Before adding AI, choose the job you want help with
A practical starting point for business owners considering an AI feature.
"Can we add AI?" leaves a lot to decide. Start with a task your team already spends time on.
Perhaps someone reads long enquiries and decides where to send them. Or staff search documents before answering a customer. Describe that work before choosing a feature.
Decide what an acceptable result looks like
Take a draft customer reply. Does it need to quote a policy accurately? Could it promise something you do not offer? Who checks it before it is sent?
For an early trial, I would keep the task narrow and have someone review the output. Keep example requests with expected answers, including missing information and questions your documents cannot answer. Try them again when the setup changes.
Count the work it adds
Someone needs to maintain reference information and review mistakes. Compare that effort with the time saved before expanding the feature.
NIST's Generative AI Profile discusses risks including confidently incorrect output and data privacy. Check what information the task needs and how a service handles it before sharing business or customer data.
Give people a way to correct the result
Imagine an assistant suggesting which team should handle an enquiry. If the suggestion is wrong, the person reviewing it needs an easy way to change the assignment and continue working.
Agree on who can pause the feature if it causes problems. Keep the previous process available during the trial, especially when a customer is waiting on the result.
When you review the trial, look beyond the most impressive examples. Check the ordinary requests and the ones that needed intervention. Ask staff whether they trust the suggestions enough to use them and where they still find themselves doing the whole task again.
