I was reminded today of an article I once read on places to intervene in a system by the late great Donella Meadows. I panicked briefly because I wasn’t sure if I could find it, but a little bit of lucky googling got me there. This quick post summarizes how I think that article relates to the kind of work I do.

In data science or analytics we’re often charged with building models to help calculate answers to questions like

  • How much should I invest in business W?
  • How many people should we hire for X?
  • Should we build feature Y?
  • How can we encourage users to take action Z?

I have always found this mandate to be troubling because I find the returns calculated from incrementally better answers to these questions rarely justify the high cost of infrastructure and personel, especially in startups.

In Meadows’ article she outlines how complex systems are more responsive to some kinds of interventions than other kinds. She comes up with a list of 12 ways to intervene in reverse order of effectiveness:

  1. Constants, parameters, numbers (such as subsidies, taxes, standards).
  2. The sizes of buffers and other stabilizing stocks, relative to their flows.
  3. The structure of material stocks and flows (such as transport networks, population age structures).
  4. The lengths of delays, relative to the rate of system change.
  5. The strength of negative feedback loops, relative to the impacts they are trying to correct against.
  6. The gain around driving positive feedback loops.
  7. The structure of information flows (who does and does not have access to information).
  8. The rules of the system (such as incentives, punishments, constraints).
  9. The power to add, change, evolve, or self-organize system structure.
  10. The goals of the system.
  11. The mindset or paradigm out of which the system — its goals, structure, rules, delays, parameters — arises.
  12. The power to transcend paradigms.

Meadows defines leverage points as “places within a complex system (a corporation, an economy, a living body, a city, an ecosystem) where a small shift in one thing can produce big changes in everything.” Producing such a change in a desired direction is the objective of almost any kind of work, but I would argue especially in data science.

The first two questions that I gave as examples correspond to the least influential ways to change a system: Constants, parameters, and numbers. The third corresponds to #2, sizes of buffers. The fourth is nebulous, but might be thought of relating to #3, the structure of materials.

I contend that the data team’s highest leverage work is the ability to change the structure of information flows (#7) by democratizing access to data. By improving the quality of information at Artsy and the ability to access it, we generated dramatic change across the organization in how people did their jobs. Hundreds of small decisions each day were made using up to date and reliable data, and the compounding gains from that volume of impact definitively justified the investment in a thoughtful analytics team.

The structure of information flows is the highest leverage point in a small to medium sized business that doesn’t belong explicitly to the realm of the executive team (although of course some of this is fractal in the sense that incentives and mindsets can be set at multiple levels). The data science or analytics team is uniquely capable of influencing this leverage point, and that is the most important reason for having one.