An Odd Concept: AI for Employee Feedback?
It sounds somewhat odd or unauthentic that AI could be used for employee feedback purposes. My initial reaction was how can you share feedback that hasn't really come from you? What if the recipient knew it was generated by AI? It doesn't have a good 'feel' about it.
The Experiment: Giving AI a Try
However, driven by curiosity, we've been conducting experiments over the past six months and have seen some promising results.
Most feedback models assume people can articulate their thoughts and feelings. I have used multiple models and it still takes me a big effort to contextualise what's on my mind into a 'feedback package' that is received as beneficial for the recipient.
The need for constructive feedback is typically triggered when expectations aren't met, whether it's related to deliverables or behaviours. The immediate focus often shifts to what's wrong, making it easy to either point out faults or stay silent when unsure of how to provide helpful feedback.
Reframing with AI: Making Feedback Positive and Actionable
Through refining our AI prompts we are now achieving consistently good reframes for when employees or managers encounter unmet expectations. This might be as simple or broad as 'poor quality code'. From here we are able to use AI to reframe this into positive, actionable feedback delivered with the SBI model, for example:
"In our recent project, I noticed the code you delivered had some issues in terms of quality.
There were instances of unclear variable naming, lack of comments, and unhandled exceptions, which can lead to maintainability and stability problems down the line.
This can make it harder for other team members to understand and maintain the code, which can in turn lead to delays in project timelines. It may also potentially result in bugs and application crashes, which could affect our end-users negatively.
It might be beneficial if you spend more time on code reviews before committing your code. It would also help if you could adopt practices like clear variable naming, writing meaningful comments, and handling all possible exceptions. We have resources on these topics I think you would find useful. I'm confident that with some focused effort, you can significantly improve the quality of your code."
Not Quite Perfect: Adjusting AI-generated Feedback
The output may not be a perfect match for the specific situation and will no doubt need adjustments, but it does a good job of contextualising negative sentiments into positive feedback. Words can be swapped out for greater relevancy and may also be used as a conversation guide when the feedback is better delivered in person.
I'm curious to hear thoughts from the HR community on the use of AI in this context as we continue to explore this approach.