How to Find Tasks to Automate

The first question in business automation should not be, “Which automation tool should I use?”

It should be:

Which recurring task consumes enough time, follows a sufficiently stable pattern, and can be automated without creating disproportionate risk?

Solopreneurs often choose automation projects based on frustration. A tedious task feels like an obvious candidate, but frustration alone does not determine whether automation is worthwhile.

A task performed for five minutes once a quarter may be irritating but commercially insignificant. A two-minute task repeated 20 times each day may deserve immediate attention.

Finding useful automation opportunities requires observing actual work, measuring its recurring burden, and understanding how the task behaves when the normal process breaks.

What Is an Automation Candidate?

An automation candidate is a task, action, or decision that can be completed partly or entirely by software with an acceptable level of cost, accuracy, maintenance, and risk.

Strong candidates usually have several of these characteristics:

• They occur frequently
• They follow a recognizable trigger
• The required input is available digitally
• The desired output can be defined
• Most cases follow stable rules
• The task involves moving or transforming information
• Completion can be verified
• Errors are detectable
• Failed actions can be reversed
• The time saved exceeds the cost of building and maintaining the system

A task does not need to be fully predictable. Variable steps may use AI, while calculations, validations, routing, and external actions remain controlled by fixed rules.

The unit of analysis should be the individual task—not the job, role, department, or complete business process.

Analyze Tasks, Not Roles

A solopreneur’s work may include marketing, accounting, sales, support, delivery, administration, and strategic planning during the same day.

None of these roles should be treated as wholly automatable or wholly manual.

“Customer support,” for example, contains several separate tasks:

• Receiving a message
• Recording the request
• Finding the customer
• Retrieving the order
• Identifying the topic
• Checking the relevant policy
• Deciding whether an exception is justified
• Drafting a response
• Sending the approved answer
• Updating the support record

Some steps are predictable. Others require interpretation, commercial judgment, or personal communication.

This task-level distinction is important beyond one-person businesses. The 2025 ILO analysis estimated that one in four workers worldwide were in occupations with some exposure to generative AI, but only 3.3% of global employment fell into the highest exposure category. The report concluded that transformation of work was generally more likely than complete job replacement.

For a solopreneur, the practical lesson is straightforward: do not ask whether AI can automate “marketing” or “running the business.” Identify the specific actions inside that work.

Begin With Observation

Most people underestimate small recurring tasks because they remember projects, meetings, and deadlines more easily than dozens of short administrative actions.

Do not construct an automation plan entirely from memory.

Observe your work for at least five representative working days. If the business has monthly or seasonal processes, include a month-end, launch, reporting period, or other relevant cycle.

Record a task whenever you:

• Copy information between applications
• Re-enter information that already exists elsewhere
• Download and rename a file
• Reformat data
• Search for the same type of information
• Check whether something happened
• Send a recurring message
• Create a document from a standard structure
• Apply the same decision rule
• Wait for another system
• Reconcile two records
• Correct a recurring error
• Perform work because a date or event occurred
• Ask yourself, “Did I already do this?”

The objective is not to document every mouse movement. It is to identify repeated units of work with a recognizable beginning and end.

Create a Task Inventory

A task inventory is a structured record of recurring work.

Use one row per task and record:

Write task names as actions.

Weak inventory item:

“Email”

Stronger inventory items:

• Save paid invoice attachments
• Label partnership enquiries
• Send onboarding instructions
• Follow up on unanswered proposals
• Extract customer questions
• Archive completed support threads

Specific tasks can be measured and redesigned. Broad categories cannot.

Use Evidence From Existing Systems

Your business already contains evidence of repetitive work.

Calendar

Review recurring events, reminders, reporting dates, payment dates, content schedules, renewals, and administrative blocks.

A recurring calendar reminder often indicates that a trigger-based automation is possible.

Email

Look for messages with repeated subjects, recipients, attachments, questions, and response structures.

Search for phrases such as:

• “Just checking”
• “Please find attached”
• “Your order”
• “Payment received”
• “Following up”
• “Can you send”
• “Here is the report”
• “Reminder”
• “Welcome”

Repeated language may reveal a task suitable for a template, rule, or automated workflow.

Project management system

Inspect tasks created repeatedly, reopened tasks, recurring checklists, and work that waits in the same status.

Frequent manual status changes may indicate that applications are not exchanging information correctly.

Spreadsheets

Look for sheets that require repeated imports, copying, cleaning, sorting, categorization, or formula repair.

A spreadsheet may be a useful operating tool, but it can also hide a manual database or reporting process.

Accounting records

Review repeated invoice preparation, receipt collection, transaction labeling, payment checking, reconciliation, and reporting work.

Financial actions may require strict controls, but their predictable preparation steps can often be automated.

Customer enquiries

Group incoming questions by topic, required information, and resolution.

Automation becomes more viable when a large share of requests follows a common path and exceptions can be identified reliably.

Browser activity

Notice websites you repeatedly open to retrieve prices, reports, order information, analytics, or status updates.

Manual checking is a strong automation signal when the information can be retrieved through an approved integration, export, feed, or API.

Checklists and templates

A checklist describes repeated logic. A template describes repeated structure.

Both are useful starting points because part of the process has already been standardized.

Look for Automation Signals

Certain patterns frequently indicate an automation opportunity.

Repetition

The same task is completed in substantially the same way across many cases.

Frequency matters because small savings accumulate.

A task that saves four minutes and occurs 100 times per month returns more than six hours. The same saving on a monthly task returns less than one hour per year.

A clear trigger

The task starts when something observable happens:

• A payment succeeds
• A form is submitted
• A file is uploaded
• A date arrives
• A status changes
• A metric crosses a threshold
• An email reaches an inbox
• A customer completes a purchase

Clear triggers reduce the need for manual monitoring.

Copying and re-entry

Information is copied from an email to a spreadsheet, from a form to a customer record, or from one application to another.

Re-entry consumes time and creates opportunities for transcription errors.

Reformatting

The underlying information remains the same, but it must be converted into another structure:

• CSV to database rows
• Notes to a standard report
• Dates into a consistent format
• Product data into a publishing template
• Form submissions into a client brief

Rule-based transformations and structured AI extraction can both help, depending on input variability.

Recurring decisions

The same conditions are checked repeatedly.

Examples include:

• Is this order eligible?
• Does this page require updating?
• Is this invoice overdue?
• Does this enquiry meet the minimum criteria?
• Is this transaction already recorded?
• Has this metric changed enough to investigate?

If the decision can be written as stable rules, it may be automated directly. If it depends on meaning or variable language, AI may prepare or classify the case.

Waiting and checking

The task exists primarily to determine whether another event occurred.

Examples include:

• Checking whether a payment arrived
• Checking whether a page is live
• Checking whether a file was uploaded
• Checking whether a client replied
• Checking whether a service recovered
• Checking whether stock became available

The system that knows the status should ideally notify the person or trigger the next step.

Repeated interruptions

A small task may consume little active time but still fragment focused work.

Examples include checking notifications, downloading files, approving routine records, or answering predictable requests.

Measure both completion time and the cost of switching attention.

Recurring errors

Frequent corrections may indicate that a validation rule, required field, standardized input, or automated calculation is missing.

Do not automate the error. Identify why it continues to occur.

Bottlenecks

Work repeatedly waits for the same manual action.

A bottleneck may justify automation even when the action itself is short, especially when delay affects customers, revenue, or later work.

Time-sensitive work

Some tasks lose value when performed late:

• Lead responses
• Failed-payment notifications
• Delivery confirmations
• expiring-domain alerts
• Stock warnings
• Security notifications
• Broken-page monitoring

Automation can improve response time without necessarily replacing the eventual decision.

Find the Trigger, Input, Rules, and Output

A task becomes easier to evaluate when it can be described through four elements.

Trigger

What starts the task?

Examples:

• Every Monday at 08:00
• When an order is marked paid
• When a new row appears
• When an email contains an invoice
• When clicks fall by more than a defined amount

Input

What information does the task require?

Examples:

• Order record
• Customer email
• Analytics export
• Uploaded document
• Product data
• Existing policy

Rules

What determines what happens?

Examples:

• Orders above a value receive a specific delivery option
• Reports use data from the previous complete calendar month
• Refund requests above a limit require approval
• Pages below a defined traffic threshold are excluded
• Missing required information causes an escalation

Output

What must exist when the task is complete?

Examples:

• Updated record
• Delivered file
• Approved draft
• Sent notification
• Reconciled transaction
• Prioritized report

If one of these elements cannot be described, the process may require clarification before automation.

Break Processes Into Individual Steps

Do not attempt to automate a large process as one indivisible unit.

Consider creating a monthly performance report.

The process might contain:

1. Retrieve analytics.
2. Remove incomplete dates.
3. Compare the current period with the previous period.
4. Calculate changes.
5. Exclude low-volume pages.
6. Identify meaningful movements.
7. Investigate possible causes.
8. Prepare recommendations.
9. Review the analysis.
10. Deliver the report.
11. Record the recommended actions.

The first six steps are largely deterministic. Investigation and recommendations require interpretation. Review requires business judgment. Delivery and recordkeeping can become deterministic again.

This separation reveals that a process does not need to be fully automatable to create substantial savings.

Automating data retrieval and calculations may remove most of the repetitive effort while preserving the owner’s judgment where it matters.

Eliminate Before You Automate

Automation is not the first possible response to inefficient work.

Evaluate each task in this order:

1. Eliminate: Does the task need to exist?
2. Reduce: Can its frequency or scope be lowered?
3. Standardize: Can the process use one format or rule?
4. Template: Can repeated structure be reused?
5. Batch: Can several cases be handled together?
6. Delegate: Should another person or specialist own it?
7. Automate: Can software execute it reliably?

Examples of unnecessary automation include:

• Automatically generating a report nobody uses
• Sending reminders for a process that should have been removed
• Synchronizing duplicate databases
• Producing several content formats without a distribution need
• Archiving files that have no legal or operational value

Removing a task produces a 100% time saving and requires no ongoing integration.

Standardize Before You Automate

A process that changes every time is difficult to automate.

Standardization may include:

• Using one intake form
• Requiring consistent filenames
• Defining mandatory fields
• Storing information in one authoritative location
• Applying one status system
• Creating a decision table
• Setting a standard delivery sequence
• Documenting approved exceptions
• Defining a completion check

If customers submit project information through email, voice notes, direct messages, and documents, automation must handle several inconsistent inputs.

A single structured form may solve most of the problem before any sophisticated system is introduced.

Measure the Recurring Burden

Frequency and duration determine the gross automation opportunity.

Use:

Monthly task burden = Monthly occurrences × Average minutes per occurrence

Divide the result by 60 to convert it into hours.

Also include:

• Time spent locating information
• Time spent correcting errors
• Time spent checking completion
• Time lost waiting for a manual step
• Rework caused by delayed or missing information
• Attention lost through interruptions

A task performed 200 times per month for two minutes has a monthly burden of 400 minutes, or 6 hours and 40 minutes.

A task performed once per month for 30 minutes has a monthly burden of only 30 minutes.

The second task may feel larger, but the first usually offers greater recurring leverage.

Estimate the Automatable Share

Few tasks lose 100% of their manual effort.

Separate the task into:

• Work software can execute
• Work software can prepare
• Work requiring review
• Work involving exceptions
• Work that must remain human

Use:

Estimated time saved = Current time − New review time − Exception time − Maintenance time

If a ten-minute task becomes a seven-minute review, the practical saving is only three minutes.

If automation eliminates eight minutes in 90% of cases but creates 20 minutes of correction in 10%, calculate both outcomes before assuming a benefit.

Use an Automation Opportunity Matrix

Classify tasks by recurring burden and process clarity.

The strongest early candidate is a high-burden task with a clear process.

A high-burden but poorly understood task deserves attention, but immediate automation may reproduce inconsistent decisions. Study the process first.

A clear but low-burden task should be automated only when setup is nearly effortless or the automation improves something beyond time, such as accuracy or response speed.

Score Automation Candidates

Use a 1–5 score for each dimension.

Do not use the total score blindly.

A task with a high time-saving score but serious legal, financial, safety, or reputational consequences may still be a poor candidate for independent execution.

Use the score to compare opportunities, then apply a separate risk filter.

Apply the Risk Filter

Before prioritizing a task, ask:

• Could an error move or lose money?
• Could it expose personal or confidential information?
• Could it create a contractual promise?
• Could it publish false information?
• Could it damage an important customer relationship?
• Could it delete or overwrite valuable data?
• Could it affect health or safety?
• Is the action legally or professionally regulated?
• Can the action be reversed?
• Would the error be detected quickly?
• Is a reliable manual fallback available?

High-risk tasks may still contain automatable preparation steps.

For example, software can retrieve an invoice, check required fields, detect duplicates, and calculate a proposed payment amount. A person can approve the transfer.

The goal is to remove repetitive work without removing necessary accountability.

Consider Data Readiness

Automation requires dependable inputs.

A task is easier to automate when:

• Information is already digital
• Fields have consistent meanings
• Records have unique identifiers
• Dates and currencies follow standard formats
• The source of truth is known
• Permissions are available
• Data is sufficiently complete
• Historical examples exist for testing

A task is harder when information is split across personal notes, email, unstructured files, memory, and several conflicting spreadsheets.

Data readiness matters particularly for AI-assisted processes. A 2026 business data study found that 82% of AI-using businesses in its sample analyzed data, compared with 68% of businesses not using AI. The result is an association rather than proof of causation, but it reinforces the operational connection between usable data and meaningful AI adoption.

Before automating, determine which source is authoritative when two systems disagree.

Choose the Appropriate Intervention

Finding an automation opportunity does not mean that full automation is the correct response.

Use the smallest intervention that solves the problem.

A template may save almost as much time as a complex integration while remaining easier to maintain.

Good First Automation Candidates

Early automation projects should be frequent, visible, reversible, and easy to verify.

Strong examples include:

• Saving form submissions to a structured table
• Delivering a purchased digital file
• Creating recurring invoices
• Renaming and organizing downloaded reports
• Recording confirmed payments
• Sending internal deadline reminders
• Backing up important data
• Retrieving weekly analytics
• Detecting broken links
• Creating tasks from approved form submissions
• Updating a status after an external event
• Notifying the owner when a threshold is crossed
• Generating a report from verified calculations
• Applying standard labels to structured records
• Confirming receipt of an enquiry

Successful early projects teach the owner how integrations fail, how exceptions appear, and how much maintenance is actually required.

Tasks That May Benefit From AI

AI becomes relevant when a useful task contains variable language, documents, images, or interpretations.

Possible candidates include:

• Classifying enquiries by meaning
• Extracting fields from differently formatted documents
• Summarizing customer feedback
• Grouping search queries by intent
• Identifying recurring support topics
• Comparing a document with a checklist
• Preparing a draft from approved source material
• Detecting unusual records for review
• Translating standardized content
• Explaining differences between reports

In 2026 adoption research, marketing and administration were each cited by 72% of current or prospective business AI users, while IT was cited by 64%. These broad categories contain large amounts of text, data processing, classification, and recurring administrative work.

However, the popularity of a category does not establish that every task inside it should use AI. Each task still requires its own business case and risk assessment.

Tasks to Keep Manual

Automation is usually a weak choice when:

• The task occurs rarely
• The process is still changing
• The objective is unclear
• Success cannot be evaluated
• Each case requires a substantially different approach
• The owner’s personal attention is part of the value
• Errors would be expensive or irreversible
• The required information is unavailable
• The workflow depends on fragile access
• Maintenance would exceed the time saved
• The owner cannot recognize an incorrect result

Tasks commonly kept manual include:

• Choosing business direction
• Setting important prices
• Making exceptional customer promises
• Negotiating sensitive agreements
• Resolving serious disputes
• Publishing personal opinions
• Evaluating professional advice
• Making consequential financial decisions
• Handling vulnerable customers
• Ending an important client relationship

Software may prepare information for these decisions without making them.

Beware of False Automation Opportunities

Some tasks look suitable but fail economically or operationally.

Annoying but rare

Frustration makes the task memorable, but its annual time burden is insignificant.

Frequent but already fast

Reducing a ten-second action may not justify an integration unless the volume is exceptionally high.

Easy normal cases with difficult exceptions

The standard path may take little time while exceptions create most of the effort. Automating only the standard path may produce limited savings.

Unstable process

The workflow changes so frequently that maintenance consumes the benefit.

Inaccessible system

A critical application may lack a reliable integration or approved method of access.

Unverifiable completion

The automation cannot confirm whether the required external action succeeded.

Output without a consumer

The system creates reports, summaries, or alerts that do not influence any decision.

AI used as a substitute for missing rules

If the business has not decided its refund policy, lead criteria, editorial standard, or pricing boundaries, AI cannot make the underlying policy coherent.

Create an Automation Candidate Brief

Before selecting a tool, write a short brief.

Task: What specific action is being considered?

Business purpose: Which cost, delay, error, or constraint should improve?

Trigger: What starts the task?

Frequency and volume: How often does it occur?

Current time: How many minutes does each case require?

Input: Which information is required?

Output: What must be created or changed?

Rules: Which conditions determine the outcome?

Exceptions: Which cases leave the standard path?

Systems: Which applications and accounts are involved?

Risk: What could happen if the result is wrong?

Reversibility: Can the action be undone?

Completion evidence: How will success be confirmed?

Expected saving: How much time or money should be saved?

Maintenance owner: Who will repair the workflow when something changes?

Stop condition: When should the automation be paused or retired?

If the brief cannot be completed, the task may not yet be understood well enough to automate.

Validate the Candidate Before Building

Run a small manual simulation.

Take 20–50 recent examples and apply the proposed rules.

Record:

• How many follow the standard process
• How many require exceptions
• Which information is missing
• How often records conflict
• Whether the final outcome can be verified
• Which cases require human judgment
• What an error would have changed

A workflow covering 90% of cases may be valuable when the remaining 10% are identified and escalated reliably.

A workflow covering 99% may still be unacceptable if the final 1% includes large payments, confidential information, or irreversible deletions.

Coverage and risk must be evaluated together.

Establish a Baseline

Measure the existing process before changing it.

Useful baseline metrics include:

• Tasks completed per week
• Median completion time
• Active minutes per case
• Error rate
• Rework rate
• Delay before starting
• Customer response time
• Number of missed tasks
• Cost per completed task
• Percentage requiring exceptional handling

Without a baseline, the business cannot distinguish real improvement from the impression that a new tool feels faster.

Calculate the Business Case

Use:

Monthly gross saving = Occurrences × Minutes saved per occurrence ÷ 60

Then calculate:

Monthly net benefit = Value of saved time − Software cost − Review cost − Exception cost − Maintenance cost

Finally:

Payback period = Setup cost ÷ Monthly net benefit

If the monthly net benefit is zero or negative, the automation does not pay back through time savings.

It may still be justified by faster customer response, fewer errors, reduced risk, or improved business continuity. Those benefits should be defined explicitly rather than assumed.

Prioritize an Automation Portfolio

Do not build several automations simultaneously.

Rank candidates in this order:

1. High burden, clear process, low risk
2. High burden, clear process, moderate risk with approval
3. Moderate burden, very low setup effort
4. High burden, unclear process requiring standardization
5. Low burden or high complexity
6. High-risk tasks with weak controls

Begin with one task whose result can be measured within a short period.

A working automation that saves two hours every week is more valuable than an ambitious system that remains unfinished.

Introduce Automation in Stages

A task does not need to move directly from manual work to independent execution.

Use progressive stages:

Observe

Record what the person currently does.

Prepare

Let software retrieve data, calculate values, or assemble the required context.

Propose

Let the system recommend an action without changing anything.

Approve

Require the owner to confirm the proposed action.

Act within limits

Allow approved, low-risk, reversible cases to complete automatically.

Monitor

Track outcomes, failures, exceptions, and changes in performance.

The stages provide evidence about reliability before greater authority is granted.

Monitor the Task After Automation

An automation candidate does not remain suitable forever.

Applications change. APIs fail. Business policies evolve. Customer behavior shifts. AI models and data distributions change.

Monitor:

• Completion rate
• Failure rate
• Exception rate
• Human correction rate
• Duplicate actions
• False completion
• Time saved
• Review burden
• Cost per successful task
• Unexpected side effects

The 2026 NIST monitoring paper notes that deployed AI performance can change because of factors including system updates, changing data, user behavior, attacks, and wider environmental changes. Monitoring must therefore produce information that leads to a defined action, such as investigation, restriction, correction, or suspension.

An alert without an owner or response rule is only another source of noise.

Conduct a Quarterly Automation Audit

Once per quarter, review:

• Which manual tasks increased in volume
• Which recurring tasks appeared
• Which automations saved measurable time
• Which created excessive exceptions
• Which workflows no longer serve a useful outcome
• Which manual approvals are now routine
• Which systems or policies changed
• Which tasks should be eliminated instead
• Which automations should be simplified or retired

Also review new services, products, markets, and content processes. Business changes create new work patterns that may not have existed during the previous audit.

Do not judge the automation portfolio by the number of active workflows. Judge it by the recurring burden removed without unacceptable risk.

Frequently Asked Questions

How do I identify tasks that can be automated?

Observe your work over several representative days and record repeated actions, triggers, inputs, rules, outputs, exceptions, time, and error consequences. Prioritize frequent tasks with stable processes, digital inputs, verifiable outcomes, and low-cost failures.

What tasks should a solopreneur automate first?

Start with high-frequency, low-risk, reversible tasks such as moving data, sending internal reminders, retrieving reports, organizing files, delivering approved digital products, creating recurring records, and monitoring clear thresholds.

How repetitive should a task be before automating it?

There is no universal frequency. Multiply occurrences by time per occurrence and compare the annual burden with setup, review, exception, and maintenance costs. A short task may justify automation when its volume is high.

Should I automate a task I dislike?

Only when the recurring burden and business case justify it. A disliked task may be rare, risky, or cheaper to complete manually. Frustration is a discovery signal, not a prioritization method.

Can a process be partially automated?

Yes. Partial automation is often preferable. Software can retrieve information, perform calculations, prepare documents, and record outcomes while a person handles judgment, exceptions, and approvals.

What is an automation audit?

An automation audit is a structured review of recurring business tasks to identify work that can be eliminated, standardized, templated, batched, delegated, or automated.

How long should I track tasks?

Track at least five representative working days. Include monthly, quarterly, seasonal, reporting, or launch activities separately because they may not appear during a normal week.

How do I calculate automation ROI?

Estimate the monthly value of time saved, then subtract software, review, exception, and maintenance costs. Divide the setup cost by the resulting monthly net benefit to estimate the payback period.

Which tasks are poor automation candidates?

Poor candidates include rare tasks, unstable processes, unclear decisions, relationship-sensitive work, irreversible actions, high-risk judgments, and tasks whose results cannot be verified.

Do I need AI to automate business tasks?

No. Many useful automations use schedules, formulas, database rules, integrations, scripts, and conditional logic. AI is relevant when a step requires interpretation of variable language, images, documents, or patterns.

How do I know whether an automation is working?

Compare it with the original baseline. Measure completion rate, time saved, error rate, exception rate, correction time, cost per successful task, and the business outcome the workflow was intended to improve.

When should an automation be removed?

Remove or suspend it when the underlying task disappears, maintenance exceeds its value, the process changes materially, failures become unacceptable, or the output no longer supports a useful decision.

Automate the Right Work

The largest automation opportunity is not always the most visible task.

Useful candidates often hide inside small actions repeated throughout the week: retrieving information, copying records, checking statuses, applying rules, preparing standard outputs, and confirming completion.

Observe those actions before selecting technology. Measure their true recurring burden. Eliminate unnecessary work, standardize what remains, and separate predictable execution from interpretation and judgment.

The objective is not to automate as much of the business as possible. It is to remove recurring friction while preserving the judgment, relationships, and decisions that make the business valuable.