The job ad is half-written, and the question behind it is, “Should you hire an AI specialist?” “AI Specialist. Experience with automation platforms, LLM integration, workflow design.” It has been sitting in a draft folder for three weeks because something about it does not feel right.
Good instinct. For most founder-led businesses between one and fifty people, the honest answer is not yet, and it has nothing to do with budget or talent supply.
A specialist needs something to specialise on. Walk one into a business where every process, priority and client quirk lives in the founder’s head, and you have hired an expensive person to interview you for six months.
Below: what the role actually involves, why the timing is usually wrong, a readiness test you can run this afternoon, and what to put in place first.
Key Takeaways
- Should you hire an AI specialist? Not until your business knowledge lives somewhere other than your own head, because a new hire cannot read a head.
- MIT research found 95% of enterprise AI initiatives return nothing measurable, and the failures start with tool-first thinking instead of process.
- An AI specialist spends their first three months learning your business rather than building anything, and in an undocumented business that stretches to six.
- Most businesses under fifty people need two to four automations running in production, not a full-time person maintaining a tech stack.
- Write down your ten most repeated tasks before you write a job ad. If none of them is documented, the hire will fail for reasons the hire cannot fix.
- Hire an AI specialist once you have automations in production, a documented process library, and a request queue longer than your capacity to build.
- A contractor or an outside partner on a fixed scope beats a permanent hire when the work is a ninety-day build rather than an ongoing function.
- Every automation built without documented context gets rebuilt from scratch the day the person who built it leaves.
What an AI Specialist Actually Does All Day
There is a gap between the role founders imagine and the role that exists.
The imagined version: a clever person arrives, looks at the business, and starts making things automatic. Emails answer themselves. Quotes go out faster. The admin pile shrinks. Six weeks in, you notice you have your Tuesdays back.
The actual version breaks into four buckets, and only one of them is building.
Discovery, which is mostly interviewing you
Before anyone automates a process, someone has to write the process down. That means sitting with you and asking how a quote gets approved, what makes a lead worth chasing, which clients get called back same-day and which can wait, what happens when a job goes wrong. Every answer lives in your head. Every answer takes a meeting to extract.
In a business with documented processes, this phase takes two or three weeks. In a business without them, I have watched it run four months, because the specialist can only pull knowledge out at the rate you have time to sit down and give it.
Integration plumbing
Connecting your CRM to your accounting system to your booking tool to your inbox. This is genuine technical work, it is slow, and it is where most of the first six months actually go. It is also the least visible work in the building. You will not feel it. You will just keep paying for it.
Building the thing
The part everyone pictures. Usually 20% to 30% of the total effort.
Maintenance and change management
An automation is not a finished object. Your pricing changes, a supplier switches systems, a team member leaves and the handover process shifts. Every one of those breaks something. Someone has to fix it, and someone has to convince the team to actually use what was built.
That last part is the one nobody costs into the job ad. Development is roughly 30% of the work. Adoption is the other 70%. A brilliant build that the team quietly ignores is worth precisely nothing, which is the failure mode behind most AI automation failures in small business.
Should You Hire an AI Specialist Before You Have a System? No
Here is the uncomfortable number. MIT research on enterprise AI adoption found that roughly 95% of AI initiatives deliver no measurable return. Not 95% of bad ones. 95% of all of them.
The ones that work share a trait: they start with a process, not a technology. The ones that fail start with a capability and go looking for somewhere to point it.
Hiring a specialist first is the purest version of starting with a capability. You have acquired a very capable hammer, and you are now going to spend six months, at full salary, working out which of your problems are nails.
The context problem
Ask yourself what would happen if a new AI specialist started on Monday and you were unreachable for a fortnight.
They cannot proceed. Not because they are not good, but because there is nowhere to look up how the business works. No record of which clients need which handling. No written pricing logic. No documented escalation rules. The information exists; it just exists inside you, and it comes out at the speed of conversation.
This is the same reason your ops manager took five months to become useful, and the same reason your last hire keeps asking you questions you have answered twice already. It is not a people problem. Your team is capable. They just do not have the context to make decisions without you, which is a different failure entirely and one more staff will never fix.
The scale problem
There is a second issue, and it is more awkward to say out loud.
Most businesses under fifty people do not have enough AI work to fill a full-time role. They have three or four processes that badly need automating, then a long gap, then a maintenance load of a few hours a week.
So the specialist automates the obvious things in months four through eight, and then needs something to do. What they do next is invent work. More dashboards. More integrations. A rebuild of the thing they built in month five. It looks like progress, and it is mostly motion.
The bus problem
The knowledge that a specialist builds up over a year sits in their head and in a set of workflows only they understand. When they leave, and technical people in a hot market leave, you own a stack of automations nobody can modify. I have picked up work from businesses in exactly this position. The rebuild costs more than the original.
Everything the specialist learned about your business was learned again from scratch, and then walked out the door. You solved a knowledge concentration problem by moving the concentration to a different person.
The Readiness Test: Should You Hire an AI Specialist Yet?
Seven questions. Answer honestly, out loud, right now. This takes ten minutes and will save you a bad twelve months.
1. Can you name your ten most repeated tasks? Not “admin”. Specific ones. “Chase unpaid invoices past 14 days.” “Re-quote jobs that went quiet.” “Send the Friday recap to the three biggest accounts.” If you cannot list ten in two minutes, you do not yet know what you would ask a specialist to build.
2. Are any of those processes written down anywhere a stranger could follow? Not in a wiki nobody opens. Somewhere current and readable. If the answer is no, your specialist’s first six months are documentation, and you can do that part yourself, faster, for free.
3. Do you already have at least one automation running in production? Something that fires without a human. If you have never shipped one, hiring a person to run a function that does not exist yet is a strange first move.
4. Is there a queue of automation requests you cannot get to? A real backlog, from real people, with real time cost attached. If the queue is one item long, that is a project, not a role.
5. Can you describe what “working” looks like in numbers? Hours recovered per week. Response time cut from four hours to four minutes. Percentage of recurring tasks handled without you. Vague briefs produce vague builds, then arguments about whether it was worth it.
6. Would you know if the specialist was doing a good job? This is the one founders skip. If you cannot evaluate the work, you cannot manage the work, and the role becomes a black box you fund monthly and hope about.
7. Is your data anywhere near usable? Are contacts in one system or four spreadsheets? Is the CRM current or abandoned in 2024? A specialist cannot automate on top of data that nobody trusts, so their first job is cleaning it, which is a data project wearing an AI job title.
Score it: six or seven yeses, and hiring is a reasonable next step. Three to five and you have a scoping problem, not a staffing problem. Two or fewer, and a hire right now will burn a year and a six-figure salary to arrive at a place you could reach in a quarter without one.
Most founders who ask me this question land on two or three. That is not a failing. It is just early, and the 90-day checkpoint is a better first move than a job ad.
What to Build Before You Hire Anyone
If the test came back low, here is the sequence that actually works. None of it requires a permanent hire, and all of it makes a future hire dramatically more effective.
Get the business knowledge out of your head
This is the whole thing. Everything else stacks on top.
You are writing down what you know so that something other than you can act on it. Who you are, what you sell, how you price, which clients need which handling, what your team does, what the strategy is this quarter, what happens when things go wrong.
Think of it as the briefing you would give a new executive on their first day, except it gets written once and never forgotten. That document set becomes the brain of your business. An AI reads it before every task. A new staff member reads it in week one. You stop being the only source.
The test that it is working: ask an AI a strategic question about your business and see whether the answer is generic or specific. If it comes back with your actual clients, your actual numbers and your actual priorities, the knowledge is out of your head. If it comes back with something you could have got from a Google search, it is not.
Get your numbers into one place
Most founders check five or six systems every morning to work out what happened yesterday. Accounting, CRM, analytics, the booking tool, an inbox, a spreadsheet someone updates on Fridays.
Pick your single most important source and connect it first. Usually whatever answers “how much money came in” or “how many leads arrived”. One connection, one source of truth, refreshed automatically. Not a migration and not a new platform. Your existing tools stay exactly where they are.
Automate two things, not twenty
Choose the highest-volume, lowest-judgment tasks. Not the interesting ones. The boring, repetitive, high-frequency ones that eat an hour a day and require no thought.
Lead response is almost always in the top two. Harvard Business Review’s research on the short life of online sales leads found that firms responding within an hour were about seven times more likely to have a meaningful conversation with a decision maker than those responding an hour later, and sixty times more likely than those waiting a day. Most businesses take four hours or more. A system takes ninety seconds, every time, including Sunday.
Dormant database reactivation is the other one that pays for itself immediately. One finance client, James, had 319 contacts his team had written off entirely. A multi-touch reactivation sequence across SMS and email recovered $49,000 from people who already knew the business. No ad spend, no new leads, just systematic follow-up on contacts that were already paid for.
Missed calls sit close behind. A dental practice I work with, Dr Claire, had two receptionists and was still losing nearly half her inbound calls at peak. Bookings rose 44% once every call was answered and qualified around the clock. That was not a staffing problem being solved by more staff.
Then, and only then, look at hiring
Now the picture is different. You have documented processes a new person can read on day one. You have data in one place. You have two or three automations running, which means you know what good looks like and can brief against it. You have a queue of the next ten things you want built.
A specialist walking into that business is producing in week two. A specialist walking into the version of the business you have today is interviewing you until Christmas.
System first, then hire. Every hire after the system is in place is worth more, because the context is already there and it stays there when they leave. This is the same argument I make about AI adoption for SMEs generally: the sequence matters more than the tools.
When Hiring an AI Specialist Is the Right Call
I am not arguing that nobody should hire for this. There are clear cases where a permanent hire is correct, and pretending otherwise would be dishonest.
You have more than fifty people and multiple departments. At that size, the internal request volume genuinely fills a role, and the coordination work of managing an outside partner across five departments starts to cost more than an employee.
AI is in your product, not just your operations. If you are building something customers use, that is a product engineering role, and it is not optional. Different question entirely.
You are in a regulated industry with data that cannot leave the building. Health, finance, legal. Sometimes the compliance position makes external help genuinely hard, and internal capability is the only workable answer.
You already have five or more automations in production and a real backlog. You have graduated. The maintenance load alone justifies the role, and you know how to evaluate the work because you have been living with it.
You are deliberately building an internal capability as strategy. Some founders decide AI fluency will be a competitive advantage in their sector and want it in-house permanently. That is a legitimate strategic bet. Just make it deliberately, with a two-year view, rather than as a reaction to feeling behind.
Notice what none of those looks like. None of them is “our competitor posted about AI on LinkedIn, and I felt slow.” That feeling is real, and it is not a hiring signal.
What to look for if you do hire
Bias hard toward business understanding over technical depth. The technical side of this work has become dramatically easier in the last two years. The hard part is knowing which problems are worth solving and getting a team to change how they work.
The best question in the interview: “Tell me about an automation you built that nobody used. What happened?” Anyone with real production experience has one. The ones who claim they do not have one either never shipped or are not telling you the truth.
Employee, Contractor, or Partner: Choosing the Shape of the Help
Founders treat this as a binary. It is not. There are three shapes, and most businesses under fifty people fit the middle two.
A permanent employee makes sense when the work is continuous, the volume is high, and the knowledge needs to compound inside the business. Expect a slow start. Expect to manage them, which is its own hour cost you should count honestly.
A contractor on a defined scope suits the ninety-day build. You know the three things you want automated, the scope is bounded, and when it is done, it is done. The risk is documentation. Contractors ship and leave, so put the handover requirement in the contract, not in a conversation.
An outside AI partner suits founders who want the system built, installed and working without becoming the person who manages it. The work is done for you, the knowledge stays documented in the business rather than in a contractor’s head, and the engagement continues as the business changes: more training, more optimisation, more implementation, more support. The advantage over a hire is speed to first result. The disadvantage is that you are dependent on someone outside the business, so pick one that documents everything and hands over cleanly.
There is also the option nobody mentions, which is often the right one: train someone you already have. Your most process-minded existing team member, given a day a week and outside guidance, will outperform an external hire in year one more often than you would expect. They already have the context. Context is the expensive part. If you go this route, the discovery call checklist is a decent framework for scoping what they take on first.
The decision is not really “employee or not”. It is “what is the smallest commitment that gets the first automation into production this quarter”. Start there. The answer to the staffing question gets much clearer once something is actually running.
The Real Question Underneath
When a founder asks whether they should hire an AI specialist, they are almost never asking about AI. They are asking a different question: how do I stop being the bottleneck?
Every decision runs through you. Every escalation lands on you. Your team waits for you because you are the only one with the full picture. You have tried tools, and they became more tabs. You have tried hiring, and it added management. So AI looks like the next lever, and hiring an AI person looks like the way to pull it.
But the constraint was never the technology, and it was never headcount. The constraint is that the business has no brain of its own. Everything it knows, it knows through you. Add a person to that, and you have distributed the chaos across two heads. Add a tool, and you have another login.
Get the knowledge out of your head and into a system. Get the numbers into one place. Automate two boring things and feel the difference. Then decide whether you need a specialist, because by that point the question answers itself, and the answer is usually a much smaller commitment than a full-time salary.
Should you hire an AI specialist? Ask again in ninety days, after something is running. You will have a much better answer, and it will cost you a lot less to be wrong.
Where to Start
If you are somewhere between “I know I should be doing something with AI” and “I am about to post a job ad”, have a conversation before you commit to either.
Book a 30-minute Discovery Call. No pitch deck. I will ask what actually eats your week, tell you honestly whether a hire makes sense in your situation, and point at the two or three things worth automating first. If the answer is that you should hire someone, I will say so.
Frequently Asked Questions
What does an AI specialist actually do in a small business?
Four things: interviewing you to document how the business works, connecting your existing systems to each other, building automations, and maintaining them as the business changes. Building is usually only a quarter of the total effort. Most of the first six months go into discovery and integration plumbing, which is slow, invisible work that founders rarely account for when they write the job ad.
How much does it cost to hire an AI specialist in New Zealand?
Advertised ranges I see sit roughly between NZD $110,000 and $180,000 depending on seniority and whether the role skews technical. Add recruitment costs, three to six months of ramp-up before meaningful output, and your own management time. The honest total for year one is usually well north of the salary line, which is why the readiness test matters before you commit.
Should I hire an AI specialist or work with an AI automation partner?
Under fifty people, a partner is usually faster to first result because they arrive with pre-built modules and have solved your problem before. A permanent hire makes more sense once you have five or more automations running and a genuine backlog. The deciding question is whether the work is continuous and high volume, or a bounded build with maintenance after it.
What qualifications should an AI specialist have?
Weight business understanding over credentials. Certifications tell you almost nothing about whether someone can spot which process is worth automating or convince a sceptical team to change how they work. Ask for two automations they have shipped into production and one that nobody adopted. The second answer is more revealing than the first, because everyone with real experience has one.
Can I train someone already on my team instead of hiring?
Often the better move. Your most process-minded team member already holds the context, which is the expensive part to acquire. Give them a day a week, outside guidance, and one specific process to automate first. They will frequently outperform an external hire in year one. The failure mode is giving them the title without protecting the time.
Is it too early to hire for AI if my business is under twenty people?
Almost always, yes. At that size, there usually is not enough continuous work to fill the role, so the person automates the obvious three things and then invents work to stay busy. Two or three automations built by a contractor or partner, plus documented processes, get you further than a salary. Revisit the question once you have a backlog you cannot clear.
How do I know if my business is ready for AI at all?
Ready means you can name your ten most repeated tasks, at least a few are written down, your customer data lives somewhere current, and you can describe what success looks like in numbers. If all four are true, you are ready to automate. If none are, start by writing down how your business actually works. That step costs nothing and changes everything downstream.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.