Build, buy or borrow: staffing AI capability you cannot yet justify full-time
6 minute read. Updated 2026-08-08.
For AI capability you can't yet justify full-time, borrow before you build. A contractor at 2 days a week can cost less than an equivalent permanent salary, using advertised UK day rates and salaries, while letting you learn what the work actually requires. Build once the need is continuous and full-time, when the arithmetic on day rates flips in favour of a permanent hire. Buy only the commoditised parts.
Three routes, one decision
When you need a capability today, you're choosing between three ways to get it: hire someone permanently, buy something off the shelf, or borrow a specialist for a defined period. Each solves a different problem. Confusing them is how companies overspend on a permanent hire for a 6-month need, or underinvest in a permanent capability by treating it as a short project indefinitely.
The decision hinges on one question: is this need continuous, or is it a phase? Cost, speed and risk all follow from the answer, not the other way round.
The figures below are advertised UK rates, not agreed ones, and the gap between what's advertised and what's actually paid tends to widen further up each range. Use them as a starting point for the arithmetic, not as a quote.
What building costs, in full
A permanent Machine Learning Engineer carries an advertised UK median salary of £76,000, using contract rates and salaries quoted in job advertisements over the six months to August 2026, published by IT Jobs Watch. A broader Artificial Intelligence hire sits at £71,000. Neither figure includes employer pension contributions, National Insurance, bonus or equity, all of which sit on top of the headline number.
Recruitment for a role like this rarely closes inside 2 months, and a mis-hire costs you the salary, the notice period, and the months of drift while the wrong person tries to do the job. None of that shows up in the salary line, but it's real money and real delay.
What borrowing costs, in full
A contractor at the median advertised Machine Learning Engineer day rate of £575, working 2 days a week for a year, comes to roughly 104 days, £59,800. That sits below the £76,000 permanent median before you've added a penny of employer cost, and it buys you the ability to stop at short notice.
The comparison isn't free of complications. Since April 2021, medium and large private-sector clients, not the contractor, determine whether an engagement is taxed as employment under the off-payroll working rules, known as IR35. Where an engagement is judged inside IR35, deductions through an umbrella company reduce what the specialist actually takes home, though no percentage is stated here. It's a cost the specialist absorbs, not you, but it shapes who is willing to work with you, and at what rate.
A day rate also includes no holiday, no sick pay, no pension and no paid time between engagements, all of which are priced into how a contractor sets their rate. The comparison above is fair on total cost, but it isn't a comparison of like-for-like security for the person doing the work.
What buying solves, and what it quietly doesn't
Buying a platform or a licensed model solves a narrow, well-defined problem: you need a capability that already exists as a product, and you don't want to build it. It's usually the right call for infrastructure that isn't your differentiator, such as a vector database or a model API.
What buying doesn't solve is who configures it, who checks its output against your data, and who owns the outcome when it's wrong. Companies that buy a platform and assume the job is done end up needing to borrow or build the missing role 6 months later, having paid for the platform and the missing role separately.
The call I'd make
For a capability you cannot yet justify full-time, borrow first. A fractional or contract specialist lets you find out what the work actually requires before you commit to a permanent cost base. If the need turns out to be continuous, converting a working relationship into a permanent hire, or simply extending the contract, is straightforward. Unwinding a permanent hire you didn't need is not.
Buy the parts of the problem that are genuinely commoditised, and use the borrowed specialist to tell you which parts those actually are. Most companies discover that fewer of their AI problems are commodity problems than the vendor's sales deck suggested.
When to build immediately instead
The trade-off runs the other way once the need is continuous and full-time. Take MLOps: the advertised median contract rate is £575 a day. At 5 days a week across a working year of roughly 230 days, that comes to £132,250, well above the median permanent salary of £87,500. If you genuinely need someone on this every working day for the foreseeable future, hiring is the cheaper route, not just the steadier one.
The mistake is applying the flexible logic to a role that was never actually flexible, and keeping someone on a rolling contract for 2 years because nobody wanted to make the hiring decision.
Revisit the decision, don't set it once
Whichever route you choose, put a date in the diary to review it: 90 days out for a borrowed specialist, 6 months for a permanent hire. The market for AI skills is moving fast enough that a decision that was right when you made it can be wrong within 2 quarters, in either direction.
A specialist you brought in for a defined problem, who's now quietly running 4 other things nobody wrote down, is a sign the borrow has become a de facto build. Formalise it, or scope it back down; don't let it drift.
What to do about it
- Work out whether the need is continuous or a phase before choosing a route.
- Do the day-rate arithmetic for your actual hours per week, not the headline rate alone.
- Don't buy a platform without also funding who configures and owns it.
- Borrow while you're still finding out what the work requires.
- Build once the role is full-time, ongoing and central to the product.
- Put a review date on the decision, not just a start date.
Questions people also ask
Is a fractional AI specialist always cheaper than a permanent hire?
Not always. It depends on how many days a week you actually need. At 2 days a week the contract route is usually cheaper than an equivalent permanent salary; at 5 days a week across a full year the arithmetic can flip, because a day rate is priced for flexibility, not full-time occupancy. Work it out for your specific hours before assuming either direction.
Who decides IR35 status?
Since April 2021, medium and large private-sector clients make the determination, not the contractor. Get this assessed properly before the engagement starts, because getting it wrong exposes you, not just the specialist, and the tax treatment materially changes what a given day rate means in practice.
What should I buy versus borrow someone to build?
Buy infrastructure that isn't your differentiator, such as model APIs or vector databases. Borrow someone to work out which parts of your actual problem are commodity and which aren't, then to configure and own whatever you've bought. Buying a platform without funding someone to run it just delays the cost, it doesn't remove it.
How long should a borrowed specialist's engagement run before I decide whether to build the role permanently?
There's no fixed answer in the figures, but 90 days is a reasonable first checkpoint. By then you should know whether the demand on the role is continuous or was a one-off project, which is the only question that decides whether build, buy or borrow was right.
Where the figures come from
Every rate and salary quoted in this article is a median or percentile of figures advertised in UK job postings over the six months to 8 August 2026. They are not rates paid, and the gap widens at the top of a range.
- IT Jobs Watch, UK contract rates, 6 months to 8 August 2026, read 2026-08-08.
- IT Jobs Watch, UK permanent salaries, 6 months to 8 August 2026, read 2026-08-08.