For most small firms, not in the next few years. AI takes over tasks, not whole jobs. Roles built mainly around one repeatable task, such as data entry, transcription or first-line email replies, will shrink, and you may need fewer new hires as you grow. Staff whose work mixes judgement, client contact and varied tasks will mostly see their jobs change.
That's the middle answer, and it has two uncomfortable edges: some roles really do shrink, and what happens to the people in them is your decision, not the software's. You can't control the technology, but you can assess each role before it arrives and decide in advance what the freed time is for. The examples come from an illustrative eight-person bookkeeping firm.
What the evidence says so far
The research is early and mostly about averages, but it points the same way.
- Employers expect churn, not collapse. The World Economic Forum's Future of Jobs Report 2025, based on a survey of more than 1,000 employers, expects 170 million jobs to be created and 92 million displaced worldwide by 2030 from all the trends it tracks, technology included: a net gain of 78 million. Data entry clerks, bank tellers and administrative assistants are among the roles it expects to decline fastest. About 40% of the employers surveyed expected to reduce staff where AI can automate tasks, while 77% planned to upskill their existing workers.
- So far, little change in pay or hours. A large study of workers in occupations exposed to AI chatbots, by economists Anders Humlum and Emilie Vestergaard, found no measurable effect on earnings or recorded hours two years after the chatbots arrived. What did change was the mix of tasks within jobs.
- The least experienced gain most. A study of 5,172 customer support agents by Erik Brynjolfsson, Danielle Li and Lindsey Raymond found an AI assistant raised issues resolved per hour by 15% on average, with the biggest gains for newer, less experienced staff and little for the most experienced.
Two caveats. First, AI agents that act on their own, rather than drafting for a person, are newer than this research, and could move faster. Second, averages hide your firm. A business with one person doing nothing but data entry will feel the change more sharply than any headline statistic suggests.
The roles most exposed in any small firm
Whatever your sector, the same handful of job shapes carry most of the exposure, because each is built around one task that AI already does well. What stays human in each is usually the part that made the person valuable in the first place.
- Data entry and document processing. Keying invoices, receipts, timesheets and forms. Invoice capture and auto-categorisation now do much of this. What stays: chasing what's missing, fixing what doesn't match.
- Transcription and minute-taking. AI note-takers produce a summary within minutes of a call ending. What stays: checking it records what was actually agreed.
- First-line email and chat replies. Routine questions answered from approved content. What stays: complaints, upset customers and anything unusual.
- Phone answering and booking. AI receptionists can take messages and book slots. What stays: the calls where someone needs a person.
- Junior research and first drafts. Background research, first versions of letters and reports. What stays: judging whether the draft is right, which is also how juniors learn.
If one person in your firm spends most of their week on one of these, that's the role to plan around first. If the tasks are spread thinly across everyone, nobody's job is at risk; everybody's week changes a little.
The same shapes turn up under other job titles. Take the receptionist at a three-vet practice (an illustration): perhaps half the week goes on phone bookings and reminder calls, a quarter on payments and pet-insurance forms at the desk, and a quarter on reassuring worried owners in the waiting room. An AI phone line and automated reminders could take much of the first half. The desk work and the waiting room stay human. That's a job that changes shape, probably towards more time with clients face to face, rather than one that disappears. The real planning question is whether the practice wants that extra desk time, or would rather not replace the role when the receptionist eventually leaves.
Split each role into tasks before you worry
"Will AI replace the bookkeeper?" is too big a question to answer. "Which of the bookkeeper's tasks can AI do, and how many hours are they?" is answerable in an afternoon. For each role, list the tasks and roughly how many hours a week each takes, then mark each one:
- A: AI can do most of it, and a person reviews the result.
- H: AI helps, but the person still does the work.
- N: no meaningful help, such as a client meeting or a judgement call.
ROLE: ______________________ Contracted hours per week: ____
Task | Hrs/wk | A / H / N | What AI would do; what the person checks
-----------------------------|--------|-----------|------------------------------------------
| | |
| | |
Share of hours marked A: ____ %My rule of thumb for reading the result: under 30% of hours in A, the job will feel as if it got a better tool. Between 30% and 60%, the job changes substantially and frees real capacity. Over 60%, the role as it stands is at risk, and you should plan what that person does next now, not when the software arrives.
The middle band is the one most owners land in, and it's worth seeing one before the bookkeeping sheet below. Picture a garden centre's 30-hour marketing assistant: eight hours a week writing product descriptions and five on social captions (both A), three answering website messages (A), three on the newsletter (H), six photographing new stock and five on in-store displays (both N). That's 16 of 30 hours in A, about 53%. Her job changes a lot, but nothing about it disappears: the photography and displays are what customers see, and the freed hours can go into more of both, or into seasonal events the centre never had time to promote.
Filled in for the bookkeeping firm's data-entry assistant, the sheet might look like this (illustrative):
ROLE: Data-entry assistant Contracted hours per week: 35
Task | Hrs/wk | A / H / N | What AI would do; what the person checks
-----------------------------|--------|-----------|------------------------------------------
Keying receipts and invoices | 12 | A | Capture reads them; she checks supplier and total
Coding bank transactions | 9 | A | Software suggests codes; she fixes the odd ones
Filing client documents | 4 | A | Auto-filed by client; she spot-checks folders
Chasing missing paperwork | 5 | H | AI drafts chasers; she decides who and when
Answering client calls | 3 | N | Clients want someone who knows their file
Month-end admin for seniors | 2 | H | AI drafts checklists; she runs them
Share of hours marked A: 25 of 35 = 71%Two lines on that sheet matter more than the total. Chasing paperwork and answering client calls are the parts clients value, and they grow as the firm takes on more clients, so they're where a redesigned role starts.
An eight-person bookkeeping firm, role by role (illustrative)
| Role | Main tasks | Hours in A | Likely outcome |
|---|---|---|---|
| Data-entry assistant (1) | Keying receipts, coding transactions, filing documents | About 70% | Role at risk as it stands; redeploy |
| Bookkeepers (3) | Reconciliations, month-end, client queries, calls | 35% to 45% | Job changes; each can handle more clients |
| Payroll administrator (1) | Payroll runs, starters and leavers, queries | About 25% | Better tools, same job |
| Client managers (2) | Advice, reviews, onboarding, relationships | 15% to 20% | More time for advisory work |
| Owner | Management, sales, final review | About 20% | Faster reporting and marketing |
The driver here is mostly software the firm already uses. Intuit has added AI agents to QuickBooks Online that categorise transactions and help reconcile accounts, and Xero's JAX is aimed at the same routine work. As those features improve, the keying and coding that fill the data-entry assistant's week shrink, whether or not the firm buys anything new.
That's the honest case. The data-entry role isn't gone this year, but its core tasks are the ones software is built to take. The bookkeepers' roles are safer than they look, because the reconciliation that AI speeds up is only part of a week that includes explaining numbers to clients and sorting out the exceptions nobody's software handles.
The more likely outcome: slower hiring, not redundancies
Suppose each of the three bookkeepers frees six hours a week once the AI features settle in. That's 18 hours, roughly half a full-time role. The firm can take on perhaps 15 more clients before it needs its next hire, or stop the month-end overtime everyone hates.
In most small firms this is what "AI replacing people" actually looks like: the vacancy you don't advertise, the temp you don't book for the busy season. Before you recruit, it's worth checking whether AI covers the gap; AI or a new hire sets out that comparison. And how much time AI can really save each week helps you put realistic numbers against each task.
One trap to avoid: if AI does all the junior tasks, juniors have nothing to learn on. Firms that stop hiring juniors entirely can find, a few years later, that nobody is ready to become a senior. Keep a training route, even a smaller one.
The gentlest version of this is a role that ends on its own schedule. Consider an illustrative two-dentist practice whose part-time secretary, who types up referral letters from the dentists' dictation, retires in eight months. The practice decides not to replace the role, and uses those months to move the typing to an AI transcription tool whose terms it has checked for patient data, with the secretary checking every letter while she's still there. Her knowledge of how each dentist phrases things and which consultants want which details goes into the letter templates. When she leaves, the dentists check their own letters in a few minutes each, and nobody has had to lose a job for the saving to happen.
If a role really does shrink
You'll usually see it coming months ahead. The backlog that used to build up at month-end stops appearing. The person starts asking for more work, or quietly stretches what's left. A software update switches on a feature that does a large part of their week. When two of those happen together, have the conversation early, while there's still time to plan a new shape for the job rather than react to an empty one.
In the bookkeeping firm it might show up like this. After an update to the accounting software's auto-categorisation, the pile of uncoded transactions waiting for the data-entry assistant at month-end falls from around 400 to about 60 over three months, and she starts asking the bookkeepers whether there's anything she can take off them. That's two signs at once. The right response is a conversation that week, starting from the redeployed week described below, rather than a quiet hope that the work comes back.
Work through the options in this order.
- Redeploy. The data-entry assistant knows the clients' paperwork better than anyone. Tasks that grow as AI arrives, such as chasing missing client records, onboarding new clients and checking the AI's categorisations, are a natural fit. In the illustration, a redeployed week might be 12 hours reviewing AI-coded transactions and fixing exceptions, 10 hours on new-client onboarding, 8 hours chasing missing records before deadlines and 5 hours of document filing the software still can't sort. The job is different, but it's a full week of useful work.
- Retrain. Support a bookkeeping or payroll qualification so the person can move into a role with more judgement in it.
- Change hours by agreement, if the person wants that.
- Redundancy, as a last resort. Follow a fair process. The rules on consultation, selection and notice depend on the employment law that applies to you, so get advice from an employment adviser before you announce anything or change anyone's duties.
Redesigning job roles once AI handles routine tasks covers how to rewrite a role so it's better than before, rather than a leftover.
What to tell your team, and when
Tell them before the tools arrive, not after the rumours start. Say what you know, what you don't, and what you've decided. If you've decided no one loses their job because of this year's AI projects, say so. If you haven't decided, don't promise it.
We're going to start using AI for [TASKS]. Here's what that means:
- The aim is [a faster month-end / taking on more clients without overtime].
- Some tasks will be done by software, mainly [data entry, first drafts].
- [Only if true:] Nobody's job is being cut because of this. If a role
changes a lot, I'll talk to that person first and we'll plan it together.
- You'll get time to learn the tools: [PLAN].
- If you're worried, or you have ideas, come to me directly
or raise it at [MEETING].Here's a realistic way it goes wrong. An owner switches the AI features on over a weekend and mentions them in passing on Monday as "a way to make us more efficient". The data-entry assistant hears "efficient" as "cheaper without me" and starts looking for another job; the firm finds out two months later when she hands in her notice, just as it needs her to check the AI's coding. Compare the template above, filled in and sent a fortnight before anything is switched on (illustrative wording):
We're going to start using the AI features in our accounting software for coding transactions and reading receipts. The aim is to get month-end done without the overtime. Some of the keying will now be done by software. Nobody's job is being cut because of this. The role that changes most is data entry, and I've already talked that through with the person in it. Everyone gets two hours on Friday afternoons next month to learn the tools, and if you're worried or have ideas, come and see me or raise it at Monday's team meeting.
Prepare for the question that comes straight after, usually from someone who works hard: "If I get faster, do I just get more work?" The honest answer depends on your plan, so have one. In the bookkeeping firm it might be: "Partly, yes. The aim is to take on more clients without anyone doing overtime at month-end, so some of the time comes back as new clients. The overtime stops first, and I'll show you the numbers each quarter." That's a better answer than a vague reassurance, because staff can check it against what actually happens.
Expect the quietest people to be the most worried. How to talk to staff who fear AI will take their job has more on those conversations.
Where people become more valuable, not less
As AI takes over drafting and keying, four kinds of work grow in value in a small firm. Checking: someone has to own the quality of what AI produces, and that person needs to know the work. Relationships: clients still want a named person who understands their business and picks up the phone. Exceptions: the odd transaction, the unusual client, the case the rules don't cover, where AI is weakest. Context: knowing which client is sensitive about fees, or which supplier always invoices late, is the knowledge AI doesn't have unless someone gives it.
The people who learn to use AI well, and to catch it when it's wrong, get more done than they did before. The customer support study found the biggest gains among less experienced staff. For a small firm, that's the more useful way to frame the question: not which employees AI replaces, but which of them you'll help get better at their jobs with it.
Follow-up questions owners ask
Can I reduce staff hours because AI saves time?
You can offer it, but a change to contracted hours normally needs the employee's agreement or a proper consultation, and the rules depend on the employment law that applies to you. Talk to an employment adviser before proposing anything. In practice, most small firms use freed time to take on more work or stop overtime, which is simpler and keeps goodwill.
Should I stop hiring juniors if AI does their tasks?
Think carefully before you do. Junior roles are where people learn the work that senior staff later check, and AI does exactly the tasks juniors used to learn on. If you stop hiring them, you may have nobody to promote in five years. Many firms keep a smaller junior intake and redesign the role around checking AI output and client contact.
Will clients accept AI doing some of the work?
Most accept it for drafting and admin as long as a qualified person reviews the result and their data is handled properly. Problems arise when clients discover it by accident or feel they're paying people rates for automated work. Be open about where AI helps, keep a named person accountable for every piece of work, and check your client contracts for any restrictions.
Further reads
- AI Change Management for Small Teams: A Practical Plan — A practical plan for bringing the team through the change.
- How to Handle Staff Who Over-Rely on AI — The opposite risk: staff trusting AI too much.
- Why AI Saves Time but Not Money, and How to Capture the Gain — Why freed hours don't cut costs unless you plan for it.
- How to Train Staff to Use AI in a Small Business — Help the people whose jobs change most.
- Should You Charge Clients Less When AI Speeds Up Your Work? — What happens to your pricing when work gets faster.
- How to Test Job Candidates' AI Skills in an Interview — Hire for the new version of each role.
- How to Get Your Staff to Actually Use AI Tools — Diagnose low AI use, fix the six usual blockers, and see how a tutoring agency went from 3 regular users out of 12 seats to 9.
- AI Myths That Stop Small Businesses Getting Started — Ten beliefs that keep small firms from trying AI, what is actually true about each, and a small first step that tests it within a week.
- How to Get Staff Buy-In When You Introduce AI — A veterinary practice introduces AI with its team rather than to it: one-to-ones, a team-chosen first job, a sceptic's veto and a written time-dividend deal.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: World Economic Forum, Future of Jobs Report 2025; Humlum and Vestergaard, NBER working paper w33777 on large language models and labour markets (revised 2026); Brynjolfsson, Li and Raymond, 'Generative AI at Work' (Quarterly Journal of Economics, 2025); QuickBooks Online and Xero AI product pages.