Plan on one to four hours per person per week for staff who spend much of their day writing, reading or re-keying information, once habits settle after the first month. Automating a whole workflow can save more on its own. Staff whose work is mostly face to face or hands-on may save less than an hour.
Those are planning figures, not promises. Published research is more modest than vendor claims, and the first month usually saves almost nothing once learning time is counted. The only number worth trusting is the one you measure in your own business, and you can do that in a fortnight without a stopwatch or any new software.
What the published research found
A small average across whole jobs. Economists Anders Humlum and Emilie Vestergaard surveyed around 25,000 workers in 11 occupations exposed to AI chatbots. In the 2025 version of their paper, users reported average time savings of 2.8% of work hours: a little over an hour in a 40-hour week. That average includes light and occasional users.
A bigger gain on one well-suited job. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,172 customer support agents given an AI assistant. Issues resolved per hour rose 15% on average, with the largest gains among less experienced agents.
Large speed-ups on the right tasks, and losses on the wrong ones. In a field experiment with 758 BCG consultants, those using AI finished suitable tasks 25.1% faster, but on a task outside the AI's capabilities they were less likely to get the right answer.
Put together, the picture is consistent: AI can make well-suited tasks much quicker, by about a quarter in the consultants' experiment, but those tasks fill only part of anyone's week, so the saving across the whole week is smaller. That's the arithmetic every realistic estimate rests on.
Where the weekly number comes from
The weekly saving for one person is roughly:
hours a week on AI-suitable tasks × how much faster those tasks get, after checking
Say an accountant works 37.5 hours and spends 15 of them drafting emails, summarising documents and writing client commentary. If AI cuts those tasks by 30% after review, that's 4.5 hours a week. A colleague who spends only 6 hours on such work, with a 20% cut, saves just over an hour. Both are honest answers to the same question, which is why a single headline figure for "a small business" is meaningless. The mix of work decides it.
Which businesses sit at the top of the range, and which at the bottom
Desk-based professional firms such as accountancy practices, law firms, recruitment agencies and insurance brokers sit towards the top, because a large share of every role is reading, writing and moving information between documents. Two to four hours a week per person is realistic once habits settle.
Firms with a field team and an office, such as trades, cleaning or engineering site services, split sharply. The office administrator and the owner can save several hours a week on quotes, scheduling emails and invoicing; the people on site may save very little, apart from writing up job notes by voice.
To put numbers on that, picture a six-person electrical contractor (illustrative): an office manager, the owner and four electricians. The office manager's quotes, booking confirmations and invoice chasers might save her around three hours a week. The owner saves perhaps an hour on quotes and supplier emails. Each electrician dictating job notes into a phone in the van, instead of typing them up at home, saves maybe ten minutes a working day, about 50 minutes a week. That's roughly seven hours across the team, and the largest single share sits with the office manager, which tells you whose workflow to set up first.
Shops, cafés and studios save least per person, because most hours are spent serving customers. The owner usually gains most, on marketing, supplier emails and the books.
An illustrative independent bookshop, run by the owner with three part-timers, shows the scale. The owner might save about 45 minutes on the weekly newsletter, half an hour on social posts and another half hour on supplier and event emails: under two hours a week. Each part-timer saves perhaps 15 minutes writing short descriptions for the website's staff picks. That's about two and a half hours across four people, most of it the owner's, and for a shop that's a perfectly good result. It just isn't the four hours a head a vendor page might suggest, and it shouldn't be budgeted as if it were.
Whole-workflow automations sit outside these per-person figures. An automation that sorts 200 incoming emails a week, at 30 seconds of human sorting each, saves about an hour and forty minutes without anyone opening a chat tool. One that captures 150 supplier invoices a month at four minutes each saves around ten hours a month. Those savings add to the per-person numbers rather than replacing them, and they're easier to measure because you can count the items.
Per-task savings you can plan with
These are illustrative estimates for common office tasks. The "with AI" column includes the time spent checking and correcting the output. Replace them with your own timings as soon as you have them.
| Task | By hand | With AI, including review | Saved each time |
|---|---|---|---|
| Routine client email reply | 8 min | 4 min | 4 min |
| Summary of a 20-page document | 40 min | 15 min | 25 min |
| Notes and actions from a 45-minute call (with a note-taker) | 25 min | 8 min | 17 min |
| First draft of a proposal or engagement letter | 60 min | 30 min | 30 min |
| Monthly commentary on a client's figures | 45 min | 25 min | 20 min |
| Coding 100 bank transactions (software's built-in AI) | 50 min | 20 min | 30 min |
| Job advert from a role description | 45 min | 15 min | 30 min |
Notice what isn't on the list: client meetings, phone calls that need judgement, reviewing and signing off work, and anything involving a physical document that has to be found first. Those make up a large share of most weeks, and AI barely touches them.
A seven-person accountancy practice, added up (illustrative)
Take the per-task figures above and apply them to a typical week in a small practice: two partners, three accountants, a bookkeeper and an administrator.
| Role | People | AI-suitable work each week | On paper, per person | Realistic, per person |
|---|---|---|---|---|
| Partner | 2 | 25 emails, 3 calls, 1 proposal | 3.0 hours | 1.5 hours |
| Accountant | 3 | 30 emails, 2 summaries, 4 commentaries | 4.2 hours | 2.1 hours |
| Bookkeeper | 1 | 600 transactions, 20 emails | 4.3 hours | 2.2 hours |
| Administrator | 1 | 40 emails, 2 calls | 3.2 hours | 1.6 hours |
| Team | 7 | 26 hours | About 13 hours |
On paper the practice saves 26 hours a week. The realistic column halves it, for the reasons in the next section: not every email goes through the AI, some drafts need rework, and people are still learning. About 13 hours across seven people is roughly two hours each, a third of a full-time role, and squarely inside the planning range.
The first month is different. If each person spends around five hours learning the tool, the rules and the checks, that's 35 hours of learning against a partial saving, so month one roughly breaks even. Tell the team that in advance, or the first fortnight will feel like a failure.
Why the saving shrinks between the spreadsheet and the week
- Review and rework. Drafts that need a second attempt, or a figure that has to be corrected, eat the saving on that task.
- Habit. People fall back to doing it by hand when they're busy, which is exactly when the saving would have mattered.
- Prompt fiddling. Ten minutes tinkering with instructions to save five minutes of writing is a common early pattern. Shared, tested prompts fix it.
- The bottleneck was elsewhere. Saving 20 minutes on a report that then waits three days for a partner's sign-off doesn't get it to the client any sooner.
- Copying between apps. Pasting text out of one system and into a chat tool and back again costs a minute or two each time. AI built into the software you already use avoids it.
- Absorption. Freed time fills up with other work. That isn't bad, but it makes the saving invisible unless you've decided where it should go.
The bottleneck point catches firms out more than any other. Say a three-surveyor practice cuts report drafting from about three hours to one and a half. Clients still wait the same eight working days, because every report queues for the principal's sign-off, which happens whenever she finds a spare afternoon. The hours were saved; the turnaround wasn't. It only changed when sign-off moved to a fixed half-hour slot each morning.
Copying between apps is the quiet one, and a quick sum shows its size. The administrator's replies save four minutes each on the table above, but if each also takes about a minute and a half to copy the client's email into a separate chat tool and paste the draft back, that's 60 minutes gone across 40 replies a week, out of 160 minutes saved. Using the assistant built into the email program instead, where the draft appears in the reply window, recovers most of that hour without changing anything else.
A shared prompt deals with most of the fiddling. This is the kind of thing the accountancy practice might keep for routine replies, tested once and then used by everyone:
Draft a reply to the client email below. Use only the facts in the
email and in my notes. Under 120 words, plain English, friendly but
brief. If the client asks something my notes don't answer, write
[ASK PARTNER] instead of guessing.
My notes: [two or three lines]
Client email: [paste]Given a client asking whether a new laptop counts as a business cost, and the note "yes if mainly for business; need receipt; record as equipment", an illustrative draft reads: "Yes, you can include the laptop as long as it's mainly used for the business. Please send us the receipt and we'll record it as equipment. You also asked whether last year's purchases can be added now: [ASK PARTNER]." That's about four minutes including the read-through, and the one gap is flagged rather than filled with a guess.
Measure your own number in two weeks
You don't need stopwatches or time-tracking software. You need a tally and honest rough minutes. Setting a baseline before you introduce AI explains why the "before" week matters so much.
- Pick three frequent tasks per person, ones they do at least five times a week.
- Week one, the baseline. Each time the task is done, note the rough minutes. No AI.
- Week two, with AI. Same tasks, same note, including checking time, plus a tick if the output needed significant rework.
- Calculate. For each task: (average minutes before minus average minutes after) multiplied by how many times a week it's done. Add the tasks up per person, then across the team.
- Repeat at three months. Savings usually improve as prompts and habits settle, and the second measurement is the one to budget on.
TIME LOG: [NAME] Week: baseline / with AI (circle one)
Task 1: ____________________
Times done (tally): ______ Rough minutes each: ___ ___ ___ ___ ___
Task 2: ____________________
Times done (tally): ______ Rough minutes each: ___ ___ ___ ___ ___
Task 3: ____________________
Times done (tally): ______ Rough minutes each: ___ ___ ___ ___ ___
Rework needed (with-AI week only): task 1 __ times, task 2 __, task 3 __
Anything slower with AI? ____________________________________Filled in by the practice's administrator for the with-AI week, it might read (illustrative):
TIME LOG: Administrator Week: with AI
Task 1: Routine client email replies
Times done (tally): 25 Rough minutes each: 5 4 6 5 5 (average 5)
Task 2: Call notes and actions
Times done (tally): 2 Rough minutes each: 11 13 (average 12)
Task 3: Chasing missing records
Times done (tally): 8 Rough minutes each: 4 3 5 4 4 (average 4)
Rework needed (with-AI week only): task 1: 3 times, task 2: 1, task 3: 0
Anything slower with AI? No, but the note-taker misheard two client
names on one call.Against her baseline averages of 8, 25 and 7 minutes, the sum is (8 − 5) × 25 + (25 − 12) × 2 + (7 − 4) × 8 = 75 + 26 + 24 = 125 minutes, a little over two hours. That's slightly above the 1.6 hours the table assumed, which is exactly why you measure. Rework on replies was 3 in 25, under one in five, so that prompt is ready to share. The misheard names earn a rule of their own: every name in a call summary gets checked against the client list.
How to read the result: if a person's saving comes out under 30 minutes a week, their three tasks were probably the wrong ones, so swap in different tasks rather than concluding AI doesn't work for them. If a task got faster but needed rework more than one time in five, the prompt or the input is the problem; fix that before scaling it to the team. And if one person's saving is far above everyone else's, find out what they're doing differently and share it.
Watch for an unusual baseline week. Suppose the bookkeeper's baseline fell at month-end, with 900 transactions instead of the usual 600 and every one rushed. Her week-two total will look better simply because there were fewer items, and her minutes per item may shift because the pressure was different; neither tells you much about the AI. Compare minutes per item rather than total hours, and if the baseline week clearly wasn't typical (a holiday, a colleague off sick, a deadline), repeat it the following week before starting with AI. One extra week costs far less than budgeting on a number that was never real.
Make it clear that a slower result is useful information, not a failure. If people think they're being judged on how much time AI saves them, the logs will tell you what they think you want to hear. Measuring time saved after rolling out AI covers the longer-term version, once the tools are in daily use.
Deciding in advance what the hours are for
Thirteen hours a week is real capacity, but it only becomes value if someone decides what it's for. In the accountancy practice, the options are concrete. If a typical small client takes about four hours of work a month, 13 hours a week is room for roughly a dozen more clients without a hire. Or it clears the year-end backlog without overtime. Or it moves each accountant a couple of hours a week from routine drafting towards advisory work that clients pay more for.
Pick one before the tools go live, and check at three months whether it happened. Why AI saves time but not money explains what happens when nobody decides, and how to calculate AI ROI turns the hours into a return you can compare with the cost.
More questions about time saved
Do time savings grow after the first few months?
Usually, for a while. Prompts get better, staff learn which tasks suit AI and stop using it for the ones that don't, and checking gets quicker as people learn where the tool slips. Savings from faster individual tasks tend to level off once the suitable tasks are covered. The next step up normally comes from automating a whole workflow rather than doing each task faster.
Why do some staff save far more time than others?
Mostly because their weeks contain more suitable work: someone answering 40 routine emails a day has more to gain than someone in client meetings. Habits matter too. Research on customer support agents found less experienced staff gained most from an AI assistant. A careful colleague who double-checks everything may save less, and that caution isn't necessarily a problem.
Should I count time saved by automations separately?
Yes. An automation that sorts emails or captures invoices saves time without anyone opening a chat tool, so it won't show in individual task logs. Measure it as items handled per week multiplied by the minutes each used to take, then subtract the time spent fixing exceptions and checking samples. Report the two numbers side by side.
Further reads
- How Much Time Can AI Save a Solo Consultant Each Week? — The same question answered for a one-person business.
- How Much Time Does AI Save a Recruitment Agency per Role? — Time saved per role filled in a recruitment agency.
- How to Check AI Is Doing Good Work, Not Just Fast Work — Make sure faster work isn't worse work.
- Is It Worth Automating a Task You Only Do Once a Week? — When a task is too infrequent to be worth automating.
- How to Find AI Savings Line by Line in Your Profit and Loss — Find where time savings could show up in your accounts.
- Will AI Replace My Employees? An Honest Answer for Small Firms — What freed-up hours mean for each role in the team.
- AI or a New Hire? How to Decide Before You Recruit — Break a planned role into tasks, see which ones AI can absorb, cost hire against AI over a year, and set a clear trigger for recruiting anyway.
- Is AI Worth It for a Small Business? How to Work Out Your Answer — A three-number test for whether AI pays in your business, a break-even table, the costs the quick maths misses, and a two-week trial to settle it.
- 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.
- How to Redesign Job Roles Once AI Handles Routine Tasks — A four-step way to rework roles after AI takes routine tasks: task inventory, honest freed-hours count, a decision on the time, and new role descriptions.
- Do I Need AI in My Business? A Decision Guide for Owners — A decision guide for owners: the three situations where AI is worth it, when waiting is sensible, a ten-question score and a 30-day test.
- ClickUp vs Asana vs Monday: Which AI Features Save Time? — All three now charge for AI differently. A timed photography-studio example shows which features save hours and what each costs a five-person team.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Humlum and Vestergaard, NBER working paper w33777 (2025 version reporting average time savings); Brynjolfsson, Li and Raymond, 'Generative AI at Work' (Quarterly Journal of Economics, 2025); Dell'Acqua and colleagues, field experiment with BCG consultants (2023).