AI is handing your team back hours. Who decides where they go?

Many people expected ATMs to end bank tellers. The machines took over much of the cash handling, and a typical city branch needed fewer tellers. That made branches cheaper to run, so banks opened more of them: 43% more in cities between 1988 and 2004. Teller jobs held steady overall, and the work moved toward helping customers (James Bessen, IMF, 2015).

The machine freed the time. The banks decided what to do with it.

AI is doing the same thing now, in every kind of role. A site supervisor finishes the daily report before leaving the truck. A scheduler fills tomorrow's routes in minutes. A salesperson walks into a call already briefed.

In a June 2026 BCG survey, 42% of frontline employees who use AI regularly said it saves them a full day a week or more. Among leaders who use it regularly, the figure was 60%. Two in three people surveyed said they get little or no guidance on what to do with the time they save (BCG, AI at Work, 2026). The survey covered 11,749 people in 14 markets, including the US, the UK and much of Europe. Canada was not one of them.

Hours nobody planned for are like a new hire who arrived with no job description, no manager and no first assignment. The hours still get used, on whatever is loudest that week.

Deciding where that time goes is a leadership job, and it is easier to do before the hours arrive than after.

Freed time can go to three kinds of work, and most firms move through them in order:

• Catch up. Work that is behind or not done well, now done faster and properly.
• Get ahead. Work you always meant to do and never had time for.
• Do new things. Work your firm could not do before, because nobody had the skills.

You don't need any numbers to start. If you want to put numbers on one task, the hours test takes about 10 minutes.

Catch up: work that is behind or done poorly

Work you are already losing money on

• Chasing late payments. Calling clients who are 30 to 90 days late, sending reminders, fixing the invoice problem that held up payment. Evidence: we found no current Canadian figure on how often firms get paid late. Your report of unpaid invoices by age will tell you how much of this work is waiting.
• Billing extra work. Changes a client asked for, added services, parts or time left off an invoice. Ask your project leads how often extra work goes unbilled.
• Following up proposals and quotes. Count this only if the person with spare time can make the follow-up call. In many firms the owner or estimator has to.

Work your busiest people would hand over

• Project coordination. Scheduling, chasing documents, keeping the job file current, booking meetings. Evidence: none found on how often people move into this work.
• Admin done by billable staff. Moving admin off a senior person's desk, or a site lead's evening paperwork, only pays if the time goes to client work or better work.

Work with your clients

• Client check-ins. A call after a job closes, or a quick check on a client who has gone quiet. Evidence: none found on staff moving into client support.

Checking the AI's work

• Reading what the AI produces, fixing its mistakes, and updating its instructions. This is part of the cost of using AI, so take it off the hours saved first. The person who did the task by hand is often best at this, because they know what right looks like. Treat it as skilled work and agree how long they will do it. Evidence: our recommendation. We found no hiring data for this role yet.

The same ideas in other kinds of firms

Get ahead: work you never had time for

Every firm has a "we should really" list. Nobody chases you for this work, so it never gets done. With AI doing the first draft, it fits in the week:

• A short client survey after each job, and someone who reads the answers.
• Writing down how your key processes work, so a new hire, or the AI, can follow them.
• Renewal or maintenance calls made before the client has to ask.
• A welcome pack for new clients.
• Tidying shared files and folders so people can find things.
• Client work you turned away or delayed because nobody had time.

Keep only the items that would make or save money, or keep a client.

Do new things: work that wasn't possible before

The bigger change comes as people get good with AI and take on work outside their training. That is work that used to need a specialist, an agency, or a hire you could never justify.

Our founder, Chase Bernier, is one example. Before Sidekick, he had never designed a website, produced marketing or built a brand. Today he builds Sidekick's website, marketing and branding, writes articles like this one, runs research that used to take a team, and builds our processes.

The research points the same way. In a 2024 BCG study, consultants with no data science training used AI to do data science tasks. On a coding task they reached 86% of the data scientists' benchmark, against 29% without AI (BCG Henderson Institute, 2024). In a field experiment at Procter & Gamble, one person working with AI did about as well as a two-person team without it, and people working outside their usual role, with AI, matched teams that included a specialist (Dell'Acqua and others, Organization Science, 2026).

There is a catch. In the BCG study, people could do the work with AI but had not learned to do it alone, and on one task AI sometimes led them astray. Someone still checks the result, especially early on.

What this can look like in a small firm

• The billing clerk builds the cash-flow dashboard nobody had time to set up.
• The office coordinator produces proposals, client presentations and marketing pieces in-house instead of sending them to an agency.
• The site supervisor or project lead builds a simple tool that answers the questions clients ask every week.

Count this work in your plans once someone has tried it and it works. Until then, treat it as upside.

Why firms usually go in this order

Catch-up work is already defined, so it comes first. New work comes last, because it needs people already comfortable with AI. This tracks the four levels of AI adoption. Catching up and getting ahead come with levels 2 and 3, when AI runs repeatable and multi-step work. Doing new things is level 4.

How people grow into the new work

Start with the leaders. A manager who hasn't used AI on their own tasks can't judge what it produces or coach anyone through it. Leaders handing off the learning is the first stall in Where AI rollouts go to die. Give people protected time to learn, for example two hours a week, and a named person to ask. Adjust this for anyone who needs an accommodation. Then there are three ways to move people into new work:

• Upskill. The person keeps their job, learns to do it with AI, and puts the freed time into catching up and getting ahead.
• Reskill. The person trains for a different role the firm needs, such as a billing clerk moving into project coordination.
• Reassign. The person moves to existing work elsewhere in the firm that uses skills they already have.

In a 2025 World Economic Forum survey of over 1,000 employers, for every 100 workers they expect by 2030 to train 29 for their current role, train and move 19 into a new one, and leave 11 without the training they need (WEF, Future of Jobs Report 2025).

Try any move as a trial, agreed in writing. Give the person 60 to 90 days on one piece of work, with one measure of success, and only with their real agreement. Keep pay, title and hours the same, review them at the end, and don't treat a poor fit as a performance problem. Agree in writing beforehand what happens if it is a poor fit, in a letter an HR professional or employment lawyer drafts, since the old job may have changed. In the UK, a trial in a new role offered instead of redundancy is four weeks by law unless you agree a longer period in writing for retraining (GOV.UK). Training pays off slowly, so judge it over the full trial, on quality as well as time saved.

How to find the work in about a week

1. Ask each manager or project lead one question: "What work would you hand over tomorrow if someone had the time?" Write down the task and the hours a week. Count it only if they have better work for the time it frees.
2. Pull your report of unpaid invoices by age. Count the clients 30 to 90 days late in a typical month. Ask your bookkeeper which invoices you gave up collecting last year, and what they added up to.
3. Look at last year's turned-away work. Jobs you declined, delayed or did late because nobody had time. Add up the hours and divide by 52 for a weekly number.
4. Write your "we should really" list, and ask: "What would we do if we had a designer, an analyst or a developer on staff?"
5. Ask the people who do the work now, one at a time, once you have decided what you will try first. Ask what they would most like to spend freed time on. They often know where the gaps are better than anyone.

Then sort what you found into the three kinds of work, and start with one catch-up item.

When there still isn't enough work

Some firms will not find enough. That is a real answer, and better to know before you say anything to your team. There are four paths. They work over months and can be combined.

1. Take on more work. If you turned client work away, take it on. If you were going to hire someone for new work in the next year, see whether the people with spare time could do it instead. It is the best outcome, if the demand is real.
2. Leave roles empty as people leave. When someone leaves on their own, the spare hours can absorb their work instead of a new hire. Most Canadian firms using AI had not cut staff when Statistics Canada last asked: in spring 2025, 89.4% of businesses that had used AI in the past year reported no change in their employment levels (Statistics Canada, June 16, 2025). Its June 2026 update did not repeat the question. Don't ask anyone about their retirement plans: it can feel like pressure to leave, and it can raise legal issues.
3. Retrain someone for work the firm needs. Pick one piece of work from the three kinds of work and run it as a written trial, as described above.
4. If you need fewer people. Decide whose work changes based on the work and skills, written down before you choose, and never on any ground protected by human rights or equality law where your staff work, such as age, sex, disability or family status, or because someone is on, or has taken, a leave. Check that neutral-sounding criteria, such as comfort with AI, don't end up singling out older staff or anyone else protected. In Canada, each province and territory has its own human rights agency (list). If a role is no longer needed, get advice on what you owe in pay and notice before you tell the person. Help with the job search and offer a reference.

What to settle before you talk to staff or change anyone's duties

1. Decide your first step before you speak. Early estimates are rough. Keep them for planning until you know what you will try first. Then tell the team promptly.
2. Only promise what you can keep. Something like: "We're trying AI on billing first. Before anything about your job changes, we'll talk it through with you." In writing, say what you plan to try and avoid promising outcomes you can't control.
3. Tell people early. A short team conversation, then one-to-ones. Say what is changing, what is not, and when you will talk again. Only about one in four employees say their employer has communicated a clear AI plan (Gallup, 2026, US workers).
4. Put any change in writing first. Agree any move with the person in writing, at the same or better pay, title and hours. The person must actually agree. A big change in duties can raise legal issues even at the same pay, so check the rules just below.

A worked example

Illustrative example. The firm and its numbers are invented.

An engineering firm has four people in billing, each spending about 30 hours a week on it: 120 hours. After trying AI on a month of invoices, the office manager expects about 36 hours a week of checking and fixing. About 84 hours a week come free.

What she found in a week of asking

That leaves about 32 hours a week with no work waiting: most of a full-time week.

What she did

• Reskill: the firm planned to hire a project coordinator next year for a new group of projects. Instead, one of the four volunteered to try that work for 90 days. The other three take over her billing, and her new work uses most of the spare 32 hours.
• Pay, title and hours stay the same for the trial, and the measure is whether the new projects stay on schedule. What happens if it is a poor fit is set out in a letter an HR professional drafted.
• Do new things: another of the four is using AI to build a monthly status report for the two largest clients, something the firm never had the skills for. It isn't counted as work for the spare hours until it works.
• Catch up: about 120 hours of very old unpaid invoices and past unbilled work give the team a backlog to work through while the trial runs.

What would need to be measured before calling it a success: hours actually freed after three months, where they went, cash collected sooner, and how the team feels about the change.

Where to go from here

If you have followed this guide from the start, you have looked at why AI matters now, whether it is safe, what it costs, where to begin, and how to train leaders first and then champions. This last step is the one many firms skip: in the BCG survey above, two in three people said they get little or no guidance on what to do with the time they save. The rollout gives people hours back, and somebody has to decide what those hours are for.

Keep the first move small. Pick one catch-up item from your week of asking, name the person who owns it, and set a date about 90 days out to check where the freed hours actually went. What you learn from that one item tells you more about the next one than any estimate.

Sources: James Bessen, Toil and Technology, IMF Finance & Development, 2015. BCG, AI at Work 2026. BCG Henderson Institute, GenAI Doesn't Just Increase Productivity. It Expands Capabilities, 2024. Dell'Acqua and others, The Cybernetic Teammate, Organization Science, 2026. World Economic Forum, Future of Jobs Report 2025. Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2025 (released June 16, 2025) and second quarter of 2026 (released June 11, 2026). Gallup AI indicator, 2026. Provincial training grant pages for BC, Alberta and Ontario, checked October 2026. GOV.UK redundancy guidance, checked October 2026.

This guide is general information for planning. It is not legal, HR or financial advice.