Every now and then it becomes briefly public that “If You Can’t Measure It, You Can’t Manage It“, is not at all what Peter Drucker or W. Edwards Deming said. Management is not a neatly solved problem of defining OKRs and managing SLIs. On the other hand, it’s also not a mystical exercise in holocracy, where everyone figures out what to do today via divination or something.
Managers are there to make painful decisions and take painful actions that are necessary to make the organization better. People often bristle at this idea. “We are not children, we bring professionalism and discipline and a sense of craft to this work and we will of course do the right thing even when it is painful!” That is fine and true of the team of seasoned professionals operating at the team level and deciding to expand feature area or perform housekeeping. It becomes harder with issues like “we’ve won a set of customers that isn’t profitable”, “we’re building something that isn’t selling”, or “we’ve hired a group of people that aren’t delivering what we need.” Identifying these states might be possible as an individual contributor in a band of professionals, but acting to correct them takes a shared agreement of power and responsibility. Someone has to be the boss in tough times, and ideally that someone will also be able to hold tough times at bay.
It is possible for that manager to work entirely from intuition and their own data sources; we can sometimes forget that business and industry predate high speed communication after all. However, we now have a great deal of data at our fingertips and are eager to use it. Managers look to data for understanding and seek support for decisions therein. This wealth of data is a great comfort to the uncertain management team, but data alone fails to drive good decisions. It’s not hard to find evidence of poor decisions supported by data, but my favorite collection right now is Lea Kissner’s talk on Metric Perversity and Bad Decision-Making.
The biggest issue with purely data-driven approaches is that the data is inherently incomplete. As Chelsea Troy writes, landing a big product change is more than typing code. Pivoting a sales team from unprofitable deals to profitable ones is ultimately measurable in gross margin, but the activities that lead to that success require massive amounts of human judgement. You can put that judgment call into a field in Salesforce so you get a chart, and you can buy human-sourced data for that field, but you can’t make the number be accurate or tell the future purely from machine data. Machines lack context.
As a person with context, we can be tempted by the idea of table flipping the OKRs and SLIs. “We all see what needs to be done, right?” Unfortunately, that way lies management by popularity contests and executive fiat.
It being mid-2024, the other evil possibility rattling around in everyone’s head is to replace management with AI.
- Measure: What did we do, did we do it well.
- Manage: what do we do now.
Perhaps we can insert the chatbot at “what do we do now” and “did we do it well”, where it serves the purpose. Instead of getting the developer out of the way, we could get the manager out of the way! Priests (oops, I mean prompt engineers) then can query (propitiate) the idol (ChatGPT) and interpret its enigmas. Well, maybe that’s just a redefinition of manager, but whatever, still an opportunity to reset an organization and perform some consultation.
The inherent flaws are still there. Situational awareness is difficult enough to produce and maintain for people, and making it possible for computers is more expensive and more difficult than doing it for people. To the degree that attempts to do so succeed, they really drive an economic process of disintermediation and globalization, not automation. In other words, look to the offshoring of in-house development and the remote assists of self-driving cars as a model for the future. Management As A Service.
Monkeynoodle.Org 