When I worked at a scaling startup, one of the tasks I was given was to get every manager documenting their 1:1s in our performance platform, once a month.
We had chosen the tool and set up the templates. I had written a 1:1 playbook and run training for managers on how to get the most out of their 1:1s. I was pretty pleased with myself.
But a lot of our managers kept their 1:1s in Google Docs anyway.
I could have treated this as an adoption problem: more reminders, another round of training, a nudge from leadership. But the Google Doc was doing a job for those managers that our process wasn’t, and the most useful thing I could do was find out what that job was.
The hardest part of building a solution is getting people to use it, and workarounds are what people do when it doesn’t fit how they work. I have seen the same pattern in almost every organisation since: the calibration decisions that live in a shared spreadsheet, the anniversary reviews someone tracks by hand.
What has changed is that the workaround isn’t only a doc or a spreadsheet any more. Most managers now have an AI tool open in another tab:
- 75% of knowledge workers use AI at work, and 78% of them bring their own tools rather than waiting for one from their employer (Microsoft and LinkedIn, 2024 Work Trend Index)
- In a June 2025 survey of 1,342 US managers, 6 in 10 use AI at work, and 94% of those use it to make decisions about their direct reports, including raises and promotions. Only around a third had any formal training in how to do it (Resume Builder)
So the manager who once kept their 1:1 notes in a Google Doc might now be pasting them into ChatGPT to write their performance review.
In this issue I’ll show you why that’s worth paying attention to, how to find out what people are actually doing (using a method I carried over from UX research), and how to design what you learn back into your performance process.
PREVIOUS NEWSLETTERS
In case you’re a new reader (or just missed them), here are the past few newsletters:
- Can you drive high performance AND stay human? Key takeaways from the Startup People Summit panel discussion about the future of the CPO role.
- Build or buy is the wrong question. How to choose the right tool for your people solution.
- Reviewed by someone who never saw my work. Why the org chart shouldn’t dictate how performance reviews work.
Shadow systems are a signal, and AI ones can go astray
Mark Lewis (Crewmojo’s CEO) calls these shadow systems: the spreadsheets, side docs and now AI prompts people build when the official process doesn’t fit how they work.
When your team builds a shadow system, it’s a signal you should listen to, because someone cared enough about the process to fill a gap in its design.
The instinct is to shut them down. I understand why. People teams know what happens when processes run outside the guardrails. The trouble is that a mandate doesn’t remove the need behind the workaround. It goes underground, people stop telling you about it, and you lose the insight it was giving you about the gaps in your process.
With AI there’s more at stake. Here are a few ways it can go astray when nobody has designed how AI should be used in performance reviews:
- Reviews that depend on the prompt. Two managers paste the same year of notes into ChatGPT. One asks for “honest and direct”, the other for “warm and encouraging”. Their teams get very different reviews for the same work.
- Self-reviews that all sound the same. When everyone writes theirs with AI, calibration ends up comparing how well people write.
- Only what’s visible gets counted. AI summarises whatever it’s given. If one person supplies a broader range of inputs than another, the outputs will vary widely.
- Personal data in personal accounts. A manager pastes a direct report’s performance concerns into their own AI account, and it’s now outside anything your organisation controls.
- Career plans that don’t exist. AI suggests a progression path that sounds great and has nothing to do with your roles, levels or capability framework.
I’m not saying people should stop using AI for their performance processes. We have asked before whether you would accept AI-generated feedback, and the need behind each of these is real. It’s worth designing for intentionally so it works the same way for everyone.
How to use contextual interviews to find shadow systems
If you ask people how they do something, or what they think of a process, what they tell you is usually different to what they actually do.
So much of what we do day-to-day runs on habit that we stop noticing the small steps and shortcuts we take. Which means when we’re asked to describe our behaviour, we focus on what we think others expect from us rather than how we actually behave.
That’s why I recommend running contextual interviews with your people to find shadow systems and improve your performance processes. Instead of asking people how they’re using AI in their 1:1s or their performance review, you ask them to show you how they last did that activity.
The method comes from UX research (Hugh Beyer and Karen Holtzblatt set it out in their book Contextual Design). In UX, contextual interviews usually happen in real time, while someone does the task. That doesn’t translate well to people processes, because nobody wants HR watching over their shoulder while they write their own business case for a promotion. So I adapt it: talk to people shortly after the moment, and ask them to walk you back through it with the real documents open.
1. Pick one moment. 1:1 prep, writing a review, a career conversation or calibration. One is enough to start. Book contextual interviews with a subset of the group whose experience you’re looking to improve, within a week of that activity being completed, while it’s fresh.
2. Choose four to six people in your target demographic. You might be interested in understanding the manager experience, a new joiner experience, or a high performer versus a low performer. Be clear about why you want to speak to that demographic in particular.
3. Make it safe. Start by telling them why you’re asking for their feedback. People are more likely to open up if it’s clear you’re looking to understand their experience so you can make it easier and more useful for them. Then explain how you’ll use what they share: you’re looking for themes across everyone, not making notes about them.
4. Ask them to walk you through it.
- Walk me through the last time you did this. Where did you start?
- What did you have open? What did you use AI for, and what did you do yourself?
- What did you give it (notes, Slack messages, last year’s review, the company values)? What did you leave out, and why?
- Can you show me the prompt?
- What did you change in what came back? When did you trust it, and when didn’t you?
- What would our process need to give you so you didn’t need this?
5. Collect the artefacts. With permission, save the prompts, templates and spreadsheets. A prompt someone wrote for themselves tells you exactly what context they needed and what good looks like to them.
6. Make sense of it, including the guardrails. For each workaround, write down the job it’s doing and what the official process is missing. Then ask what would need to be in place for it to work the same way for every manager and every employee. For example:
- Shared context everyone starts from, like your company values and capability framework
- A consistent tone, so the review someone gets doesn’t depend on their manager’s prompt
- Clear lines on what data goes in, and what never does
From workaround to workflow
Once you’ve finished your interviews, work through three questions for each workaround. Then pilot it with the same people you spoke to, and check back after the next cycle. You’ll know it’s working when nobody needs the workaround any more, without anyone having to ban it.
Does it still solve a real problem? Some workarounds exist because your process is missing something people need, like a place to keep running notes between 1:1s. Others are left over from an old way of doing things, like a tracker someone built for a review cycle you’ve since changed, which they keep updating out of habit. Build the first kind into your process. For the second, let the person know the original problem has been fixed, so they can stop.
Is it one person’s need or everyone’s? If the same workaround turned up in three of your six conversations, it’s a gap for everyone, and that’s where to start.
What does it become in the workflow?
- Spreadsheet columns become fields in your form
- A side meeting becomes a workflow step, triggered at the right moment
- A manual “who’s due for a promotion” tracker becomes a trigger, like a start-date anniversary
- An AI prompt a manager wrote for themselves becomes an AI step HR sets up once, with your values, tone and data rules built in, so every manager starts from the same place
- The calibration spreadsheet becomes a report you can ask questions of
A flexible platform makes this part easier. In Crewmojo you can build each of these into the workflow itself, rather than asking people to fit their work around a fixed template.
One thing to try before your next review cycle
Pick one moment, book four to six walkthroughs, and ask to see how your people approached it. Then identify the themes about where you could streamline the process for them.
If you’re rethinking what your own process asks of people, our guide on how to design a performance management process that fits your organisation walks through setting expectations, coaching and reviewing as separate decisions rather than one bundled cycle.
I’d love to hear what you find. What’s the workaround your team uses outside your performance process, and is AI part of it yet? Hit reply and tell me.
Want to talk through what this looks like for your organisation?
Book a demo