You’ve identified a workflow or process that looks like a good candidate to automate with AI.
A natural question is: at what point should a human come in to review its work?
For example, imagine you’re going to use AI to respond to a sales enquiry. You consider two ways to design it:
- The AI reads the customer’s enquiry, identifies the product, checks its details and then prepares a quote for an account manager to review
- The AI does all of this, but instead of sending the quote to an account manager for review, it sends it directly to the customer, including a price and delivery date
In the first case, the AI is preparing information that will be validated by a human before it goes to the customer. In the second, it has directly sent the customer something they may rely on and act on.
To make this tangible, what if the price the AI produced was incorrect in both cases?
This is why, when redesigning a process or workflow with AI, it’s important to consider the points at which you will have a human in the loop.
If you involve a human too much, you lose the time you saved by automating. If you involve one too little, you leave yourself exposed to costly mistakes that then need to be corrected.
In our example, you might decide that routine enquiries within agreed pricing and availability rules can proceed automatically, while unusual discounts or delivery commitments require human review.
The answer doesn’t necessarily depend on one fixed point in the process. It may depend on the value of the quote, the AI’s level of confidence, whether the request falls outside agreed rules and how difficult the action would be to reverse.
So, as you move forward, keep this question in mind: where should AI automation assist us, and where should a human intervene?