The AI revolution is here — so it’s no wonder that many managers, and their team members, are using generative artificial intelligence (AI). We already know that large language models (LLM) like Google Bard and GPT-3 can help us speed up all manner of tasks, and certain aspects of the performance review process are no exception.
But most of us have also heard of that time when ChatGPT suggested that the peregrine falcon (a bird) was the fastest marine mammal. So while generative AI does have the potential to disrupt the performance review process, it’s certainly not the complete answer.
The key is “to think of AI as your assistant, not your replacement,” recommended Theresa Fesinstine, an HR executive turned AI educator for HR teams and founder of peoplepower.ai.
With that in mind, let’s see how your new assistant can help you write performance reviews filled with genuine, specific, and useful content.
How Generative AI Can Help With Performance Reviews
Performance reviews take time. A 2017 Adobe survey of 1,500 US office workers found that, on average, managers spend 17 hours per employee just preparing for the performance review itself. But even with this significant time investment, many companies are still falling short.
WTW’s 2022 Performance Reset Survey found that only 26% of surveyed US organizations consider both their performance management and pay-for-performance programs highly effective. In contrast, 41% say they don’t believe these programs are effective at their organizations.
This combination means that many managers and employees are likely wondering if AI can help make this process easier and more effective.
“AI can be a game-changer for performance reviews, making them quicker to write and more impactful,” said Fesinstine. “Instead of spending hours trying to find the right words, AI can give you a solid starting point, offering up varied and specific phrases. So, when you're giving a shout-out to multiple employees for being team players, AI can help you say it in different ways. That keeps your praise fresh and makes each person feel uniquely valued.”
Some of the other ways that generative AI technology can be used within the performance review process include:
- Increasing the speed of writing each review
- Offering meaningful vocabulary
- Diversifying language
- Generating employee goals based on defined criteria
- Suggesting employee development plans
- Prompting ideas for key competencies
- Putting together summaries of performance data collected throughout the year
- Creating a first draft (which can then be edited)
- Developing personalized learning plans for career progression
- Helping employees summarize their achievements and goals
Nathan Deily, cofounder and chief people officer of nth Venture, mentioned that most of us have already learned that generative AI is useful for completing a specific portfolio of tasks — when it’s given the right kind of focus and instructions. But it does need to be used in the right way.
How NOT to Use AI-Powered Performance Reviews
Imagine being the employees who find out their annual performance reviews weren't even written by their managers — but by AI. That’s one surefire way to harm trust, motivation, and employee engagement (not to mention self-esteem).
And that means team leaders and employees using AI during the performance review process need to take care. Fesinstine noted that while AI can give us a helping hand, it can’t — and shouldn’t — do all the work. “Using it to spit out generic, one-size-fits-all feedback won't do your team any favors,” she cautioned.
Managers and employees should not rely on AI to create a complete performance review — but it can be used as a starting point if you’re unsure how to structure things or are facing writer’s block. This starting point should then be edited and adjusted as necessary because, as Fesinstine noted, “a performance review is a chance for real talk about how someone is doing, and that requires human insight.”
But, given the right prompts, generative AI can form part of an effective performance management system.
AI Prompts for Managers and Employees
The quality of AI output depends on the quality of your input, so don’t expect decent results unless you’re using detailed prompts.
Prompts Managers Can Try
“The first place to start in this process is to keep an ongoing short bulleted list of successes and challenges for each employee,” suggested Fesinstine. “Then in order to get the most out of AI tools like ChatGPT, Claude 2, and Bing Chat, the trick is in asking the right questions.” She recommended managers try the following prompt to kick things off:
- How can an employee improve in areas such as teamwork? Take into account that they were coached on Y feedback this year.
At Lattice, we also suggest the following prompts:
- Using the following employee traits [list traits here], offer alternative vocabulary, using a positive tone.
- Check these comments about five different employees [insert comments here] for signs of unconscious bias.
- Analyze this employee's top achievements [list them here] and pull out any key themes. Which hard and soft skills are most evident, and which may need development?
“These kinds of questions can guide the AI to give you more targeted and useful feedback that you can then tailor further,” recommended Fesinstine.
Prompts Employees Can Try
Employees can also leverage AI to help prepare for their review, using prompts like the following:
- Turn this bulleted list of successes into a narrative account of my accomplishments, making sure to highlight any key themes. [Add list here.]
- I’ve worked at my company for Y years. Last year my performance feedback included [add specific details here]. Using this information, create a list of questions my manager might ask me.
- My current job title is X with a salary of Y, and I want to work toward Z. Here are my accomplishments [add them here]. Write a request for a promotion, including reasons why I should be considered.
Remember that good prompts need good data.
Using AI might initially seem like a way to save time, but as you can see, it still requires a significant amount of data collection. To avoid generating boring, boilerplate responses, it’s worth experimenting with your prompts. Try adding as much detail as possible, but stick to clear and concise language.
As Deily noted, performance reviews are a particularly challenging domain for generative AI to work well in: “There is a lot of very specific context and nuance to be considered in what is often a very subjective process.”
Without the right data and detailed prompts, AI may be less of a helpful assistant and more of a potential risk.
Risks to Watch for When Using AI to Write Performance Reviews
We already touched on the fact that AI isn’t perfectly reliable. That means human resources leaders and legal teams need to carefully consider the risks, and how to reduce the chances of those risks impacting their people.
Bias
“Watch out for biases in the language the AI might use — it's only as unbiased as the data it was trained on,” highlighted Fesinstine.
This is well illustrated in a 2023 experiment by Textio: It found that when ChatGPT was prompted to write feedback for specific job titles, the results often showed gender bias. For example, ChatGPT used “she” 90% of the time when writing a review for a receptionist, and it used “he” in 100% of reviews generated for a construction worker.
For other roles including doctors and lawyers, the feedback was completely gender-neutral. When prompts included gendered pronouns, ChatGPT tended to write longer feedback for female employees, and generally, the additional feedback was critical.
Privacy
“Be cautious about privacy,” warned Fesinstine. “Do not include specific names of employees or any other personally identifiable information.” That’s because any information entered into AI isn’t protected by your company’s data protection regulations, so it’s best to assume it’s not secure.
Misuse of Intellectual Property
Generative AI is trained using data lakes — huge collections of information pulled from a wide range of sources. But often, the intellectual property owners of this information haven’t consented to it being used in this way. If the AI tool you’re using creates performance review comments based on examples from a copyrighted book, that may be considered copyright infringement.
Because AI is a new technology, the legal implications of using it in this way are still being established. If possible, check that the provider of any tool you’re using can confirm it was trained using open-source, non-copyrighted content.
Distrust
If managers are not open about using AI during the employee evaluation process, the trust between managers and their teams may be negatively affected. Employees may be discouraged or upset when they discover their performance appraisal wasn’t written by their manager — but by AI.
A 2021 article on the effects of using and disclosing AI-generated performance feedback found that AI feedback tends to improve employee performance more than human feedback does, but only when an employee doesn’t know the feedback was generated by AI. When this information is disclosed, job performance decreases and employees tend to trust the quality of feedback to a lesser degree.
What can HR leaders do to ensure the responsible use of AI?
It’s almost inevitable that AI will be used within at least some parts of individual performance evaluations. But it’s up to HR leaders to shape how that might look at their company and to introduce the appropriate guardrails.
“HR has the opportunity to play a lead role in the rollout and utilization of AI — in the drive of efficiency, creativity, and risk mitigation,” said Fesinstine. She recommended having clear rules around what’s okay and what’s not when it comes to using AI at work. “Training sessions can help everyone get up to speed on how to use AI responsibly. And don't just set it and forget it — keep an eye on how things are going with regular checks,” she said.
Deily added that HR’s role here should be “to reinforce the importance of the performance management task, provide some clarity on the applicability (or lack thereof) of certain AI tools, and caution against the frightening scenario of having an employee find out that their manager cared so little about them and their review that they punted the task to ChatGPT.”
Developing Policies for the Responsible Use of AI at Work
At Lattice, we created a policy covering the use of generative AI, including that ultimately, whoever creates the content effectively owns the final work output. That means they’re responsible for:
- Verifying the data sources
- Assessing and checking for any bias
- Reviewing and contextualizing the output before use
When AI Isn’t the Answer
If HR leaders have decided the use of AI for specific tasks — including performance reviews — isn’t allowed, it’s important to publicize this within the company. Clearly explaining your reasons can help managers and employees understand why they’re not yet permitted to use AI for this aspect of their job.
Some HR leaders might still not be convinced that AI can accurately summarize what makes an employee great at their job, while others may be concerned about privacy issues or the potential for bias. Others may simply prioritize the human element of this essential review process.
If you’re not permitting the use of AI, it’s worth revisiting this position regularly, since this technology changes and improves rapidly. And if HR leaders are comfortable with managers and employees using AI, it’s worth standardizing that process to some degree, too.
It’s also important for HR leaders to state whether using AI for specific tasks, including performance reviews, is simply not permitted.
Ensuring a Fair and Accurate AI-Powered Performance Review Workflow: A Manager’s Checklist
Before you use AI during the review process, we recommend you:
- Check (and follow) your company’s policy on the use of AI — which may include restrictions on its use.
- Talk to your team about how you can all use AI responsibly during reviews, and what this might look like.
- Ensure you’re using AI to build on your feedback rather than create something unspecific from scratch.
- Double-check all AI outputs with a human eye.
- Ask for employee feedback, thoughts, and opinions regularly.
- Improve or adjust your process as necessary.
Integrating AI, the Right Way
Preparing for performance reviews can be a daunting task for first-time managers — and that’s without all the questions about how AI can help.
The first step is to check what your company’s policies are around the use of AI. Then, use the sample prompts above and tweak them until you’re getting an output you’re happy with. Review that output carefully and supplement the review as needed.
For everything else, we’ve got you covered with our workbook Preparing for Performance Reviews as a First-Time Manager.
