Rocket Science for Business: Running Operations based on Controls Theory
I apologize in advance that this post might be slightly technical.
As organizations grow, executives start relying on data from several sources to make decisions. Organizations with a growing gap between desired and actual results can use control system theory to better manage and predict results using their existing data sources. This is not so different from autopilot systems for satellites, rockets and aircraft. I want to draw parallels between the kinds of algorithms I used design in my previous jobs as a flight controls engineer at Sikorsky & Boeing to the data science algorithms that I used at Facebook & PayPal to provide business insights.
An automated system with a dashboard to provide key metrics, alarms and cues can significantly reduce the time spent on data gathering and insight generation, enabling managers to focus on non-trivial long term strategic decisions.
Quick Primer on Controls Theory
Control theory is an interdisciplinary branch of engineering and mathematics that deals with the behavior of dynamical systems with inputs. The external input of a system is called the reference(desired result). When one or more output variables (organization performance) of a system need to follow a certain reference over time, a controller (management) manipulates the inputs to a system to obtain the desired effect on the output of the system. — Source
Control Theory is typically used in two forms;
(1) Goal Seeking: For example autopilot guiding an aircraft from point A to point B
(2) Disturbance Rejection; For example algorithm of an aircraft that maintains a a certain altitude or airspeed.
Relevance to Business
Both of these concepts are based on sound engineering principles that can be used to model and manage business operations. An organization looking to increase profits or grow customer base can be modeled as a Goal Seeking organization while an organization seeking to minimize costs, as a Disturbance Rejection model. Disturbance Rejection acts as a GPS guidance system providing optimal recommendations to executives based on real-time data. The diagram below is a simplistic example of how a control system works.
STEP 1: Hypothesis/Model development
The typical process begins by conducting interviews and collecting secondary research within your organization. The goal is to understand what factors impact your business and desired results. In controls theory terminology this would be called “System Identification” Coming up with a hypothesis here is very important but can be changed later based on new data. This activity helps develop an understanding of the company’s products, processes and people. Additionally, analyze company’s appropriate databases that may include financial data, production logs, requirements repository and customer databases. Based on this initial research you can generate insights into where some opportunities and challenges may lie.
As an example when I was at PayPal I wrote an algorithm that would constantly estimate our share-of-wallet with clients by combining (1) quarterly earnings reports and (2) PayPal’s transactions data.
STEP 2: Design Strategy
Next model the processes into a control loop diagram. This is the process of designing an autopilot for your business where you as the pilot determine what goals should be set for your company. The model also allows you to understand what data is relevant to your business.
When I was at Facebook I was responsible for manage our marketing team’s budget and our strategy was to hit our annual budget goals. Having a data pipelines that estimated where we were headed helped in accurate forecasting and eventually a full utilization of the annual budget.
STEP 3: Enable Data Pipelines
Determine how to effectively collect data on a periodic basis to feed into the model. You will need a team that specializes in automating data synthesis in order keep impact on your management teams to a minimal.
At both PayPal and Facebook this process was very similar. We used SQL+R/Matlab+Tableau/Qlikview to build data pipelines that refreshed daily (if not more often).
Bringing it all together
Every time data is fed into the model it will generate:
(1) Insights: trends that increase executive’s situational awareness
(2) Alarms: Alerts that indicate if any of the KPIs are reaching their thresholds, and
(3) Cues: Suggested actions for executives.
These three pieces of information should result in actions taken by the organization to move in the direction of desired results. Once in place, this results in a cyclic process of experimentation and iteration that can be used to update the model as well.