How To Succeed In Performance Reviews When AI Holds The Pen
AI is influencing your performance evaluations. It is writing summaries, development plans, and creating distributions. Here's how to manage it and get ahead.
Now Enrolling for the April cohort of my leadership accelerator program for mid-career professionals who want to double down on the soft skills that get them ahead in the age of AI. Proven curriculum and well loved by past students. We have 2 spots left! Learn more here.
I am a coach to senior leaders and executives at top Silicon Valley companies like OpenAI, Google, Amazon, and Instacart, who manage teams of 50+ to 2000+ people. This performance review season, they all used AI to supplement their work. Some use it to draft written feedback for their team. Some use it to suggest an improvement plan. Some use it to stress test an on-the-border rating. Others use it to find teams with ratings that deviate from an expected average. It’s no longer a question whether AI models are influencing performance conversations and evaluations. It does. And it is also a reality you need to get ahead of.
Where AI Strengthens Performance Reviews
Humans are biased, forgetful, and emotional. There are some areas where AI does better than humans when it comes to measuring value delivery and identifying areas for growth:
Humans have recency bias: we tend to forget easily and overindex on recent events. We remember the last product launch or the latest exchange we had with someone. We are not great at remembering what happened 12 months ago when the strategy was first decided, or those wins that came early in the quarter.
Humans have negativity bias: we tend over-emphasize negative events and generalize them. That one time you forgot to send meeting notes? Now you’re someone who’s not good at following up.
Humans have similarity bias: we overvalue the work of people who we see as similar to us. Two of your team members each had successful product launches. One reminds you of yourself when you were younger. As a manager, you are likely to give more credit to that person.
AI has the advantage in that it can process a large amount of context and store it over time. It doesn’t treat two people differently if the data is the same. And it can quickly analyze patterns and spot variations.
Where AI Can Hurt Fairness And Adverse Incentives
What’s also important to know is that AI has its own set of biases, most of it arising from imperfect context. Here are some areas where AI tends to be more biased than humans:
AI overemphasizes measurable activity: Since the AI is not (yet) a person who wanders the halls, catching nuanced conversations and interactions, it tends to overemphasize what is logged and written down (and accessible to it). Small gestures and relationship-building work tend to get overlooked.
AI may decontextualize a signal: A teammate who goes against the rules in an exceptional situation may get dinged for not following protocol. A team effort on a project may be falsely attributed to the person who wrote up the final report. AI currently isn’t great at judgment calls that require human interpretation.
AI can amplify existing bias in the system: If the company has 5 performance metrics related to short-term revenue and 1 metric on long-term investment, AI will over-value short-term gains in performance analyses. If the AI doesn’t measure team members helping each other, it will reward the lone-wolf mentality.
Similar to choosing a metric for a company goal, any AI (metric) is an imperfect measurement of reality. Any AI (metric) can and will be gamed. Knowing how an AI (metric) might be gamed and staying focused on the intention behind the AI (metric) is critical to staying on the winning path.
How To Set Yourself Up For Success
With AI as a core part of performance reviews, the key to managing it effectively is giving it context and data that accurately measure your performance. More than ever, what is tracked, measured, and written will influence the outcome.
Keep a log of tasks completed in your AI tool: use a combination of daily voice note dumps, meeting notes, and dumps from calendar and email. If you keep timesheets or track work in another software, do regular exports. You want to avoid your manager getting an AI summary that doesn’t include a project she thought you did, and doubting whether she remembered it correctly.
Make it measurable.
Know what proxy metrics the AI in your organization uses for performance, and focus on how you are being measured for those. If you’re not sure, start by just asking it. For example, some companies use Slack or email response time for “engagement.” Other track “token used” or “app run time” for AI expertise. Figure out what matters.
Figure out what you’re doing that’s not measured. Then get it measured, or consider dropping it. If it’s not tracked or visible, it didn’t happen.
Complement AI with People: Have AI do your own review from your manager’s and skip’s perspectives. Know where it says your weak spots will be based on the data and context it has access to. Then, proactively address this through storytelling and influence with critical stakeholders at the company. This way, the human judgment with AI support will provide the complete picture.
Just get started: This is one of those the sooner you do it, the more helpful it’ll be scenarios. The log is more valuable the longer it is kept. The proxy metrics will be more improved the longer you work at it. The new metrics will gain more credibility the sooner they are implemented. Perfection is the enemy.
AI brings a lot of benefits to the performance review process: time savings, context retention over time, processing large amounts of information quickly, and being blind to organizational politics. However, without careful management, AI will amplify its own bias, at times to your detriment. It’s no longer a question of whether you need to account for AI as a stakeholder in your promotion process: the question is how to do it in a way that helps you and your organization succeed.
That’s all folks! See you next week at 3:14 pm.
Yue
I tried one pretty interesting case. For a report who was on the fence between meets and exceeds expectations, I asked the AI model for its take. It not only helped me justify the meets rating, but also helped me fine-tune my message to encourage them to keep up their growth momentum.
That said, I wouldn't recommend this for every manager. I'm a "strong opinions, loosely held" type of person. So I trust myself to leverage the help without being overly influenced by it.