Start with the work, then choose the AI.
Before choosing a model or planning an assistant, spend time with the people doing the work. Their day is a much better starting point for deciding what to build.
Follow one task from beginning to end.
Pick a task that happens often. Ask someone to walk you through the last time they did it: where the information came from, which tools they opened, who they waited for, and how they knew they were finished.
Look closely at the handoffs. Repeated copying, searching, summarizing, and checking can reveal useful opportunities. They can also reveal a simpler process problem that should be fixed first.
Understand what makes the task difficult.
A repetitive task is not automatically a good candidate for AI. It may depend on information nobody has recorded, judgment that is hard to explain, or an exception that changes everything.
Find out what good work looks like and what happens when something goes wrong. That helps define where AI could assist, where rules are enough, and where a person should remain responsible.
Describe the improvement before the solution.
Write down the change you want in plain language: give a support specialist the right context before they draft a reply; help an account manager prepare for a customer meeting; make an approved answer easier to find.
Then establish a baseline. How long does the task take today? Where do mistakes happen? What would the team actually value? With those answers, you can compare possible solutions against a concrete need.
Our starting point
In an opportunity sprint, we bring workflow discovery and early prototyping together. The aim is to leave with an opportunity you can explain, assumptions you can test, and a next step that fits the business.