Picture an operations director in Leeds with a job advert half written. Her team is drowning in order queries, supplier chasers and month-end reconciliations, so the obvious fix is one more person. Before that advert goes live, though, a sound AI workforce strategy asks a different question first: which parts of this job could an AI worker carry, and which parts genuinely need a human?
The short answer is this. Start with the work, not the tools. Break the role into tasks, then score each task by how often it happens, how rule-bound it is and how much damage a mistake would cause. High-volume, rule-bound, low-risk tasks go to AI first; judgement-heavy or high-risk tasks stay with people, with AI in a supporting role.
That sounds tidy on paper. In practice, most businesses skip the mapping and jump straight to a tool, which explains a great deal about why so many AI projects stall.
Why the hiring decision now comes with a second question
The economics of a new hire have shifted. Since April 2025, employers in Britain pay secondary Class 1 National Insurance at 15%, and the threshold at which they start paying fell to £5,000 per employee, as the explanatory notes to the 2025 NICs Act set out. For a small firm, each extra salary now costs noticeably more than it did two years ago.
At the same time, AI use has become ordinary. According to the ONS review of AI in UK businesses, published in July 2026, about 35% of businesses with ten or more employees used AI by June 2026, up from about 12% in September 2023. Yet only 10% of those adopters described their use as extensive.
So adoption is wide but shallow. Many firms have a chatbot subscription or a writing assistant, while the actual workflows of the business run exactly as they did before. Consequently, the decision about the next operations hire rarely includes a serious look at what AI could take on.
The hiring market is already moving
Labour market data hints at the same shift. In an August 2026 Bank of England staff blog on AI exposure and hiring, researchers found that online adverts for the most AI-exposed occupations fell 15% over three years, against 6% for the least exposed. Customer service adverts fell 23% and administrative adverts 22%.
However, the authors are careful. They write that confident attribution to AI “remains premature”, because remote working, pandemic over-hiring and cost pressures all hit the same period. The honest reading is that employers are hiring fewer junior office roles, and AI may be one reason among several.
What an AI workforce strategy actually is
An AI workforce strategy is a plan for which work your business hands to AI workers, in what order, under whose supervision and with what measures of success. It treats AI as a set of roles with defined duties, rather than as a piece of software someone happens to open.
That framing matters because AI agents now do more than answer questions. They can read an email, check a CRM record, draft a reply, update a ticket and pass the case to a person when something looks wrong. Our guide to how agents differ from chatbots covers that distinction in more depth.
Most organisations remain early on this road. McKinsey’s State of AI survey for 2025 found that 62% of respondents’ organisations were at least experimenting with AI agents. Only 23% said they were scaling an agentic system anywhere in the enterprise, and no more than 10% were scaling agents in any single business function.
Notably, the same survey found that the small group of high performers were nearly three times as likely as others to fundamentally redesign their workflows. In other words, the winners changed how work flows, rather than bolting AI onto old processes.
Map the work by volume, rules and risk
The practical heart of any AI workforce strategy is a task map. You take the role you were about to hire for, or the team that feels overloaded, and list every recurring task. Then you score each one on three dimensions.
- Volume. How often does the task happen each week? A job that occurs 300 times matters more than one that occurs twice, because small savings compound.
- Rules. How predictable is the right answer? If a competent new starter could follow a written procedure, the task is rule-bound. If it needs experience, negotiation or taste, it is judgement-heavy.
- Risk. What happens if the task goes wrong? A mistyped delivery note is annoying. A wrong refund, a missed regulatory deadline or a careless reply to a vulnerable customer is serious.
- Data access. Does the information live in systems an AI worker could reach, such as your CRM, helpdesk or accounting tool? Or does it sit in someone’s head and a shared inbox?
- Exception rate. Roughly what share of cases fall outside the normal path? This tells you how much human escalation you need to design in.
Score volume, rules and risk from one to five. Then plot the results. Anything high on volume, high on rules and low on risk sits in your first wave. Anything high on risk stays with people, whatever its volume.
A worked example with real numbers
Take an illustrative wholesale business with a three-person operations team. The director wants to hire a fourth person at a salary of £28,000. Before posting the advert, she maps a typical week.
| Task | Weekly volume | Rules (1–5) | Risk (1–5) | Verdict |
|---|---|---|---|---|
| “Where is my order?” emails | 240 | 5 | 1 | AI first wave |
| Logging new orders from email into the CRM | 180 | 4 | 2 | AI first wave, human spot checks |
| Chasing overdue supplier confirmations | 60 | 4 | 2 | AI first wave |
| Credit note approvals | 25 | 3 | 4 | AI drafts, human approves |
| Handling complaints from key accounts | 10 | 1 | 5 | Human only |
If each order-status email takes three minutes, those 240 emails consume 12 hours a week. Logging orders at four minutes each adds another 12 hours. Supplier chasers at five minutes add five more. Together, that is roughly 29 hours, or most of the new post she planned to create.
The verdict is not “fire nobody, hire nobody”. Instead, the map shows that the pressure comes mainly from repetitive, rule-bound work. An AI worker could absorb much of it, while the existing team spends more time on credit decisions and key accounts, where their judgement actually earns money.
Sequencing your AI workforce strategy
Once the map exists, the order of work becomes clearer. A sensible AI workforce strategy runs in waves, each with a defined owner and a defined measure.
Wave one: high volume, low risk. Customer status questions, appointment booking, lead qualification and data entry between systems usually sit here. Automatic lead handling is a common starting point, as our piece on following up with leads automatically explains.
Wave two: AI drafts, humans decide. Refund recommendations, credit notes and contract summaries fit this middle tier. The AI worker prepares the case and a named person signs it off. Over time, you can loosen approval thresholds as accuracy data builds up.
Wave three: cross-system work. Management reporting, reconciliation and pipeline hygiene involve several systems at once. They deliver large savings, yet they depend on clean data and good access controls, so they come later.
Measure each wave before starting the next
Each wave needs a baseline. Record how long tasks take today, how many errors occur and how quickly customers get answers. Then compare after four to six weeks. Without a baseline, you will only have anecdotes, and anecdotes do not survive a budget meeting.
The traps that sink AI workforce plans
Analysts expect many projects to fail. In a June 2025 forecast, Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. Gartner also warned of “agent washing”, where vendors rebrand existing chatbots or RPA tools as agents.
Those three causes map neatly onto mistakes that a task map prevents. Below are the most common ones in small and mid-sized firms.
- Starting with the tool. A team buys a platform, then hunts for something to automate. The map reverses that order.
- Picking the most impressive task. Leaders often choose a complex, high-risk process for the pilot because it makes a good story. Choose the boring, high-volume one instead.
- Ignoring escalation. Every AI worker needs a clear route to a human. If nobody owns the exceptions, customers fall through the gap.
- Hidden, personal workflows. Staff often build private prompts that only they understand. We explored that risk in our article on workflows trapped in one person’s chat history.
- No owner. An AI worker without a named human supervisor drifts. Somebody must review its output, tune it and answer for it.
Who should own the plan
In a smaller business, the owner is usually the operations director or the founder. Larger firms may split the role between operations, IT and a data protection lead. Whoever holds it needs authority over process changes, not just a software budget.
Training matters too. The ONS found that only 11% of businesses with ten or more employees said more than half their workforce had received AI-related training. Meanwhile, about 40% of medium and large firms responded to AI by training existing staff. Your people will supervise these AI workers, so they need to understand what they are supervising.
Some firms build this capability in-house. Others work with partners such as Riophany, which design, connect and then run AI workers across CRM, finance and customer operations, with staff keeping control of decisions. Either route works, provided the task map comes first.
The advert that never went out
Back in Leeds, our illustrative director might still hire that fourth person. However, she would now write a different advert. Instead of “operations assistant, high-volume admin”, the role might read “customer operations lead, supervising automated order handling and owning key account escalations”.
That is the quiet shift that a good AI workforce strategy produces. It rarely removes the need for people. Rather, it changes what you hire them to do, moving them from carrying the volume to deciding what happens when the volume does something unexpected.
Frequently asked questions
What is an AI workforce strategy?
An AI workforce strategy is a plan for which tasks a business hands to AI workers, in what order, with what supervision and with what success measures. It starts from a map of the work rather than a choice of software. Good plans treat AI workers as roles with defined duties and a named human owner.
Which tasks should a small business hand to AI first?
Start with tasks that are high in volume, follow clear rules and carry low risk if something goes wrong. Typical examples include order status questions, appointment booking, lead qualification and copying data between systems. Keep high-risk or judgement-heavy work, such as complaints from key accounts, with people.
Does AI mean I should stop hiring operations staff?
Not necessarily. Mapping the work often shows that AI can absorb repetitive volume, while people remain essential for approvals, exceptions and relationships. Many businesses still hire, but they hire for supervision and judgement rather than manual processing.
How long before an AI worker shows results?
For a well-scoped, high-volume task, most teams can measure a difference within four to six weeks, provided they record a baseline first. Measure handling time, error rates and customer response times before and after. Cross-system work such as reporting or reconciliation usually takes longer because it depends on data quality and access controls.