How much time can you reclaim with AI?

The research behind this estimator

Most businesses know roughly how long their work takes, but few have measured where that time actually goes. This estimator borrows from a long line of research on measuring work in service businesses and offices. Here is what it draws on and why it asks the questions it does.

Working on a task and waiting on it are different

Researchers who study office and service work separate two kinds of time. The first is the time someone is actually working on a task. The second is the total time from when the task arrives to when it is done, which includes every hour it sits in an inbox or a queue. Karen Martin and Mike Osterling found these two measures, along with how often work has to be fixed, are enough to understand most office and service processes. Michael George found that in most service and transactional processes, the work is being actively handled less than 10% of the time it takes to finish. That is why the estimator asks both “how long does someone work on it” and “how long does it take from start to finish.”

Small tasks add up to large yearly totals

Robert Kaplan and Steven Anderson at Harvard Business School showed that a business can understand the cost of its work with just two numbers: how long a task takes and how often it happens. Multiplying those together across a team and a year turns a 90-minute task into a figure leaders can plan around. The estimator uses the same approach, and it counts 48 working weeks in a year rather than 52 to account for holidays and time off.

Busy teams create long waits

Queueing research explains why work slows down even when everyone is busy. John Kingman showed that waiting time rises sharply as people approach full capacity, especially when the amount of work arriving and the time each task takes vary from day to day. John Little showed that the time a task spends in progress is tied directly to how much work is in progress at once. Niklas Modig and Pär Åhlström describe this as the difference between keeping people busy and keeping work moving. This is why saving hands-on time gives your team hours back but may not shorten how long your clients wait, and why the estimator shows both.

Redos are hidden time

Martin and Osterling also measure how often work arrives at the next step complete and accurate, without needing to be corrected, filled in, or clarified. When a step has to be redone, it repeats both the work and the wait in front of it. The step-by-step view asks how often each step is redone or sent back for this reason.

I ask for totals because people estimate them best

A study of office workers published in the Journal of Occupational Psychology compared what people reported about their work with what trained observers recorded. People were good at naming the tasks they do, less accurate at judging how their time divides across tasks, and least accurate at estimating the time for any single activity. For this reason, the quick estimate asks about one task you know well, the step-by-step view is optional, and redos are answered with “rarely, sometimes, or often” rather than a percentage.

AI savings include the time to check the work

AI can produce a first draft quickly, but someone still has to review it. Recent research by Silvia Bartolucci and Pierpaolo Vivo shows that when AI errors slip past review and come back as rework, a workflow can slow down overall even while each task gets faster. The estimator builds this in: when you choose “AI helps” for a step, it adds review time back before showing your savings.

What this estimate leaves out

A closer look at your workflows covers things a short estimator cannot. It measures how much of each person’s week is already committed, which shows where work truly gets stuck. It separates waiting on clients from waiting on your own team, a distinction Frances Frei’s research at Harvard Business School shows is central to service businesses. It also counts requests that only exist because something went wrong the first time, such as status checks and follow-ups, which John Seddon calls failure demand. These are the measures the AI Opportunity Assessment adds.

Sources

• Bartolucci, S., & Vivo, P. (2026). Queue & AI: When faster tasks slow down the workflow. arXiv preprint 2605.27202. https://arxiv.org/abs/2605.27202
• Frei, F. X. (2006). Breaking the trade-off between efficiency and service. Harvard Business Review, 84(11). https://hbr.org/2006/11/breaking-the-trade-off-between-efficiency-and-service
• George, M. L. (2003). Lean Six Sigma for Service: How to Use Lean Speed and Six Sigma Quality to Improve Services and Transactions. McGraw-Hill.
• Hartley, C., et al. (1977). Subjective time estimates of work tasks by office workers. Journal of Occupational Psychology, 50(1). https://doi.org/10.1111/j.2044-8325.1977.tb00355.x
• Kaplan, R. S., & Anderson, S. R. (2004). Time-driven activity-based costing. Harvard Business Review, 82(11).
• Kingman, J. F. C. (1961). The single server queue in heavy traffic. Mathematical Proceedings of the Cambridge Philosophical Society, 57(4), 902–904. https://doi.org/10.1017/S0305004100036094
• Little, J. D. C. (1961). A proof for the queuing formula L = λW. Operations Research, 9(3), 383–387. https://doi.org/10.1287/opre.9.3.383
• Martin, K., & Osterling, M. (2013). Value Stream Mapping: How to Visualize Work and Align Leadership for Organizational Transformation. McGraw-Hill.
• Modig, N., & Åhlström, P. (2012). This Is Lean: Resolving the Efficiency Paradox. Rheologica Publishing.
• Seddon, J. (2003). Freedom from Command and Control: A Better Way to Make the Work Work. Vanguard Education.