Hire an AI person, engage a firm, or do it yourself.
The three real options, what each one costs over twelve months with your own numbers in it, and the cases where the answer is not to pay anybody. Written by a firm that sells one of these three, which is exactly why the case against us is on the page.
The short answer
For most businesses under about twenty people, the first useful spend is neither a hire nor a consultant: it is six weeks of getting the AI licences you already pay for into the work people actually do. Engage a firm when several workflows need to run this quarter and nobody internally has the time. Hire when AI is how your product works rather than how your business runs, and the queue of work is permanent rather than a backlog you are clearing once. The number that decides it is not the salary against the fee, it is the fully loaded twelve-month cost of each against what actually ships in that time.
Three options, not two
The question is usually asked as a choice between hiring and paying someone. There is a third option that beats both often enough that leaving it out would make this page an advert, so it is listed first.
Do it yourself first
Under about twenty staff, no AI in the product, and one person internally with appetite and a few hours a week.
Licence seats you already pay for, plus six weeks of somebody's attention.
It stalls when that person gets busy, and nothing is written down, so the knowledge leaves with them.
Engage a firm
You need three or four workflows running this quarter, nobody internally has the time, and the requirements touch systems that already exist.
A fixed fee for the build, then either a retainer or nothing if you take it in house.
A working system nobody inside can change. This is the failure mode to price for, and the handover is what prevents it.
Hire in house
AI is how the product works rather than how the business runs, and the queue of work is permanent.
A fully loaded salary from day one, against output that starts somewhere in month three or four.
Twelve months of salary spent learning what a six-week engagement would have taught, and a single point of failure if they leave.
The numbers
What twelve months of each actually costs
Most comparisons of this decision argue it in prose and quote a salary. The two costs that decide it in practice are the ramp months a first hire spends learning your business, and the manager time an engagement does not consume. Both are in here, and the model will tell you to hire when the numbers say so.
Put your own numbers in
The defaults are real: the salary is the one we advertise ourselves, and the engagement fee is the twenty thousand people most often arrive here holding a quote for. Change any of them. Everything runs in your browser and nothing is sent anywhere.
Hiring in house
$120,920
- Salary plus on-costs$94,080
- Recruitment and setup$8,000
- Your time directing it$14,040
- Models and tooling$4,800
- Productive months in year one9 of 12
- Cost per productive month$13,436
Engaging a firm
$34,160
- Engagement fee$20,000
- Retainer for the rest of the year$0
- Internal owner time$9,360
- Models and tooling$4,800
- Productive months in year one9 after handover
- Who holds the knowledgeThe firm, unless you take it
Twelve months of hiring costs $120,920. The engagement as you have specified it costs $34,160, which is $86,760 less.
Those two numbers do not buy the same thing, and the difference is not a saving until you make them comparable. A hire is a year of capacity. The engagement is a fixed scope, plus whatever ongoing work you set the retainer to. Your retainer is zero, so the engagement column currently covers 3 months of build and nothing after it. If work will keep arriving, put a real monthly figure in before you compare.
Cost is also not the whole decision, and this model does not pretend it is. A hire compounds: in year two there is no ramp and no recruitment, and the number that matters becomes what they ship rather than what they cost. An engagement front-loads the output and leaves a dependency unless the handover is real. Use the money as one input and the questions below as the others.
What actually decides it
Cost is fourth on this list on purpose. It is the question people ask, and the fourth most useful answer.
Is the work permanent or is it a backlog?
A backlog is a project and projects suit engagements. A permanent queue is a role. Most organisations describe a backlog and budget for a role, which is how a first hire ends up maintaining three automations and being bored by month five.
Who owns it on the day the work stops?
Name the person before anything is built. Not a team, a person, with the authority to change what runs. An engagement with no named internal owner produces a system that decays quietly, and a hire with no internal sponsor produces the same thing more expensively.
How close is this to the product?
If your customers touch the AI, the knowledge belongs inside the company and a hire is the honest answer eventually. If AI is how your operation runs, a governed install and an internal owner beats a hire for at least the first year.
What does twelve months of each actually cost?
Run your own figures through the model above rather than accepting either side's framing. The two numbers that swing it are the ramp months a first hire spends learning your business, and whether the engagement leaves you able to change what was built.
Can you hire the person you are imagining?
The engineer who can build agents, integrate them into systems you already run, and hold a governance conversation with your board is not sitting in the market at the salary most plans assume. If the search takes five months, the model above has already answered the question.
The case against us
When not to hire a firm like ours
We sell one of the three options on this page. Here is where it is the wrong one, stated plainly, because a comparison that always lands on the author is not a comparison.
- Your requirements are simple and stable. An off-the-shelf tool and a few days of attention will beat any engagement on cost, and we will tell you that in the audit rather than three months in.
- You already employ an engineer with capacity and appetite. Give them the six weeks first. If they get it running, you have saved the fee and gained the internal owner every path needs anyway.
- Nobody internally is going to own what gets built. The most expensive outcome on this page is a working system with no owner, and no amount of engagement quality prevents it.
- You want the certificate rather than the governance. We prepare organisations for ISO/IEC 42001 and cannot certify anyone, so if the actual requirement is a certificate this quarter, an accredited body is the call to make first.
Where an engagement does win, what we sell is the AI Operating System: the harness installed over the Claude, ChatGPT, Copilot or Gemini organisation you already pay for, with the agents and the governance on top. The client owns the harness, which is the part that decides whether the dependency in option two ever becomes real.
Frequently asked questions
Should a small business hire an AI person or use a consultant?
Under about twenty staff, usually neither at first. The work that pays back in the first quarter is enablement: getting the licences you already buy into people's actual tasks, writing down how the business does things so an assistant can follow it, and picking one workflow to automate properly. That is weeks of focused effort, not a permanent salary. Hire when AI is load-bearing in the product or the operation, and the queue of work is permanent rather than a backlog you are clearing once.
What does an in-house AI engineer actually cost in New Zealand?
More than the salary. Sentry AI advertises NZ$84,000 for a mid-level AI engineer, which is a real market anchor rather than an estimate. On top of that sit KiwiSaver, ACC levies, recruitment, hardware, model and tooling spend, and the manager time to direct the work. The number that matters is the fully loaded twelve-month cost against what actually ships in that time, and a first hire spends a meaningful part of year one ramping, because they are learning your business as well as building in it.
Is a $20,000 quote for AI automation normal?
It is a normal number and a meaningless one without a scope. The question that makes it answerable is what you own at the end. A twenty-thousand-dollar engagement that leaves you with one working workflow, the documentation to run it, and the ability to change it yourself is different from the same figure spent on a pilot that dies when the consultant leaves. Ask what happens on the day the engagement ends, and price the two answers separately.
When is a consultant genuinely the wrong answer?
When the work is continuous and close to the product. If AI is how your product works rather than how your business runs, the knowledge has to live inside the team, and paying an outside firm to hold it is a slow way to build a dependency. Also when nobody internally has the authority to make decisions, because a consultant with no decision-maker produces a document instead of a system.
Can you do both, a consultant and an internal hire?
That is the most common shape and usually the cheapest. An external team installs the operating system, ships the first two or three workflows, and writes down how it works. Someone internal, often a person who already knows the business rather than a new hire, owns and extends it. The failure mode of the pure consulting path is a system nobody inside can change. The failure mode of the pure hiring path is twelve months of a salary spent learning what the first engagement would have taught in six weeks.
How do the model costs compare between the two paths?
They are the same, which is the point people miss. Seats and tokens cost what they cost whether an employee or a consultant configures them. As at September 2026 a Claude Team seat lists at US$25 a month and Microsoft 365 Copilot at NZ$45.60 a user a month, both before tax, and neither price moves because of who set it up. The difference between the two paths is entirely the labour and the speed, so compare those and hold the licence line constant.
What is the honest case against hiring Sentry AI?
If your requirements are simple and stable, an off-the-shelf tool plus a few days of someone's attention will beat any engagement on cost. If your team already has an engineer with capacity and appetite, give them the six weeks first. And if nobody internally is going to own what gets built, do not start, because the most expensive outcome on this page is a working system with no owner. We would rather say that here than three months into an engagement.
Decide it on evidence, not on a quote
The free AI Opportunity Audit maps what your business already runs and what is worth building first. It takes about two minutes and it is the same map we would build in the first week of any engagement, so it is useful whichever of the three options you pick.
Ask an AI about this page
Opens the assistant with this page loaded: read it, summarise it, cite it.