid.
Pay: [your range, from your benchmarking].
To apply: send your CV and 200 words on how you'd handle a guest
message that asks about dogs, a late arrival and a refund in one
email.The application question does most of the screening. Someone who says the AI should draft the dog and arrival answers but send the refund request to a person has already shown the judgement the job needs. For more on the wording, see writing a job advert for an AI-savvy admin assistant, whose principles carry over.
Where good candidates tend to come from
General job boards work, but the strongest applicants for small-business automation roles often come from places where they already show their work:
- Automation platform communities. Zapier, Make and n8n all run community forums, and people who answer other users' questions well are showing exactly the skills you need.
- Platform partner or expert directories, where the vendors list people and firms who build on their tools. These skew towards agencies, but many list solo specialists.
- Contractors you've already worked with. A freelancer who built one good automation for you is a candidate for the ongoing role, with a track record you've seen.
- Your own team and network. An operations person at a similar business who has automated their own work is often a better fit than a pure technologist.
Whichever route, agree a probation or trial period and use the 90-day plan below as its yardstick. It gives both sides a fair, specific basis for deciding whether the role is working, rather than a general impression.
Screening with a walkthrough and three questions
Ask shortlisted candidates for a ten-minute screen recording walking through an automation they built and maintained, with any client data hidden. Then use the interview for three questions whose answers separate builders from maintainers:
- "Tell me about an automation that broke after it went live." Weak answer: "Mine don't really break." Strong answer: a specific failure, how they noticed, what they changed so it wouldn't happen again, such as adding an alert when a field comes through empty.
- "How do you decide whether a step should use AI?" Weak: "AI makes everything smarter." Strong: "If the step is copying a field or following a fixed rule, no AI. If it's reading free text or drafting wording, AI, with a person approving anything customers see."
- "How would you hand this over if you left tomorrow?" Weak: "I'd be around to help." Strong: a description of their documentation, where credentials live, and a test the next person could run.
A paid task built on your own tools
The practical task tells you what the interview can't. Keep it to three or four paid hours, set it up in a sandbox with invented data (never real guest records), and make it resemble a real item from your backlog. The campsite group's task:
Using the Make account and test form we've set up, build a scenario that takes a booking form submission, drafts a pre-arrival email using the attached fact sheet and an AI step, and flags any booking that mentions a dog to the office. Include at least one test that shows what happens when the arrival time is missing. Write a one-page handover note.
Two candidates' results, scored by the operations manager with an outside reviewer:
| Criterion (0 to 3) | Candidate A | Candidate B |
|---|---|---|
| It works on the test bookings | 3 | 3 |
| Missing arrival time handled and alerted | 1 (guessed "after 3pm") | 3 (flagged to the office) |
| AI draft used only the fact sheet | 2 (invented a shop opening time) | 3 |
| Handover note usable without them | 1 (half a page of bullet points) | 3 |
| Questions asked before starting | 0 | 2 (asked what "flag" should look like) |
| Total out of 15 | 7 | 14 |
Candidate A's interview had been more polished. The task showed who would leave the group with something maintainable. It doesn't matter whether candidates use AI to help them write the scenario or the note; what matters is whether the result works and whether they can explain every step of it. If you'd rather find a contractor for a single project first, hiring an AI automation freelancer on Upwork or Fiverr covers the marketplace route.
Accounts, ownership and the first 90 days
Before the first day, create the accounts the specialist will work in, all in the business's name: the automation platform, any AI provider accounts with monthly spending caps, and a password manager. Give them the role they need, not the owner's login. Put ownership of everything they build in the contract or employment terms, and agree a documentation standard: every automation gets a short page saying what it does, what it connects, what can go wrong and how to pause it.
Then agree a 90-day plan. The campsite group's, filled in:
- Days 1 to 30: audit existing automations and accounts; time the current process for the first two backlog items; build booking confirmations in shadow mode, where drafts are produced but staff still send their own.
- Days 31 to 60: confirmations and pre-arrival messages live; FAQ drafts in shadow mode; a weekly 30-minute review with the operations manager.
- Days 61 to 90: FAQ drafts live for reception to approve; documentation complete for everything built; monitoring alerts tested; next quarter's backlog agreed.
Timing matters for a seasonal business. The group started the specialist in February, eight weeks before the season, so the first automations met real traffic gradually. Starting in July would have meant building while reception was at its busiest and least able to test anything. Staff also need an introduction to what's changing; onboarding new hires onto your AI tools and rules covers the rules side for everyone.
Why first automation hires stall
- No owner for the processes. The specialist automates reception's work, but nobody at reception is responsible for saying what "right" looks like, so every draft becomes a debate.
- A single point of failure. Six months in, only the specialist understands how anything works. Documentation from day one, and a second person who can pause every automation, prevents it.
- Personal accounts. Scenarios built in the specialist's own account look fine until they leave. Everything goes in business accounts from the start.
- Tool sprawl. Five platforms for six automations means five bills and five places for things to break. Agree the core platform early.
- A backlog that runs dry. The first builds finish and the role drifts into general IT support. Review the backlog quarterly and resize the role if it's shrinking.
Does the hire pay? A campsite group's two-year sum
From the backlog, the finished automations save about 45 staff hours a week in season and 12 off-season: 45 × 26 + 12 × 26 = 1,482 hours a year. At an illustrative $20 an hour of staff time, that's about $29,600 a year, once everything is live.
A contractor at $45 an hour (inside Upwork's published ranges for this kind of work), two days a week for 44 weeks, costs 704 hours, or about $31,700 in year one. Savings in year one are lower than the full figure because automations go live one by one; call it two-thirds, or about $19,800. So year one costs more than it saves. In year two, the role drops to one day a week for upkeep and new work, about $15,800, while savings run at the full $29,600. Over two years the group spends about $47,500 and saves about $49,400, and every year after that is clearly positive. Add platform and AI usage costs, which for this backlog run to a few hundred dollars a year, and the conclusion holds.
That's the honest shape of a first specialist hire: it rarely pays in the first year and often pays well after it, provided the backlog is real. Run the same sum with your own hours and rates before you advertise. If the numbers only work on the most optimistic assumptions, a fixed project with a short retainer, or training someone you already employ, will get you most of the benefit for much less.
Hiring an automation specialist: follow-up questions
Should I train someone I already employ instead?
Often that's the better first step. An operations or office manager who knows your processes, has a few hours a week and likes tidying systems can learn Zapier or Make and prompt design faster than an outsider can learn your business. Pair them with an outside expert for the first build and a review of their work. Hire when the backlog outgrows the hours they can give.
What job title should I use?
Use words candidates search for and that describe the work: automation specialist, operations automation, no-code automation or workflow automation, with AI in the description. Avoid titles like prompt engineer or head of AI for a small-business role; they attract the wrong applicants and set expectations the job can't meet.
How do I check a candidate's references for this kind of role?
Ask referees about automations that ran for months after the candidate built them, how problems were handled, and whether anyone else could maintain the work from the documentation. Those three answers tell you more than general praise. Speak to at least one referee who was the candidate's client or manager, not a colleague.
Do they need to be able to code?
Not for most small-business work. Comfort with no-code platforms, webhooks, simple data formats and reading API documentation covers most tasks. Some scripting helps with edge cases, but a specialist who writes custom code for everything leaves you with systems only they can maintain. Ask how they decide between a no-code step and code.
Further reads
- AI or a New Hire? How to Decide Before You Recruit — Whether the work needs a person at all.
- Hire an AI Specialist or Upskill Your Agency Team? — The hire-or-upskill question in more depth.
- Zapier vs Make vs n8n for AI Automation: Which Fits Your Business? — The platforms your specialist will be choosing between.
- How to Stop Zapier and Make Automations Breaking Silently — The monitoring habits to expect from day one.
- What Is a Fractional Chief AI Officer and Do You Need One? — A senior, part-time alternative for strategy rather than builds.
- AI Consultant Handover Checklist: What You Need Before They Leave — What documentation good automation work leaves behind.
- What Is n8n, and Should a Small Business Use It? — n8n explained for owners: how executions are counted, what the AI nodes do, cloud or self-hosting, a wholesaler's first workflow and who should skip it.
- No-Code AI Tools vs a Custom Build: When Templates Stop Working — Six signs a Zapier or Make workflow has outgrown no-code, the middle options to try first, and a worked cost comparison for a care agency's shift offers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Upwork hourly rates guide (2026); Zapier, Make and API pricing pages (checked September 2026).