I spent years building and selling HR tech. Now I help vendors and enterprises get workforce AI products to work in the real world. Here I write about people and work data as a product, why AI-native companies will need far more of it, and whether that’s a good thing.
The prize is a virtuous circle. Recruiting adds to what the organisation knows about the roles that succeed. Learning adds what people are actually building. Performance adds where work is going well and where it isn't. Planning draws on all of it, and in turn shows where the next hire or course is needed. Each project deepens the information about work, people, performance and contentment, and the next project starts from a richer base.
Few organisations get that effect. Many are running AI experiments in several places at once: pilots within recruiting, knowledge bases for projects, an agent for customer service, and so on. Each of these can deliver real local value. But rarely do the initial platforms or products aim to connect to, or learn from, other systems. So each AI pilot is effectively an island, assessed on its own discrete ROI rather than on its ability to contribute to a deeper level of organisational intelligence and data.
Why findings stay inside each pilot
Most AI in HR pilots are scoped around their own goal and their own data: time to hire, course completions, calibration time. What they learn stays there. The cause is rarely the AI. Systems identify the same person differently, define skills differently and hold data current to different dates. Connecting two pilots means reconciling who is who and what each term means before anything else can happen.
What compounding needs
In my experience, experiments only compound when they share knowledge, context and data sets. The alternative is isolated data sets, or tools that draw mainly on market data and generalised norms about work and skills. Those describe the average organisation, not this one.
What it takes
The base is unglamorous: agreed identifiers, governed taxonomies, data kept current and a record of where it came from. It is the same data, context and governance work that decides whether any single AI product performs, and it is what makes the second pilot better than the first. This isn't an easy sell!
