How to Find the Manual Tasks Worth Automating First

The manual tasks worth automating first are the ones that happen often, follow a pattern, use digital inputs you can access and can be checked easily when something goes wrong. Find them by listing the recurring work in one team, measuring how often each task occurs and how long it takes, then scoring each on value, feasibility and risk. Start with the task that scores well on all three, not the one that sounds most impressive.

Key takeaways

• Good first candidates are frequent, repetitive, rule-guided and low in risk.
• Measure before you choose. A one-week task log beats opinions in a meeting.
• Score every candidate on value, feasibility and risk, and rank them.
• Do not automate a broken process. Simplify it first.
• Pick a first project small enough to launch quickly and visible enough that the team notices.

Step 1: Build an inventory of recurring work

Start with one team, not the whole company. Operations, finance, sales support and customer service are common starting points, because their work is repetitive and already digital.

Collect tasks in three ways:

• Ask the people doing the work. Useful questions: What do you copy from one place to another? What do you do every day that needs no real thought? What do you put off because it is tedious? Where do things wait for someone?
• Run a task log. For one week, each person notes every recurring task, how long it took and what triggered it. A simple shared spreadsheet is enough.
• Look at the systems. Shared inboxes, ticket queues and spreadsheet trackers show where manual handling piles up.

Write each task as a verb and an object: "enter emailed purchase orders into the ERP", "chase overdue invoices", "compile the weekly sales report". Vague entries such as "admin" cannot be assessed.

Step 2: Describe each task properly

For every task on the list, capture the same facts:

• Trigger. What starts it: an email, a form, a date, a status change.
• Input. What information arrives, and in what form.
• Steps. What the person does, including the decisions.
• Systems. Which tools are opened, read from and written to.
• Output. What exists at the end: a record, a message, a document.
• Frequency and time. How often it happens and how long each instance takes.
• Exceptions. How often it goes off the normal path, and why.
• Cost of an error. What happens if it is done wrong, and how quickly that would be noticed.

This takes a few minutes per task and usually reveals something by itself. Teams often discover steps nobody can explain, or approvals that exist only out of habit.

Step 3: Score each task

Rate every task from 1 to 5 on three dimensions. The numbers are rough, and that is fine. The purpose is to compare tasks with each other.

Dimension	What to ask	Scores high when
Value	How much time, delay or error does this task cause?	It is frequent, time-consuming, causes waiting or produces costly mistakes
Feasibility	How easy is it to automate with what we have?	Inputs are digital, steps follow a pattern, systems have APIs, examples exist
Safety	How contained is the damage if it goes wrong?	Errors are easy to spot, easy to undo and do not reach customers or money directly

Multiply or add the three scores and sort the list. Tasks that score high on value but low on feasibility go on a later list, with a note on what would need to change. Tasks that score high on value and feasibility but low on safety can still be done, with a person approving each output.

Signals of a strong candidate

• The task happens daily or many times a week.
• Someone reads one thing and types it into another.
• The person doing it can explain the rules in a few minutes.
• There are plenty of past examples with correct outcomes.
• The systems involved can be reached through an API or export.
• A mistake would be caught at the next step.

Signals of a weak candidate

• It happens a few times a year.
• Every instance is different and relies on experience that is not written down.
• Inputs arrive on paper, by phone or in someone's memory.
• The process is changing or disputed.
• It depends on relationships, negotiation or sensitive judgement.
• An error would be serious and hard to reverse.

Step 4: Check whether the process needs fixing first

Automating a bad process makes a bad process run faster. Before building anything, ask of the top candidates:

• Can any step be removed entirely?
• Is the same data entered twice because two systems are not connected?
• Is there an approval nobody ever rejects?
• Could a better form at the start remove the need to interpret free text later?

Sometimes the right answer is a simpler process or a basic integration, with no AI involved. That is a good result. Use AI for the steps that need reading or judgement, and plain rules for the rest. Our post on what AI automation is explains how to tell them apart.

Step 5: Choose the first project

From the top of the ranked list, pick one task that meets these conditions:

• It can be delivered in weeks, not quarters. A quick result builds confidence and teaches you how your team works with automation.
• It has a clear owner. One person who knows the process, will test the output and will decide when it is good enough.
• It is measurable. You know the current time per task and error rate, so you can compare afterwards.
• A human can stay in the loop. The first version produces drafts or suggestions that a person approves.

Resist the temptation to start with the largest, most complex process. Big projects have long feedback cycles, and early mistakes are expensive. A small success gives you a tested pattern to reuse.

A worked example

For example, an operations lead at a trading company could end up with a list like this after a week of logging:

Task	Value	Feasibility	Safety	Decision
Enter emailed purchase orders into the ERP	5	4	4	Start here, with drafts for approval
Answer "where is my order" emails	4	4	4	Second project
Compile the weekly sales report	3	5	5	Quick win, mostly rules
Chase overdue invoices	4	4	2	Later, with human approval on each message
Negotiate supplier terms	4	1	1	Not a candidate

The scores here are illustrative. Yours will differ. The point is that the ranking makes the choice obvious and defensible, and it gives you a roadmap beyond the first project.

Mistakes to avoid when choosing

• Choosing by excitement. The most interesting idea is rarely the most valuable.
• Counting time only. A short task that delays ten other people is worth more than a long one that blocks nobody.
• Ignoring the people who do the work. They know the exceptions. If they are not involved, the automation will miss them.
• Skipping the baseline. Without a before measurement, you cannot show the result afterwards.
• Trying to cover every exception. Automate the common path and route the rest to a person.

Related guides

• 10 AI Workflow Automation Examples by Department
• How to Automate Your Entire Business with AI, Step by Step
• Business Process Mapping for AI Automation: A Simple Method That Works
• AI Meeting Notes to CRM: Automating Summaries, Tasks and Follow-ups
• Service: AI Workflow Automation

Frequently asked questions

How many tasks should I automate at once?

One. Finish it, measure it and learn from it before starting the second. The inventory and scores give you the order for what comes next.

What if the highest-value task is also the riskiest?

Either keep a person approving every output, or start with a lower-risk task and return to it once you have experience. High value does not justify skipping controls.

Do I need technical knowledge to run this exercise?

No. The inventory, task descriptions and value scores need only knowledge of your own operations. Feasibility is the one area where an engineer's view helps, mainly to check what your systems allow.

How long does this assessment take?

For one team, a week of task logging plus a working session to score the list is usually enough. It does not need to be perfect. It needs to be good enough to rank the options.

Sources and further reading

Primary documentation and research behind this guide.

• BPMN.org: Business Process Model and Notation
• n8n documentation

A ranked list of tasks, with honest scores, is the most useful document you can bring to any automation project. This mapping step is where Growmerz starts every AI workflow automation engagement. If you would like help building the list for your team, book a free AI audit.