Process Selection: What Can and Cannot Be Automated

Not all tasks should be automated. The best RPA candidates are high-volume, rule-based processes that work with structured digital data in stable software environments.

03 / 8

By the end of this lesson you can

Separate rule-based work from judgement-based work
Tell structured data apart from unstructured data
Score a process on volume and application stability before writing any code

The problem

Poor Candidate (Requires Human Judgment)

Negotiating customer contract terms, reviewing artistic marketing banners, or responding with empathy to upset customer complaint letters.

The fix

Prime Candidate (Deterministic & High Volume)

Extracting standard vendor invoice details, reconciling payroll spreadsheets against bank statements, or entering approved purchase orders into SAP.

Deep Dive

01Rule-Based vs Judgment-Based Processes

Before building an automation, you must determine whether the task is suitable for a robot.

• Rule-Based Tasks (Automate these): Processes where every step has a clear answer. If you can draw the entire process on paper using simple yes-or-no branches, a robot can handle it. • Judgment-Based Tasks (Keep these with humans): Tasks that depend on empathy, taste, negotiation, or personal experience. For example, resolving an upset customer's complaint requires emotional understanding that a robot does not have.

If/Then Logic

Every path must be clearly defined in advance.

Exception Pathways

Known edge cases must have a predefined recovery rule.

Human Escalation

Tasks requiring subjective approval should route to a human supervisor.

02Structured vs Unstructured Data Requirements

A robot needs to know exactly where to find information.

• Structured Data: Information stored in tidy, predictable formats—such as Excel sheets, database tables, or digital web forms. Every piece of data is in a known location, making it easy for the robot to read. • Unstructured Data: Information that arrives without a standard layout—such as casual customer emails, hand-written notes, or voice recordings. A standard robot cannot easily understand where the key information is without extra AI assistance.

Structured Inputs

Spreadsheets, CSVs, database tables, and consistent web forms.

Semi-Structured Inputs

Standard digital PDF invoices and receipts (handled via OCR / Document Understanding).

Unstructured Inputs

Casual emails, phone recordings, and handwritten doctor notes.

03Stability and Volume: The Feasibility Scorecard

Two business conditions make an automation worth building:

• High Volume: The task must happen frequently enough to justify building an automation. Automating a job done twice a year saves almost nothing; automating a job done 400 times every day saves thousands of hours a year. • Application Stability: The software being used should have a stable design. If a website changes its button positions every two weeks, the robot will require continuous maintenance.

Candidate Category	Real-World Workflow Example	Feasibility Verdict
High Volume + Structured	Entering 500 vendor invoice PDFs daily into SAP ERP	Ideal Candidate (High ROI, Fast Payback)
High Volume + Rule-Based	Daily bank reconciliation between core banking and Excel	Ideal Candidate (Zero Typos, Immediate Compliance)
Low Volume + Unstable	Filling a quarterly survey on a portal redesigned monthly	Poor Candidate (Maintenance exceeds time saved)
Judgment-Based + Subjective	Evaluating candidate resumes and conducting culture interviews	Do Not Automate (Requires human nuance and empathy)

Scroll table sideways→

Vocabulary you now own

Structured data: Information in predictable rows, columns or fixed fields. The robot always knows where the amount and the date live.
Unstructured data: Free-form content such as a casual email or a handwritten note. Needs AI or document understanding before a robot can act on it.
Application stability: How rarely the target screens change. Frequent redesigns mean frequent automation maintenance.

Remember this much

1If a step needs empathy, taste or negotiation, it is not an RPA step.
2High volume is what turns saved seconds into a business case. A monthly task rarely pays back the build.
3A beautiful automation on an unstable screen becomes a permanent maintenance ticket.