“What could we do with AI?” opens too many doors at once. A more useful starting question is: “What happens every week, has a clear result, and takes more of our team’s time than it should?”
Daily Tasks
List the work you want to improve
Ask each department head for three lists: work that keeps repeating, work that needs constant follow-up, and work that depends on one experienced person. Describe each task as an action: check quotes for missing details, organize sample feedback, or update production progress. That gives you a much clearer scope than “build a sales agent.”
Six Selection Criteria
Six checks before you choose
Frequency: The task happens often enough for savings to add up and for you to get feedback quickly.
Rules: Most decisions follow known criteria the team can explain.
Data: The required information is accessible, and the outcome can be recorded.
Risk: Errors can be detected or reversed, and actions can be held for human approval before execution.
Measurement: You can compare time spent, missed details, rework, delays, or response times.
Owner: Someone uses the output and has time to explain what is right, wrong, or missing.
Score each candidate from 1 to 5 against these criteria. Look for a combination of useful value, accessible data, manageable risk, and a person ready to own the pilot. The biggest theoretical saving can wait if those conditions are missing.
Three Starting Points
Three useful starting points
Organizing information
Pull the same fields from emails, spreadsheets, chats, or images into a structured record or task list. The output is straightforward to check, and the process can improve the quality of your records.
Checking and following up
An agent can check for missing information, exceptions, overdue work, or inconsistencies and send its findings to the responsible person. This lets you test its ability to spot problems while people keep control of the decisions.
Preparing drafts
Prepare quote replies, social posts, purchasing lists, or production reports for someone to approve. This can reduce time spent assembling a first draft while keeping final decisions with the team.
When to Wait
Tasks to leave for later
- A single failure could cause major financial, legal, or safety consequences.
- The process is still changing frequently, and departments do not agree on the objective.
- The data is scattered and unrecorded, yet the agent is expected to somehow “know” it.
- Leadership wants it, but the people doing the work have no time to help.
- The project is designed for a presentation and cannot connect to everyday tools.
Is the scope manageable? One person can explain the inputs, expected result, and exceptions. A small group can test it without disrupting the wider business. If the pilot stops, they can safely return to the previous process.
Your Starting Baseline
Measure the current process first
Before the pilot, record how long the task takes, how many people are involved, what gets missed, and where work waits. Use the same measures after launch. Staff feedback matters, but you also need to see whether the work improved.
A well-chosen pilot gives you useful results and a method for the next project. We can help you compare the candidates and define a manageable starting point.
Choose your first workflow with usNext articleWhy AI pilots stall after the demo →