Why the next hire might not be a person.
Most ops hires in a £1-20m business are doing rules-based work. An agent now does that work for a fraction of a loaded salary. Here's the maths and the limits.
When a workflow outgrows the person doing it, the default in a UK SME is to hire a junior. That has been the right move for 20 years. It is no longer the right default, and the gap between defaults will define which businesses are leaner three years from now.
Read the job spec honestly
Pull the last three ops job specs you posted. Bookkeeping assistant, customer service rep, operations coordinator. Now count: how much of each role is sequenced, rules-based work? Sorting tickets by topic. Reconciling returns across three systems. Pulling numbers from five sources into a Monday slide. Onboarding suppliers on a 12-step checklist. In most £1-20m businesses, the honest answer is most of it.
McKinsey's State of AI Trust 2026 calls the space between agent capability and SME adoption the "agentic gap": the work is automatable, the buyers have not noticed yet. Deloitte's State of AI in the Enterprise 2026 puts customer support, supply chain, and knowledge management at the top of the functional impact list. That is the work in most ops job specs.
The maths on one role
A DTC brand hires a customer service coordinator on a £32,000 salary. Fully loaded with national insurance, pension, software, holiday cover, and a desk, the real number sits closer to £42,000. The role is mostly triage, shipping queries, and returns, with a slice of judgement work on top.
The shape matters more than the line items. Three small agents own the rules-based work. Two or three hours a week of senior time covers the judgement. The gap compounds. Apply it to four roles and there is a £120k margin point that was not there before. Real payback ranges by workflow type sit between two and fourteen months.
Where agents quietly go wrong
Agents fail when they are handed judgement work dressed as routing. Stanford's Digital Economy Lab analysed 51 enterprise AI deployments in its March 2026 Enterprise AI Playbook and flagged the same failure mode: dashboards stay clean, reports keep flowing, decision quality decays underneath. Deloitte's 2026 enterprise survey echoes it on governance: weak oversight is the leading cause of stalled value.
The example is the boring one. Software flags a revenue dip as significant when an experienced operator would know it is seasonal. Software suppresses an outlier that does not fit its heuristics, and leadership gets a tidier but materially less accurate picture. The thing that makes a good operations manager is that they spot what does not fit the rules. That does not transfer. Hand agents the rules-based work, the synthesis, and the routing. Keep contextual judgement, exception filtering, and weak-signal detection with a person.
Lean teams are the new operating model
The output of a 15-person team is now reachable with five people and a stack of agents. BCG's AI Radar 2026 reports 90% of CEOs expect agentic AI to deliver measurable returns inside two years, and frames agents explicitly as workforce, not tooling. McKinsey's State of AI 2025 found the productivity gains concentrate in firms that redrew roles rather than bolted AI onto existing ones. The next workflow that gets too big does not automatically become a hire.
The shape that wins is small teams running agents, not no teams.
Bain's Technology Report 2026 found 90% of leaders say their data foundations are not fit to scale. The UK Government's January 2026 AI Adoption Research backs it up: 71% of firms have no clear use case and 60% cite limited AI skills as the blocker. The failure mode is not the model, it is ownership.
Where the maths still says hire a person
Three categories where a human remains the right answer for years. The British Chambers of Commerce and Atos found in March 2026 that 54% of UK firms are actively using AI, almost entirely as support for existing roles rather than substitution.
Senior judgement roles.
Heads of department, finance directors, commercial directors. Exception handling, relationships, the long memory of what went wrong.
Customer-facing trust.
Account management for your biggest customer. Sensitive complaints. The bespoke side of B2B sales.
Genuinely creative.
The first creative pass on a campaign. The editorial line on a launch. Agents help draft; they don't yet originate at a publishable level.
What the next hire actually looks like
What changes when agents own the rules-based slice is the role you wanted to hire for. The next req gets sharper. Less doing, more deciding. More senior, harder to find. But the right role.
Before you post the next ops job
- If most of the role is rules-based work with structured inputs, you are hiring for work an agent should do.
- The wage bill stops being where you find your margin. Redrawn roles are.
- The right shape for a £1-20m business in 2026 is fewer, more senior people on top of agents that do what a junior used to do.
Frequently asked questions
When should a UK SME hire an agent instead of a person?
When most of the role is rules-based work with structured inputs and a known output. Ticket triage, returns processing, supplier onboarding, weekly reporting. An agent handles that reliably for a fraction of a fully loaded salary. The judgement, relationship, and exception-handling slice stays with a person.
How much does an AI agent cost compared to an ops hire?
A fully loaded UK ops hire on a £32,000 salary costs around £42,000 a year once national insurance, pension, software, and cover are included. The equivalent set of agents covering triage, shipping queries, and returns runs at roughly £11,500 a year including monitoring and tuning. The gap is about £30,000 per role per year.
What roles should still be hired as people in 2026?
Three categories. Senior judgement roles (heads of department, finance directors, commercial directors). Customer-facing trust roles where the relationship is the asset. Genuinely creative work where the founder's voice or editorial line matters. Agents support these roles. They do not replace them.
Why do most SME AI projects fail?
Ownership, not technology. Bain's Technology Report 2026 found 90% of leaders say their data foundations are not fit to scale. The UK Government's January 2026 AI Adoption Research found 71% of firms had no clear use case and 60% cited limited AI skills. Without a named owner and a clear workflow, pilots quietly stall.
What is the agent-plus-champion model?
A central team owns the platform, governance, and standards. Inside each function, a named champion knows the workflows and the tools and translates between the two. McKinsey and BCG describe this as the dominant pattern in successful enterprise AI deployments. It scales down to SMEs: in a 12-person business, the founder is the central team and one operator per function is the champion.
Wondering whether your next hire should be a person.
The Clerq Diagnostic looks at the role you're about to post against the agents that could do the rules-based slice, and prices the build. From £2,500. 20-minute intro call first.