Identify Repetitive Tasks to Automate: A Half-Day Workshop

Sixty percent. According to Asana's global Anatomy of Work research, that is how much of a knowledge worker's time goes to "work about work": chasing updates, asking for status, switching between tools, entering the same information a second time. The same study found that employees spend only 27 percent of their time on the skilled work they were actually hired for. Those figures are not a direct measure of repetitive work; they include meetings too. But the direction is clear: at most desks, work takes more time than doing the work, and the first step is to identify repetitive tasks to automate before any tool gets bought.

The McKinsey Global Institute's November 2025 report gives a similar order of magnitude: technologies already demonstrated today could technically automate activities that account for 57 percent of working hours in the US, 44 points of that through software agents and 13 through robots. That is an activity-hours calculation, not a job-loss forecast. And that is exactly why it is useful: automation opportunity hides inside activities, and looking for it in job titles is a waste of time.

So which activities are those in your company? You find out in a half-day internal workshop, no consultant's report required. This piece gives you that workshop's minute-by-minute plan, the task inventory table and the four-criteria scoring formula. For the broader picture, see our roadmap for business process automation; this article covers the first step on that map, the "where do we start" question.

What counts as repetitive work, and how is it different from "work about work"?

Repetitive work is work where the same steps are performed on the same kind of input at regular intervals and the outcome follows a rule: invoice reconciliation, sending account statements, keying an order from a chat message into the ERP, sending a shipment tracking notification. "Work about work" is broader: meetings, status requests, waiting for approvals. The automation workshop focuses on the first group; the second is a process design problem.

The two groups call for different fixes. An order entry done thirty times a day is an automation candidate; a "where is this order" meeting held three times a week is solved by making order status visible. Put both on the same list and your scoring gets muddled.

The repetitive tasks we see most often in small and mid-sized companies: matching electronic invoices against orders, posting bank statement lines into the accounting system, sending account statements to customers, typing order emails into the ERP by hand, messaging tracking numbers to customers one at a time, and collecting payroll and accounting documents at month end. This list comes from field observation, not from a study; producing your own is the whole point of the workshop.

A note for readers in Turkey: mandatory e-invoice and e-archive systems, monthly social security filings and dealer orders arriving over WhatsApp put reconciliation and chat-to-ERP entry at the top of nearly every inventory we have run there.

How do you identify repetitive tasks to automate in half a day?

The workshop takes four hours, one person from each department attends, and the output is a scored task inventory. Schedule it from nine in the morning to one in the afternoon; a workshop that spills into the afternoon turns into a meeting. No computers needed; paper, pens and a wall are enough.

• 09:00-09:20, framing: state the purpose: "whose day are we going to free up". Take the question "whose job is going away" off the table in the first minute. That one sentence shapes the rest of the session.
• 09:20-10:20, individual inventory: everyone thinks through their own week and writes each repetitive task on a sticky note: task, frequency, duration. Rule: one task per note. Ten to fifteen notes per person is normal.
• 10:20-10:40, break.
• 10:40-11:30, grouping: notes go on the wall; the same task written under different names gets merged. "Order entry" and "copying from email into the ERP" are one task.
• 11:30-12:30, scoring: each task gets 1 to 5 points on four criteria (below). The person who does the task assigns the score, not their manager.
• 12:30-13:00, ranking and the top three: sort by total score and, for the top three, answer "who, with which tool, in how many weeks".

The facilitator should not be a department head; if it is, people write down what they think the boss wants to hear. A neutral person from accounting or IT, or an outsider, produces a healthier result.

Which columns belong in the task inventory?

The inventory table has seven columns: task name, who does it, frequency (times per day, week or month), minutes per occurrence, systems used, what happens when it goes wrong, and whether the rule can be written down. The last two columns are the heart of the scoring, and most inventories leave them out.

The "what happens when it goes wrong" column measures error cost. A mistyped tracking number costs one phone call; a mistyped invoice amount costs a reconciliation crisis. The "can the rule be written down" column measures suitability for automation: if the person doing the task can put the steps on one page, the rule is clear; if they say "it depends", either the rule is clear but nobody has written it yet, or the task genuinely needs judgment.

In the workshop you test rule clarity like this: ask two people who wrote down the same task to describe their steps out loud; if the descriptions diverge, lower the score. Task mining tools measure the same thing as "variance across users".

The scoring formula: frequency, duration, error cost, rule clarity

Each task gets 1 to 5 points on four criteria and the scores are multiplied; the highest product is the best automation candidate. Multiplication is preferred over addition because a task scoring 1 on any single criterion (done once a month, or with a rule nobody can write) is a poor candidate even if its total looks high; multiplying sinks it automatically.

• Frequency: monthly = 1, weekly = 2, daily = 3, 5 to 10 times a day = 4, more than 10 times a day = 5.
• Duration: up to 2 minutes each time = 1, 5 minutes = 2, 15 minutes = 3, 30 minutes = 4, more than 30 minutes = 5.
• Error cost: nobody notices = 1, internal correction = 2, delay visible to the customer = 3, financial loss = 4, legal or reputational damage = 5.
• Rule clarity: judgment needed every time = 1, mostly exceptions = 2, half and half = 3, rare exceptions = 4, fully rule-based = 5.

Example: typing order emails into the ERP. Thirty times a day (5), seven minutes each (2), an error mixes up a shipment (3), the rule is clear but the product code is sometimes missing (4). Product: 120. Posting the monthly bank statement: once a month (1), four hours (5), an error triggers a reconciliation crisis (4), rule is clear (5). Product: 100. The low-frequency, long-duration task ranks below the high-frequency one, and that is exactly what the formula is for. Twelve times a year at four hours is 48 hours; thirty times a day at seven minutes is about 870 hours a year.

A separate note on tasks with an error cost of 5: they are automation candidates, but candidates for human-approved automation. The software prepares a draft, a person approves it. We covered what to hand over and what to keep in detail in our guide on the human-AI division of labor.

Task mining tool or workshop?

For a company under fifty people the workshop is enough; task mining tools earn their keep at enterprise scale, where they record and cluster the desktop interactions of hundreds of users. What the tools do is the automated version of the workshop: actual time spent, frequency, variance across users and complexity get measured instead of estimated.

Three names stand out. Celonis combines task mining from desktop data with process mining from system logs. UiPath Task Mining clusters screen recordings with machine learning and extracts task variants. Microsoft Power Automate (formerly Process Advisor) offers both and is the easiest entry point for companies already on Microsoft 365.

The difference: process mining sees work that leaves a trace in a system, task mining sees work done outside systems (spreadsheets, email, chat apps). Most repetitive work in a small company sits in that second group, so if you buy a tool at all, buy the task mining side. But given prices and setup effort, in a fifty-person company the half-day workshop produces the same list.

Do not automate a broken process

Automation applied to an inefficient process magnifies the inefficiency. The principle circulates as a quote attributed to Bill Gates; verified or not, the field confirms it every week. Fix the process before automating any task that scored 2 or below on rule clarity; otherwise the software just produces the confusion faster.

A typical example: sending customer account statements. The customer list lives in three separate spreadsheets and nobody knows which one is current. Automating the send in that state means automating the sending of statements to the wrong address. One list first, then automation. In the workshop, tag tasks like this "fix first"; they enter the automation queue in the second round.

And then the rule on irreversible actions: steps that move money, delete records or send a contract to a customer keep a human approval screen. They stay in the automation queue; they just do not run unattended.

The inventory of a 35-person textile wholesaler

An example from the field; the company is anonymized and the figures rounded. A 35-person textile wholesaler ran this workshop in March 2026. Four hours produced 47 repetitive tasks; after grouping, 31 remained. The top three after scoring:

• Entering dealer orders received over WhatsApp into the ERP: 40 orders a day, 6 minutes each, shipping errors, clear rule. Product: 150.
• Messaging tracking numbers to customers: 40 a day, 2 minutes, customer phone calls, fully rule-based. Product: 75.
• Building the weekly sales report in a spreadsheet: once a week, 3 hours, internal correction, clear rule. Product: 50.

For the first task, an AI model now extracts product, quantity and dealer from the chat message and creates a draft order in the ERP; sales support approves it. The setup is close to the flow in our n8n business automation scenarios. Six weeks later, time per order had fallen from 6 minutes to 1: 200 minutes a day, roughly 70 hours a month freed up. Tracking notifications followed in the second month, the report in the third. The remaining 28 tasks are waiting, but the company now knows which task it is postponing and why.

One detail worth pausing on: before the workshop, management's first candidate was "a customer chatbot". It did not appear once among the 31 tasks, because nobody was writing answers to customer questions ten times a day. Scoring brought measured work to the top; the enthusiastic idea never even made the list.

Frequently asked questions

Is half a day enough, or do I need a task mining tool?

Under fifty people, half a day is enough. With hundreds of users and many systems, a task mining tool replaces the workshop, but producing the first list with a workshop is still a cheap way to start.

What if employees push back with "my job is being taken away"?

Set the frame in the first twenty minutes: the goal is to free up the day. Most tasks that surface in the inventory are tasks nobody enjoys; having employees score their own tasks dissolves most of the resistance.

How do I know whether a task is suitable for automation?

By the product of the four criteria. Tasks scoring 1 on frequency or rule clarity are poor candidates regardless of duration.

What is the difference between process mining and task mining?

Process mining reads system logs, task mining reads desktop interaction. In a small company, repetitive work mostly happens outside systems, so the second one sees more.

What happens after the first three tasks?

Update the inventory after three months; the first automation usually creates new repetitive tasks (approvals, checks), and the second scoring round comes out more accurate.

How do I measure success?

With the duration column: minutes per task, before and after. The workshop already gives you the baseline, the number everyone goes looking for later.

So what should you actually do?

• Put the workshop on the calendar: four hours, one person per department, no manager as facilitator.
• Use the seven-column inventory; do not skip the "what happens when it goes wrong" and "can the rule be written down" columns.
• Multiply the four criteria, do not add them. A task scoring 1 on any criterion belongs at the bottom of the list.
• Set aside low rule-clarity tasks as "fix first". Speeding up a broken process makes the problem bigger.
• Fit the first three tasks into six weeks and keep the duration column as your baseline measurement.

An automation opportunity map is a ranking that emerges from forty sticky notes on a wall; it is not something you buy as a software product. While the notes are still on the wall, the "which task, and why first" argument settles itself. Once your inventory exists, working out which tool handles the top three and in how many weeks takes about half an hour; send us the table and we will run the numbers with you.