Data at Work: A Check, Not a Verdict

Learning data at work helps when it tells people about the task and shows managers patterns across a team. It causes harm when it ranks people or watches them. The difference is a design choice: learners see their own progress against their own past, managers see skills and teams rather than names. A person makes every decision that matters.

Feedback is the reason to collect learning data in the first place, and the research on it carries a warning. A review of 607 effects found that feedback improved performance on average, but in more than a third of cases it made performance worse (Kluger and DeNisi, 1996). It worked less well the more it drew attention away from the task and onto the person.

A newer meta-analysis puts detailed feedback about the task at an effect of 0.99, against 0.24 for praise, rewards and punishment (Wisniewski and colleagues, 2020).

Ranking pulls attention onto the person. In a randomised trial inside a national training programme in Zambia, telling trainees their rank lowered their exam scores by about a third of a standard deviation. For the bottom third of the group the drop was twice as large, and it was larger still when the ranking was public (Ashraf and colleagues, 2014). The people a training programme most needs to reach were the ones the league table pushed further back.

A field experiment with a sales force found the same pattern at work: when rank feedback was removed, sales went up (Barankay, 2012).

Monitoring adds stress without adding performance. Two meta-analyses of electronic monitoring at work, one covering more than 23,000 employees, found a small rise in stress, a small drop in job satisfaction and no gain in performance (Ravid and colleagues, 2023; Siegel and colleagues, 2022). Monitoring that was open about what it collected went with better attitudes than monitoring that was invasive.

Learning also needs people to admit what they get wrong. In a study of 51 work teams at a US manufacturer, the teams where people felt safe to own up to mistakes asked for help and feedback more often and talked openly about what had gone wrong (Edmondson, 1999). Data that feeds a verdict on the person teaches people to hide their mistakes, and a hidden mistake cannot be fixed.

In schools, comparing a learner with their own past is a way round all of this. It shows real progress to a learner who will never top the class, and it avoids the distortion of judging everyone against the group they happen to sit in. In FeedForward, Marking and Feedback reads every score against the learner’s own average for this reason.

Put together, the design follows from the evidence. Each learner sees their own record: which skills are improving, which mistakes keep coming back, and how much help they needed this month compared with last. Managers and training teams see the same data rolled up by skill and by team, never by name, with groups large enough that nobody can be singled out. There are no leaderboards. Nothing from the learning record goes into a performance review, and any AI that flags a pattern hands it to a person, who decides what happens next.

This is stricter than the school version, where the educator sees each learner by name. The difference is the relationship. A teacher’s job is to help that learner, and nothing the teacher sees decides the learner’s pay or position. A line manager holds both. So at work the named record stays with the learner, and with a trainer who has no say over their job.

The law is moving the same way. Under the EU AI Act, AI used to monitor or evaluate workers’ performance and behaviour counts as high-risk, with duties that apply from 2 December 2027. Those duties include human oversight, the ability to override or stop the system, and informing workers’ representatives before it is used. The Act already bans AI that infers emotions at work.

In Spain, works councils have had the right since 2021 to be told how any algorithm that affects working conditions makes its decisions.

The design above goes further than the law requires, because the law sets a minimum and the research shows where the minimum falls short.

FeedForward is built and running for schools and academies. Its workplace version will be built to this design.