Start from tasks, not job titles: list each person's weekly tasks with hours, mark which ones AI now drafts or does, subtract the time checking takes, then decide deliberately where the freed hours go. Rewrite each role around judgement, client work and review duties, agree changes with the person, and take employment advice before altering contracts, hours or pay.
If you skip the redesign, the freed time doesn't vanish. It leaks into slower email, longer tea breaks and a vague sense that AI "didn't really save anything". The sections below produce a task inventory for each affected role, a clear decision on what the freed time is for, a rewritten role description, and a plan for the conversation with each person, which is the part that most often goes wrong.
Inventory the tasks in each role, not the job title
Job descriptions are too coarse for this. "Prepares survey reports" hides six different tasks, some of which AI now does well and some it can't touch. Ask each person to keep a simple log for one normal week, or sit with them for 30 minutes and estimate together from their calendar and sent items. List every task that takes more than about an hour a week.
Then add three columns: what AI now does for that task (does it, drafts it, assists, or nothing), the hours after AI including checking, and who checks the AI's work. Here's what one row set might look like for a report administrator at an illustrative surveying firm:
| Task | Hours a week before | What AI does now | Hours after, including checking |
|---|---|---|---|
| Formatting reports from surveyors' drafts | 14 | Applies template, numbering, photo captions | 5 |
| Booking inspections by phone and email | 8 | Drafts confirmation emails | 7 |
| Chasing clients for access details | 4 | Drafts chasers from a template | 3 |
| Invoicing | 3 | Nothing yet | 3 |
| Filing and archiving | 3 | Nothing | 3 |
The totals tell you where the change is concentrated: this role goes from 32 hours on these tasks to 21, and nearly all of it comes from one task. That's typical. AI rarely thins every task a little; it usually removes most of one or two.
Count the freed hours after checking time
Three adjustments stop you overestimating. First, subtract checking time. A draft that needs reading against the source still costs minutes, and those minutes belong in the "after" column. Second, allow for the dip: the first month with any new tool is slower, so measure the "after" figure once people have used it for at least four weeks. Third, only count tasks where the AI use is actually established. If one person uses it and two don't, the saving isn't in the role yet; it's in the person.
All three show up in one illustrative case. An online homeware shop has two customer-service assistants. In the first week of AI-drafted replies, one of them reckons she's saving about 9 hours a week, and the owner starts planning to move stock-checking into the role. Measured properly in week six, the picture is different: replies are drafted in seconds, but checking order details and softening the AI's tone takes about 3 minutes each on 90 emails a week, and the second assistant still writes most replies by hand because the drafts get product names slightly wrong. Net saving across the role: about 4 hours a week, not 9. Planned on the first figure, stock-checking would have landed on a role that didn't have room for it.
Then apply a threshold. If a role frees up only an hour or two a week, leave it alone and let the time absorb. If it frees a fifth of the week or more, redesign deliberately, because unplanned time that large gets filled with whatever is loudest. Why AI saves time but not money goes into how freed hours quietly disappear if nobody claims them.
Decide what the freed hours are for
This is a business decision, and it's worth making before you talk to anyone, so you're not improvising in the meeting. There are six realistic options:
| Use of freed time | When it fits | Watch out for |
|---|---|---|
| More of the same work | You have a waiting list or turn work away | Only works if the bottleneck was this role, not someone upstream |
| Move tasks down from more senior staff | Senior people spend hours on work a trained colleague could do | Needs training and a clear sign-off rule |
| Bring outsourced work in-house | You pay a freelancer or bureau for something routine | Check the in-house quality and capacity before cancelling the supplier |
| Faster or better service | Response times or quality lose you work | Set a measurable target, or the time drifts |
| Leave the next vacancy unfilled | Someone is leaving anyway and demand is flat | Don't spread the leaver's work so thin that errors rise |
| Fewer hours | The employee wants it, or demand has fallen | Contract change: needs agreement and advice first |
The first row's warning catches people out. Take a picture-framing shop where AI now drafts order confirmations and progress updates, freeing about 5 hours a week for the shop assistant. The owner's plan is "more of the same": take on more framing orders. But the assistant was never the constraint. The framer can make about 30 frames a week, and he's already at 28. Extra orders taken at the counter would only lengthen the waiting time from two weeks to three. The freed hours went instead on something the assistant could finish herself: photographing finished work for the shop's online listings, which the owner had been doing on Sunday evenings.
Most small firms end up with a mix: some tasks moved down from senior staff, some capacity for more work. Be honest with yourself about which you're choosing. People can tell the difference between "your time is now for client calls" and "we'll see".
Rewrite the role around checking and judgement
A redesigned role usually shifts from producing something to checking and finishing it, plus the new tasks you chose for the freed hours. That shift needs saying on paper, because "check the AI's draft" is a real responsibility with real consequences if it's skipped. For the wording itself, writing job descriptions with AI has prompts; the structure below is what should change.
ROLE: Report administrator VERSION: 2 (Oct 2026)
PURPOSE (one line):
Get accurate, on-brand survey reports to clients on time, and keep
inspections booked and clients informed.
RESPONSIBILITIES THAT CHANGED
- Finalise AI-formatted reports: check every photo caption against
the surveyor's notes, check numbering and defect references.
(was: format reports by hand)
- Own the report template and the AI instructions that apply it;
update both when a surveyor changes the standard wording.
- Log any AI error that reaches a surveyor for sign-off.
NEW RESPONSIBILITIES (from the freed hours)
- Client updates during the job: booking changes, access, report ETA.
(moved from surveyors)
- Post-report follow-up call within 5 working days.
WHAT DOESN'T CHANGE
- Invoicing, filing, inspection booking.
MEASURES
- Reports issued within 2 working days of surveyor sign-off.
- Zero caption or reference errors reaching clients.
AI TOOLS USED IN THIS ROLE AND THE RULES THAT APPLY:
see AI usage policy, section 3.
Keep the old version. If questions come up later about what changed and when, a dated before-and-after is far better than memory.
Protect the way junior staff learn
Routine work is often how people learn a trade. A junior surveyor who writes forty condition reports learns what matters in a building partly by struggling to describe it. If AI now drafts those reports from notes, the junior gets faster output and fewer lessons.
Build the learning back in on purpose. Have juniors write a set share of reports unaided, say one in four, and compare their draft with the AI-assisted version in a short review with a senior. Make them explain every change they made to an AI draft: "why did you change this sentence?" is a better teacher than proofreading. And set out a sign-off ladder, so it's clear when someone moves from "a senior checks everything" to "checks a sample". The surveying firm's version, filled in:
SIGN-OFF LADDER: junior surveyors, AI-assisted reports
Rung 1 Senior checks every report in full.
Junior writes 1 report in 4 with no AI draft.
Move up after: 20 reports with no correction a client
would have noticed.
Rung 2 Senior checks every defects section; samples the rest.
Unaided reports drop to 1 in 6.
Move up after: 3 months and 40 reports at rung 2,
zero errors reaching clients.
Rung 3 Senior checks 1 report in 5, chosen by the senior.
Any error found returns the junior to rung 2 for a month.
The drop-back rule on rung 3 is what makes seniors comfortable signing off. Without it, a ladder only goes one way and people stop trusting it. The tutorials on setting up human review for AI work and handling staff who over-rely on AI cover the checking habits in more depth.
Pay, grading and the conversation with each person
Talk to each affected person individually before anything is announced to the team. Come with the inventory, the proposed new description and the reason. Then listen: they know the tasks better than you, and they'll often spot a problem in your plan within minutes.
A structure that works for that conversation:
- What AI now does in your role, with the numbers from the inventory.
- What we'd like the freed time to go on, and why.
- What stays the same: pay, hours, reporting line (if true).
- What you'd need, such as training, a different sign-off rule or a trial period.
- What happens next, and when you'll get the final version in writing.
The opening sentence of that conversation sets its tone. Compare two ways the surveying firm's owner could start with the report administrator:
AVOID
"AI's doing a lot of your job now, so we need to work out what
you're going to do instead."
BETTER
"The report formatting that took you about 14 hours a week now
takes about 5. I'd like most of those 9 hours to go on client
updates, which the surveyors are struggling to keep up with. Your
pay and hours stay the same. What would you need to take that on?"
The first version is about what the person has lost, with no figures and no plan. The second gives the number, the proposal, the reassurance that's true, and a question that hands them part of the design.
Prepare for the question people ask most once the checking duty is on paper: "If the AI gets a caption wrong and I miss it, is that my fault?" Answer it with the role description rather than reassurance. For the report administrator, the honest answer might be: "You're responsible for checking captions against the surveyor's notes, which is what the new duty says. The surveyor still signs off the whole report, so an error reaching a client has been missed twice, and we'd look at the process before anyone's performance. If the AI starts making a new kind of mistake, log it and tell me, and we'll change the instructions." That puts the responsibility in writing, sets out the second check, and gives the person a way to raise a problem without it becoming blame.
If the change adds responsibility, consider whether it warrants a regrade. If it touches hours, pay or contractual duties, stop and take advice from an employment adviser or solicitor before proposing it, because contract changes usually need agreement and the rules depend on where you operate. When someone is anxious about where this leads, talking to staff who fear AI will take their job and the more direct will AI replace my employees? help you answer honestly.
Worked example: a surveying firm's report team
Take an illustrative nine-person surveying firm: the owner, four surveyors, two junior surveyors and two administrators. Over six months it added AI to three tasks: formatting reports, drafting report sections from site notes and dictated observations, and drafting routine client emails.
The inventory showed the freed time was concentrated in two roles. The report administrator freed about 11 hours a week, as in the table above. Each junior surveyor used to spend around 10 hours a week drafting report sections; with AI drafts and their own checking, that fell to about 5, freeing 10 hours between the two. The four surveyors saved little overall: less drafting, but slightly more time reviewing AI-assisted sections, which roughly cancelled out.
The decisions: the administrator took over client updates and booking changes from the surveyors, which gave each surveyor back about an hour a week, and started a follow-up call after each report. The juniors each took one extra supervised inspection a week, plus a fortnightly review where one unaided report is compared with its AI-assisted equivalent. Nobody's hours or pay changed; the administrator's role was regraded six months later after taking on the client-update responsibility. The firm's capacity rose by roughly two inspections a week without a hire, which was the point.
If a role shrinks further than you expected
Sometimes the inventory shows that most of a role's hours are now handled by AI and there's no sensible work to move into it. Face that early rather than letting the person sit half-idle, which is worse for them and for the team. Look first at combining roles: two part-emptied roles often make one full, better one. Look next at retraining for work you currently buy in or struggle to recruit for.
A six-person letting agency shows both moves together. AI now handles most of the lettings administrator's tenant-referencing paperwork, freeing about 14 hours of a 37-hour week, and drafts most of the maintenance coordinator's contractor chasers, freeing about 12. Neither role has 12 or more hours of sensible new work in it. Combined, though, the remaining work in both comes to roughly 48 hours: more than one full-time role, less than two. The maintenance coordinator takes the combined admin role full-time. The lettings administrator, who had asked before about getting out of the office, keeps the 11 hours of admin that don't fit and fills the rest of her week with check-in and inventory visits, which the agency had been paying a freelance inventory clerk to do. Both conversations happened with the numbers on the table, and both people agreed to a three-month trial before anything was confirmed in writing.
Only after those, and only with proper advice, consider whether the role is still needed, because any move towards redundancy carries legal process requirements that vary by jurisdiction and contract.
Whatever you decide, don't leave AI as the silent explanation. People accept hard decisions better when they can see the numbers and the options you considered than when they piece it together from rumours.
Check the new roles after 90 days
Three months in, check four things. Is the freed time going where you decided, visible in calendars and job counts rather than in impressions? Have error rates on AI-assisted work stayed flat or improved? Are the juniors progressing on the sign-off ladder? And does each person think the new role is workable? Ask them one to one, because group meetings produce polite answers.
If the time has leaked away, it's usually because the task you chose for the freed hours was vague. Make it specific, with a measure, and check again in a month.
A four-person bookkeeping practice shows how that looks. AI now drafts its monthly client summaries, freeing about 6 hours a week for one bookkeeper, and the new duty was written as "more proactive client contact". At 90 days her calendar shows no change and she can't point to anything specific she's done with the time. The rewrite: "Call each of your 18 monthly clients once a quarter to go through their numbers, and log the call in the practice system." That works out at about six calls a month, each with a clear purpose, and when the quarter ends the log shows 17 of 18 calls done. The time was always there; it needed a job with a count attached.
Questions owners ask about changing roles
Can I cut someone's hours because AI now does part of their job?
Changing hours usually changes the employment contract, which normally needs the employee's agreement and a fair process. The rules differ between legal systems and depend on the contract wording, so take advice from an employment adviser or solicitor before proposing it. Often the better route is to redeploy the time and let natural turnover adjust headcount later.
Should people get paid more for reviewing AI output?
If the redesigned role carries more responsibility, such as signing off work that goes to clients or checking other people's AI-assisted drafts, it may justify a regrade. If the role simply swaps typing for checking at the same level, pay may not need to change. Judge it the way you'd judge any role change: responsibility, skill and accountability, not whether a tool is involved.
What if AI only saves each person an hour or two a week?
Then you probably don't need a redesign. Small savings get absorbed into the rest of the job, and forcing a formal change over two hours creates more worry than value. Note the saving in your task inventory, revisit it when you add the next AI use, and redesign once the freed time in a role reaches a meaningful share of the week.
Further reads
- AI or a New Hire? How to Decide Before You Recruit — Decide whether freed time removes the need to recruit.
- How to Choose and Support an AI Champion in a Small Team — Pick the person who keeps the new ways of working going.
- Who Should Own AI in a Small Business? Roles and Responsibilities — Give AI oversight a named owner in the new structure.
- AI Change Management for Small Teams: A Practical Plan — Plan the wider change around the role redesign.
- How Much Time Can AI Really Save a Small Business Each Week? — Realistic weekly savings to sanity-check your inventory.
- How to Get Staff Buy-In When You Introduce AI — Get people behind the changes before you announce them.
- Should You Charge Clients Less When AI Speeds Up Your Work? — What to do about your prices when AI cuts the time a job takes: the rules for hourly billing, when fixed fees are fair, and how to tell clients.
- Reduce Owner Dependency: Use AI to Capture What Only You Know — Get what's in your head into a form staff can use: an interruption log, an AI interview prompt, decision rules with thresholds, and a shared assistant.
- How to Write Job Adverts With ChatGPT Without Biased Language — A four-step method for bias-free job adverts with ChatGPT, with a requirements sheet, drafting and audit prompts, sample outputs and a full before and after.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.