This will be a bit of a sidetrack, but I find this interesting. I work for a Danish muniplacity, and we do BI on a lot of things, one of them is various degrees of employee performance, but because we're the public sector, we also do a great deal of research on how these benchmarks are viewed by employees and managers. We have extremely detailed tools for viewing week-to-week performance, progression, regression and changes over the year and we have an entire team dedicated to presenting this data in a way that managers can understand and use it to help their employees do better.

Only our internal research into the effects of this, shows that things haven't really changed that much. If you enter a performance review on a good story about a recent success, that will mean a lot more than your statistical data. This is true even if a success story is only viewed as a success inside your own team, but not necessarily by the organisation. So we're spending all these resources on building BI, that makes it easier for managers to engage their employees, but it's still the old traditional ways that count, and that means recency bias as you call it, is still really important, even when it shouldn't be.