# Should I Automate This Task? A Yes/No Flow

A decision tree for any recurring task you are considering automating. Print this out, walk through it with the team that does the work today, and you will know whether automation is the right call.

> Brought to you by ByteWorthy. byteworthy.io

## Walk through these questions in order

### 1. How many times per week is this task done?

- **Fewer than five times per week.** STOP. The build cost will not pay back. Keep doing it manually.
- **Five or more times per week.** Continue to question 2.

### 2. Does the task have predictable inputs and predictable outputs?

- **Yes, both are predictable.** Continue to question 3. (Rules-based automation is the right path.)
- **Inputs are unpredictable, or outputs require judgment.** Continue to question 4. (You will need a human review checkpoint or an AI layer.)

### 3. (Predictable path) Are the rules written down?

- **Yes.** Build a rules-based automation. This is the cleanest case. Estimated cost: $3,000 to $7,000.
- **No.** STOP. Document the rules first. Automation cannot work against rules that only exist in someone's head. Once documented, return to this question.

### 4. (Judgment path) Is the judgment call low-stakes or high-stakes?

- **Low-stakes (worst case is rework, not harm).** Build an AI layer with a human review checkpoint. Estimated cost: $5,000 to $12,000.
- **High-stakes (worst case affects patient care, legal liability, or financial damage).** Continue to question 5.

### 5. (High-stakes path) Can the AI's recommendation be reviewed by a human before action?

- **Yes, every recommendation goes to a human.** Build it. AI as a draft generator, human as final approver. Estimated cost: $8,000 to $20,000.
- **No, decisions must be automated end to end.** STOP. Do not automate. The risk of an automated mistake outweighs the time saved. Keep this task manual.

## Worked example

**Task:** Generating prior authorization appeal letters for denied dental claims.

- Question 1: How many times per week? 8 to 12 appeals per week. **Continue.**
- Question 2: Predictable inputs and outputs? Inputs (denial reasons, patient files) are structured. Outputs (appeal letters) follow a template but require some judgment about which clinical evidence to include. **Judgment path.**
- Question 4: Stakes? Low to medium. Worst case is the appeal fails and we file again. **Build with human review.**
- Question 5 (skipped because we are on the medium-stakes path): n/a

**Verdict: Build it. Estimated cost: $5,000 to $12,000. Expected ROI: 4 to 6 hours of admin time per week, 2x to 3x increase in appeal success rate from consistency.**

## When to come back to this flow

- Every quarter, walk through your team's repeated tasks. The threshold for automation drops as your team grows.
- Every time a process changes meaningfully, re-run the flow. A predictable task can become unpredictable when an upstream system changes.
- Every time you hire someone new for repetitive work, ask if automation could replace the headcount cost.

## Want help running the flow?

We do this with clients in the 00-discovery stage of every engagement. byteworthy.io/services/automation