In early 2012, Donna Morris, Adobe’s senior vice president of global people resources, was on a trip to one of the company’s offices in India. An Indian newspaper reporter asked her: “What is new or innovative in human resources these days?”
“Well,” she said, “we plan to abolish the performance review format.”
The line ran on the paper’s front page the next day. According to an account on the website of the investor John Doerr, she was jet-lagged, and she hadn’t yet discussed the idea with her HR staff or with Adobe’s CEO.
Now something had to take the review’s place.
Silence all year, then a verdict
What Adobe built was called Check-in. It dropped the annual review, the ratings, the rankings and the long appraisal forms, and replaced them with ongoing conversations between managers and their people about expectations, feedback and growth. In July 2013 Morris wrote that the company was saving 80,000 hours of managers’ time and that attrition was down year over year. Those figures are Adobe’s own.
Behind it sat a simple bet: a once-a-year verdict often comes after the work is done, so talk while it can still change.
An annual review tries to do two jobs in one meeting: settle pay and help someone get better. In 2016 the management scholars Peter Cappelli and Anna Tavis wrote in the Harvard Business Review that General Electric concluded, after an experiment in 1964, that the two discussions were best kept apart. It is “extraordinarily difficult,” they wrote, to talk openly about problems “while also dishing out consequences such as low merit pay.” The year-end review was also an excuse to save feedback until both people had likely forgotten what happened months earlier.
A satnav that stays silent for the whole drive, then reads out every wrong turn once you’ve parked, wouldn’t get many people home.
Adobe wasn’t alone. Deloitte counted close to 2 million hours a year going into its performance process. In a 2015 HBR article, Deloitte’s Ashley Goodall and the consultant Marcus Buckingham described a redesign built on weekly check-ins between team leaders and each team member.
When more feedback hurts
Talking more often only helps if the talk helps. In 1996 the psychologists Avraham Kluger and Angelo DeNisi pooled 607 effect sizes from studies of feedback. On average feedback improved performance, but more than a third of the feedback interventions made it worse. Their analysis suggested feedback loses power as it pulls attention away from the task and toward the self. A weekly verdict on the person is still a verdict, delivered more often.
Removing ratings is a different move from adding conversations, and the two can come apart. In 2016 the advisory firm CEB published its analysis of companies that had dropped ratings. At most of them, it said, employee performance fell by around 10%. Managers had more time, but they spent 10 fewer hours on informal conversations, not more, and the quality of their conversations fell 14%. By CEB’s estimate, fewer than 5% of managers could manage people well without ratings. The same analysis said that making informal performance conversations more frequent could lift performance by up to 12%. This is a consultancy’s survey analysis, not a peer-reviewed experiment, so it is a warning sign rather than a verdict.
Adobe’s own people saw the risk. Megan Taylor, Adobe’s director of business partnering, said at a conference that continuous feedback is hard to sustain if it isn’t happening on its own, Cappelli and Tavis report. Deloitte started out numberless like Adobe, then switched to giving employees several numbers four times a year. And dropping scores doesn’t drop bias, the HBR authors warn: discrimination and faulty assumptions “still creep into qualitative assessments.”
Much of the good news about continuous conversations comes from the companies that adopted them.
The report nobody reads
Imagine a team lead who discovers in December that a new hire has spent months building a weekly report that nobody opens. The review form has a box for it. Written up now, it arrives in the same meeting as his raise, and he hears it as a judgment on him.
The same fact, raised in a short talk the first month, would have been a question about the work: who is this for, and is anyone using it? The December version tells him what went wrong. The earlier one could have changed what he built next.
That earlier talk usually happens in a one-on-one, and a format like SBI aims to keep it on what happened.
Models drift too
AI systems raise a version of the same problem. A model that passed its tests before launch has had one review, and the world keeps moving after it. The US National Institute of Standards and Technology’s AI Risk Management Framework, published in January 2023, notes that AI systems may be trained on data that “can change over time, sometimes significantly and unexpectedly.” It asks that a system’s functionality and behavior be “monitored when in production,” with plans that include ways to capture input from users, to appeal and override, and to respond to incidents.
The limits echo CEB’s warning. The framework is voluntary, and it leaves the choice of metrics and thresholds to human judgment. Monitoring that nobody acts on is the machine version of a check-in nobody holds.
Years later, with its headcount nearly tripled by its own account, Adobe added a central web tool where employees and managers write down goals and feedback, with conversations every quarter. There is still no formal rating.