Labour costs and tax pressure: where AI helps (and where it does not)

By Jason Sibley

A note before you read this: nothing here is tax or legal advice. GOV.UK figures are cited as context only, for accuracy — for anything specific to your business, speak to an accountant. This piece is about where AI genuinely reduces operational pressure around labour costs, not about how to structure your payroll or claim any allowance.

The problem

Since April 2025, the employer Class 1 National Insurance rate rose to 15%, the threshold at which employers start paying it dropped to £5,000 a year (down from £9,100), and the Employment Allowance — which reduces an eligible employer's annual NIC liability — rose to £10,500, with the previous £100,000 liability cap removed for most employers. Whatever the net effect for your specific payroll, the direction of travel for many small employers has been an increase in the cost of employing people, and the Office for National Statistics has been tracking the consequence directly: its own Vacancy Survey feedback "continues to suggest that smaller firms may not be recruiting because of increases in labour costs." ONS data for June to August 2026 shows vacancies at businesses with 1–9 employees fell 23.5% over the year — the largest annual decrease of any size band.

This is the real backdrop against which "should we use AI instead of hiring" conversations are actually happening in small UK businesses right now. It's a legitimate question. It's also one that gets muddled easily, because AI can absorb some kinds of workload growth and cannot touch others at all.

What people get wrong

They treat "AI instead of a hire" as a like-for-like swap. It rarely is. A new employee brings judgement, relationship continuity, and the ability to handle genuinely novel situations. AI absorbs repeatable, well-defined tasks. Framing the decision as "AI or a person" for a role that's mostly judgement-based sets AI up to fail a test it was never suited for.

They don't distinguish between labour cost pressure and workload pressure. Sometimes the real problem is that a role has become more expensive to fill (NIC changes, wage inflation, recruitment costs) even though the volume of work hasn't grown much — that's a cost problem AI in the right pocket of work can genuinely ease. Other times the workload itself has grown and needs a person's judgement regardless of cost — that's a scaling problem AI generally can't solve by itself.

They ignore the Employment Allowance conversation entirely because "we're too small for it to matter." It's genuinely worth checking with an accountant whether you're eligible and claiming what you're due before assuming a hire is unaffordable — this is a straightforward compliance question, not a controversial one, and it's separate from any AI decision.

They expect AI to fill a specific skills gap it can't fill. If the actual bottleneck is a licensed trade, a regulated service, or a role requiring in-person presence, no amount of AI adoption changes that. Be honest about which category your bottleneck falls into before spending time exploring an AI-shaped solution to a people-shaped problem.

They assume the ONS vacancy data is describing someone else's business. It's easy to read a statistic about "smaller firms not recruiting because of labour costs" as a description of businesses more financially stretched than yours. In practice, the data covers the entire 1–9 and 10–49 employee size bands (the categories most UK small businesses actually sit in), which makes it a fairly direct description of the pressure a large share of readers of this piece are currently under, not an abstract national trend happening elsewhere.

Why the numbers actually matter here

It's worth sitting with the scale of the recent change for a moment, without drawing any conclusion from it yet. The employer secondary NIC threshold (the point at which an employer starts paying National Insurance on an employee's earnings) fell from £9,100 a year to £5,000 a year from April 2025, while the rate itself rose from 13.8% to 15%. For a business with several employees, that's a materially different total cost of employment than it was before, even before accounting for the Employment Allowance offset. The allowance rise to £10,500, and the removal of the previous £100,000 liability cap for eligibility, partially offsets this for many small employers — which is exactly why "have you checked with your accountant whether you're claiming the right allowance" is a genuinely useful first question, entirely separate from any AI decision.

The ONS vacancy data adds a second layer to this: costs have not merely risen in the abstract: businesses are visibly responding by not recruiting, at a scale that shows up clearly in the size bands where individual small businesses actually sit. This matters for the AI conversation specifically because it means the "should we hire or use AI instead" question isn't a hypothetical one being had by a handful of early adopters — it's happening across a measurable share of the smallest employers in the country, right now.

A practical frame: cost pressure vs capacity pressure

If the pressure is cost (you could handle the workload with the team you have, but a new hire's total cost of salary, employer NIC, recruitment and onboarding time doesn't currently pencil out), look first at whether the actual constraint is a repeatable task eating hours that AI can take a genuine first pass at (see our piece on growing without hiring another pair of hands for the practical version of this). This is squarely where AI adoption reduces pressure without touching headcount at all.

If the pressure is capacity (the work itself has grown beyond what any amount of task automation can absorb, because it needs judgement, presence, or a licensed skill), AI can still help around the edges (faster admin, faster first drafts, less time on the repeatable third of the job), but it will not solve the underlying need for another person. In this case, the more useful question is how to make the next hire's economics work, which is a conversation for your accountant about current NIC thresholds, the Employment Allowance, and realistic total cost — not an AI question at all.

Most small businesses are living in some mix of both. The mistake is applying an AI-shaped answer to the capacity half of the problem, or accepting a "we just need to hire" answer for the part that's genuinely a repeatable-task cost problem.

A worked example, kept deliberately simple

Picture a business considering a part-time admin hire, mainly to handle invoicing, basic customer replies, and diary management. Before assuming the answer is "hire" or "don't hire," it's worth separating the question properly: how much of that role is genuinely repeatable-task work that an AI-assisted workflow could absorb (invoice reminders, first-draft replies to routine enquiries), versus how much requires the judgement and continuity a person brings (handling an upset client, making scheduling trade-offs that need real context)?

If the honest answer is that most of the role is the repeatable kind, the right move might be building that AI-assisted workflow into how the existing team works, rather than a new hire at all — which sidesteps the current NIC and cost pressure entirely, because there's no new employer cost to weigh against it. If the honest answer is that the role genuinely needs a person's ongoing judgement, then the right move is having the accountant conversation about what that hire actually costs under current thresholds and allowances, rather than trying to talk yourself out of it with an AI tool that was never going to solve a people-shaped problem.

The one decision to make this week

Don't decide "should we use AI to avoid hiring." Decide: for the specific role or workload we're stuck on, is the honest constraint cost or capacity — and have we actually checked our current employer NIC position and Employment Allowance eligibility with an accountant before assuming the answer is no either way?

Separating those two questions (the AI question and the accountant question) stops them contaminating each other. You might genuinely need both: an accountant confirming what a hire would actually cost given current thresholds and allowances, and a properly scoped AI habit absorbing the repeatable third of the work regardless of what you decide about headcount.

A soft word before you go

Again — not tax advice, and for the actual numbers, talk to an accountant. If the AI half of this is what you want help thinking through, My AI Apprentice's SMB AI sessions are built for exactly that kind of practical sorting exercise. No obligation — the frame above works whether or not you bring us in.