Stop hiring for AI skills you should be buying as a system

If you are deciding between hiring an in-house AI specialist and bringing in a consultant, the honest answer for most UK SMBs is neither. Buy a system instead: working software, wired into your business, doing a defined job from the first week. A hire gives you the capacity to maybe build that system over the next year. A consultant gives you a plan for someone else to build. Buying the system removes the maybe and the someone else.

That is the kind of claim an AI agency would make, so let's do the maths properly.

"Hire for judgement. Buy the machinery. A hire gives you the capacity to maybe build the system next year. Buying the system deletes the maybe."

Dean Cookson, founder, Operosus

What does an in-house AI hire actually cost?

Start with the salary. The median advertised salary for an AI engineer in the UK is £90,000, based on job vacancies posted in the six months to June 2026. That is before employer's National Insurance, pension contributions, equipment, software and the management time it takes to direct someone working in a discipline you do not practise yourself.

Then there is the small matter of finding them. In the CIPD's Resourcing and Talent Planning report, 64% of organisations that tried to recruit said they had difficulty attracting candidates, with senior and skilled positions the hardest to fill. AI skills sit at the sharp end of that competition. You are bidding for the same shortlist as banks, scale-ups and consultancies, and they can pay over the median without blinking.

Suppose you beat the odds and land a good one. They still need to learn your business, choose a stack, prototype, test and deploy. Months of salary go out the door before anything useful runs. You have also created a single point of failure: when one person holds all the knowledge of how your automation works, their notice period becomes your biggest operational risk.

None of this means AI engineers are poor value. It means a full-time engineer is the wrong unit of purchase for a business that needs four or five specific jobs automated, not a research function.

Doesn't a consultant solve that?

A consultant solves a different problem. If you genuinely do not know where AI fits in your business, a few days of good advice is worth paying for. But look at what most engagements actually deliver: a readiness assessment, a maturity score, a roadmap, a workshop or two. Those are inputs. Not one of them answers an enquiry, drafts a tender response or chases an unpaid invoice.

The gap between "we know what to automate" and "it is automated and running" is precisely the gap the consulting model leaves open. You finish the engagement holding a document, plus the same build problem you started with, because somebody still has to make the thing. Often the recommendation at the back of the deck is, conveniently, that you hire.

Hire for judgement. Buy the machinery.

What does buying the system actually mean?

It means paying for an outcome rather than a person or a plan. A system, in this sense, is software with a job title: it drafts your tender responses, produces and publishes your content, runs your outbound email, qualifies and routes your leads. It is specified in plain English, it plugs into the tools you already use, and you can check whether it is doing the job the same way you would check on a member of staff.

This is how we build at Operosus. Bidwell drafts tender responses for UK SMBs. Contentwell runs content production. Emailwell handles outbound and lifecycle email. Each was built once, then hardened across real client use, which is the part that matters: when you buy a system that already runs elsewhere, the expensive trial and error happened on someone else's clock. Bespoke client systems get assembled from the same proven components rather than started from a blank page.

Three properties separate a system from a hire or a report:

1. It is specified against a job, not a person. "Produce a compliant first draft for every tender we shortlist" is checkable. "Lead our AI initiatives" is not.
2. The cost is a known number. A build price and a running cost you can put in a spreadsheet next to the £90,000 salary line and compare honestly (our guide to AI automation costs in the UK gives you the comparison framework).
3. It does not resign. The logic lives in software and documentation, not in one employee's head.

How do the three options compare?

That last cell deserves honesty. Buying the wrong system, or buying from a vendor who disappears, is a real risk. You manage it the same way you manage any supplier: insist the system runs on accounts you control, that your data stays in your own database, and that the spec is written in language you understand. If a supplier resists any of those, walk away. Our guide to choosing an AI agency gives you the full question list.

When should you actually hire?

Two situations. First, when AI is your product. If you are building machine-learning software to sell, you need engineers on the payroll, and nothing in this article applies to you.

Second, when the systems multiply. Once a business is running several automated systems, coordinating them becomes a genuine job: deciding what gets built next, watching the numbers, owning the supplier relationships. That is a worthwhile hire. But notice the order. It is an ownership role that comes after the systems exist, not a build role that comes before. The hire inherits working machinery on day one instead of a blank backlog and a vague mandate.

Most SMBs asking "consultant or in-house?" are nowhere near that point. They have a handful of processes eating staff time and a sense that AI should be able to help. For that business, recruiting into a market where nearly two thirds of employers report difficulty attracting candidates, at a £90,000 median salary, to eventually build something a specialist could deliver as a finished system, is the slow and expensive route dressed up as the safe one.

So what do you do on Monday?

Skip the question of who to employ and ask a sharper one: which three processes consume the most skilled time in your business, and what would "done by software" look like for each? Write those down in plain English. Then buy, or commission, the system that does the first one, with your data in your own accounts and a spec you can hold the supplier to. Rather build it yourself? We'll show you how, free, on do it yourself, and the same three-process list is the right place to start.

If the systems earn their keep and multiply, hire someone to own them later. They will walk into a business where the machinery already works, which is a far better first week than a job description that just says "AI".