Choosing What to Automate First: A Simple Scoring Method

By Camderson Academy

When teams first get excited about automation, they usually start in the wrong place. Effort goes into the impressive idea instead of the recurring task that would actually pay back. Or nobody measures how the work runs today, so nobody can say whether the automation helped. Choosing well is most of the battle, and this simple scoring method helps you do it in an afternoon.

Start with an inventory, not an idea

List the recurring work in your area: reports, inbox triage, data entry, follow-ups, approvals, reconciliations. For each item, note roughly how often it happens and how long it takes. Do this with at least one person who does the work. People who do a task every day see details that people who manage it often miss.

Score each candidate on four fits

For each task, give a score from 1 to 5 on four questions:

• Frequency: how often does it happen? A daily task scores higher than a quarterly one.
• Format: how consistent and digital are the inputs? The same spreadsheet every week scores high. Free-text emails in many styles score lower.
• Rules: how clear are the decisions? If a new hire could follow the steps as simple if/then rules, it scores high. If it needs judgment every time, it scores low.
• Risk: how low is the cost of an error? This one runs in reverse: 5 means low risk. Internal work that is easy to fix scores high. Anything touching money, safety, legal matters or people's jobs, and hard to undo, scores low.

Add the four scores for a total out of 20. As a rule of thumb, 15 or more is a strong candidate, 10 to 14 is possible with human review, and under 10 is best left manual for now.

The most common scoring mistake is reading Risk the wrong way round. Score it as "how safe is this?" or you will push the riskiest work to the top of your list.

A worked example

Imagine an operations lead at a small services company scoring three candidates:

The weekly summary is the clear first pick, and it needs no AI at all, because fixed rules can do the calculation reliably. Inquiry sorting is a reasonable second project with a person checking each label at first. Refund approvals stay manual: they involve money, need judgment and are hard to undo.

Score the design you plan to build

Risk depends on what the automation is allowed to do. If AI only sorts and logs incoming emails for a person to handle, a wrong label costs that person a minute. If AI sends replies straight to customers, a wrong answer reaches a customer. Score the version you actually intend to build. When a total lands just below a threshold, ask whether a narrower design, with fewer actions or a human check before anything goes out, would earn back a point of Risk.

A high score does not mean you need AI

Some of the best candidates need no AI at all. AI earns its place where inputs vary and need interpreting, such as classifying requests, pulling details out of messages or drafting a first response for someone to review. A useful way to think about it is a ladder of independence:

1. Ask: you prompt an AI assistant and act on the answer yourself.
2. Automate: fixed rules move information with no AI involved.
3. AI-assisted workflow: fixed steps, with AI doing one step, such as classify, extract or draft.
4. Agent: AI chooses its own steps and uses tools toward a goal.

Climb only as high as the task needs. Every step up calls for more oversight.

Measure the starting point, then start small

Before you build anything, record how the work runs today: how long it takes, how often it goes wrong and how long errors take to fix. Without that baseline, you cannot show the automation helped, and you cannot spot it quietly making things worse.

Then pick one strong candidate, design it on paper first, including what happens when something unexpected arrives, and pilot it with a human check before anything leaves the building. Name one owner who is responsible for it. Expand only once the pilot has run cleanly for a few weeks.

For organizations formalizing their approach, NIST's free AI Risk Management Framework (AI RMF 1.0), released in January 2023 for voluntary use, is a helpful reference. Its core is organized around four functions: govern, map, measure and manage.

Template: automation opportunity scorecard

The takeaway

Do not start with the flashiest idea. List your recurring work, score it on Frequency, Format, Rules and Risk, use AI only where inputs need interpreting, measure the baseline and pilot one small win with a person in the loop.

Sources

• National Institute of Standards and Technology, "NIST Risk Management Framework Aims to Improve Trustworthiness of Artificial Intelligence", January 26, 2023

This article shares general practice for choosing automation projects. Follow your organization's IT, security and AI policies.