Applied AI
How to choose the first process to automate with AI
By Julián Medina · Director of SourcingUp and creator of CommerceUp
Choose a frequent task with accessible information, a result you can verify and mistakes you can recover from. A focused task whose current performance you can measure is usually a good place to start.
Compare the effort across the whole workflow
Drafting sales replies, classifying inquiries and comparing proposals may all look promising. Compare their frequency, preparation and review time, and the consequences of an error. If a ten-minute task takes two hours of information gathering first, the main difficulty lies upstream. That changes what is worth automating.
An example of how to choose
Imagine a team that answers stock inquiries and negotiates bespoke contracts. Drafting an availability response from a current source produces something straightforward to check. Negotiating terms requires handling exceptions and making commitments. I would start with the draft, with someone responsible for approving it before it is sent. This illustrates how to choose a task; it is not a claim about savings.
What would justify taking the next step?
Collect both routine and difficult examples of the same task. Ask the team to evaluate the results without silently fixing the mistakes. Record missing information and the amount of human intervention needed. If only the person who designed the test can make it work, their knowledge still needs to become a shared process.
Put it into practice
- Describe a specific input and output.
- Identify the source of each piece of information.
- Choose who reviews the result and when the workflow must stop.
- Compare total time and errors with the current process.
The thinking behind this guide
A guide by Julián Medina. The scenarios are illustrative, not measured customer outcomes. My work at SourcingUp.