AI tools can only leverage what they can see. For far too long, business has poorly measured and understood human behaviour in the workplace. We are likely to need to better understand the human activity, motivations and drivers to get sustainable value from AI.
In a recent post on the Stratechery blog by Ben Thompson on Open AI’s new Deepseek research capability, he concludes with a discussion of where the tool failed badly, an industry analysis with one large private party. Deepseek couldn’t see that business’ information so produced a misleading industry structure without it. For a reader without knowledge, there’s no clue something is missing. It is a Rumsfeld ‘unknown unknown’ as Ben Thompson notes. He goes on to make a powerful point about new value in secrecy in strategy in a world of AI research.
This blindness raised for me another critical question. What else is often invisible in organisations and might mislead business decisions made by AI? The most obvious answer is data on people. For too long, understanding people, performance, motivations and decision making in organisations has been at best random and sporadic. Avoiding the danger of unknown unknowns is going to demand more than faith in AI, it is going to take a new model of how we understand people and surface performance. Productivity measures alone will leave too many blackspots.
When you consider the HR data that most organizations have it is the information collected on joining, changes in roles or benefits that force updates, list of titles, performance grades, some engagement survey questions and productivity data. Much of this is not a system of human data, it is the output of other systems.
When supermarkets like Tesco started to look at leveraging data for loyalty, marketing and digital engagement of consumers, they found their data on products was all driven by their merchandising systems. It was detailed SKU supply chain and logistics data required to procure and stock items. Unfortunately engaging customers more effectively required different data like whether items were low fat, low sugar, feel good, healthy, suitable for lifestyle choices, treats, ingredients or meals. All the inputs to human food decision making. We don’t decide what we want eat in a logistics process. They had reams of product data and were blind to what consumers wanted to know and the key elements of consumer decisions.
Organisations are going to have to bring this same analytics into their organisations because when you consider it most organisations don’t have the data today. Some examples:
- who works on what office on what day of the week?
- the informal networks of exchange of information resources and power between roles
- capabilities, skills and talents from previous roles and organisations.
- individual motivations and goals
- non-work challenges and opportunities that may be driving employee behaviour
Until organisations start to leverage different richer kinds of HR data, there is a real danger the human elements of performance are overlooked entirely. That will have devastating effect on people, the value of AI and performance.
