The last 3 weeks have been hectic here at OrganizationView. We’ve had the perfect storm of children’s school holidays (3 weeks long here in the mountains), massive amounts of fascinating client work, hiring, illness and an office move. I hope you understand why getting this newsletter out missed a few weeks.
After 11 years of WFH I’m back in an office, albeit one with mostly just me. I’ve lost the majestic, panoramic views of the snow-covered mountains and the lake but have acquired what must be one of the few ski-in/ski-out offices in the HR industry. I’ve been able to hang the Vitsoe
shelves along the longest wall, fill them with my rather large number of geeky books (“Data Analysis Using Regression and Multilevel/Hierachical Models
” anyone?), a subset of my collection of obscure Scottish text-based art
and spread screens and workstations around. I think my children fear I won’t come home much.
In People Analytics I’m noticing we’re at a junction point. Overall I’m seeing the type of statistics and modelling many people are doing are becoming simpler, probably more appropriately chosen. I think this is a postive. At the same time I spend my time working with more and more sophisticated ML models, mostly deep-learning-based, which require us to think of the logistical issues of scaling and optimising compute. We’re about to hit 100 million rows of feedback data, each with associated metadata, in our systems - tiny in comparison to the 499 billion tokens that GPT-3 was trained on but still enough to cause challenges.
Where does this leave us? As PA practitioners in organizations I think we have to continue to focus on small-data, probably with an assumption of messy data. However we should build a strategy where some tasks which do require big data need to be bought not built.
It used to be the case that many ‘AI products’ being sold to HR teams could probably be built by a smart internal team in a relatively short time. I think those days are disappearing. As a PA team the need to have a strong ‘when to buy, when to build’ strategy is becoming critical.