How to Choose Your First Process to Automate With AI

The first AI project sets the tone for everything that follows. Pick well, and you build confidence, internal support and a reusable foundation. Pick badly, and AI becomes the thing that “didn't work here”. The good news is that the best first projects share a recognizable profile.

Look for these signals

Strong first candidates usually have most of the following characteristics:

• High volume: the task happens many times a week, not once a quarter.
• Repetitive structure: the steps are broadly the same each time, even if the inputs vary.
• Unstructured inputs: emails, PDFs or free text that rule-based tools struggle with.
• A clear owner: someone who feels the pain and can judge whether the output is right.
• A measurable cost today: hours spent, turnaround time or error rates you can count.
• Recoverable mistakes: errors can be caught by review before they cause harm.

Score your candidates

List five to ten candidate processes and score each from 1 to 5 on four dimensions: business value, technical feasibility, data readiness and risk (where 5 means low risk). Multiply value by feasibility, then use data readiness and risk to break ties.

The winner is rarely the most exciting idea. It is usually a document-heavy or inbox-heavy process that a small team handles every day.

Define success before you build

Measure the current state first: how long the task takes, how often it happens and how many errors occur. Then agree what “good” looks like, for example a target share of items handled without manual entry, with every exception routed to a person. Without a baseline, you cannot prove value later.

Keep people in the loop at the start

Launch with review on every output, then reduce review where the data shows the system is reliable. This staged approach builds trust and gives you the evidence to expand.