Which processes to automate with AI first

The instinct is to automate the exciting thing: the customer-facing chatbot, the AI feature for the pitch deck. The processes that actually pay back are usually the ones nobody wants to demo: the repetitive, high-volume, rules-heavy work happening in the back office right now.

The four questions that rank a process

Score any candidate process on four things:

• How often does it run? Daily beats monthly. Volume is where automation compounds.
• How manual is it? Copy-paste between systems, re-typing, chasing information by hand. The more human keystrokes, the more there is to remove.
• How rules-based is it? Clear inputs and a repeatable decision automate cleanly. Pure judgment with no pattern does not, yet.
• What does a mistake cost? High-stakes steps keep a human in the loop; low-stakes, high-volume steps are the safe place to start.

What this looks like in practice

Document intake keyed into a system by hand. Quotes assembled from a price list and a set of rules. Support questions answered from a knowledge base. Data moved and reconciled between two tools that were never meant to talk. Reporting someone rebuilds every Monday. None of it is glamorous. All of it is hours, every week, forever.

Start with one, measure, then widen

Pick one process. Measure the hours it costs today. Automate it, keep a human on the exceptions, and count the hours back. That number funds the next one, and it turns ‘we should do something with AI’ into a track record you can build on.