Management is continually taking control action. This includes upper- and under-level managers who seek to achieve business objectives by adjustment or influence of human resources, financial resources, technological resources, and natural resources. It also includes managing your personal life, volunteer activities, and wherever you make decisions to optimize desired outcomes. Although managers have many important and useful experiences, often they do not have an automatic control background. As control engineers we can help managers design their control strategies. Doing so may be an opportunity for you to understand their issues and to add value to your organization.

For control, we know that you must first understand the process attributes. Also, you need to understand feedback control techniques appropriate to those attributes. Then you will be able to select and set up a control strategy. As control engineers we can share this viewpoint with managers. By doing so, you can add value to your enterprise; and be recognized for contributing. Take care of how to present your solution when you seek to tell your superiors how they should do their job. You don’t want your meddling in their territory or your attempt to override their autonomy to lead them to say, “Stay in your lane.”

The human response to criticism or unasked-for help is also a process.  Some humans are less willing to accept help than others!

Before considering how to apply control to managerial decision making, let’s review a few basics about automatic control.

Basic 1: Identify the cause-and-effect mechanisms

The first rule of control is to understand your process. If you want to control something, there must be a cause-and-effect mechanistic relation between the desired process response (controlled variable, CV) and the action taken to cause the CV to change (controller output or manipulated variable, MV). If there is no mechanistic connection, then changing the MV will not have any effect on the CV. There are probably several possible MVs for an application. For instance, a passenger in the back seat of a car can steer the car by shifting weight from one side of the seat to the other, but the effect is small. For control, choose an MV with a strong impact on the CV, that is least expensive to use, and that has little upset on other process desirables.  Identify the mechanistic way that disturbances and other influences would cause the CV to deviate from a desired value.

Maybe, the best control strategy is to prevent those disturbances from upsetting the CV, rather than taking compensatory feedback action.

In my experience technical folklore, mythological stories, understanding grounded in long past no longer relevant operations, superficial misunderstanding of other’s reports that remain in the minds of decision makers, and such erroneous knowledge could be perceived as mechanistic understanding. This leads decision-making astray. The decision makers who could be using technical folklore could be either managers, operators, or novice engineers.

Basic 2: Be sure the measurement is valid

Many things may be easy to measure, but that doesn’t mean that they are true or valid indicators of the desired CV objective. For instance, friendliness toward managers could be a measure of morale, but it might only be self-serving obsequiousness. Be sure that the data you are using as a measure of the desired result is both a valid and a comprehensive measure.  There may be a tendency to select metrics of performance that fit one person’s bias or their intent to glorify themselves. If the CV is not a valid metric for the objective, the control action will not work.  It could even steer the process in the wrong direction. If the CV is not comprehensive, if it only relates to a fraction of the desired comprehensive outcome, then the myopic control will steer the process off target.

Basic 3: Feedback control is inventory management

Feedback control is inventory management. For example, to control level in a tank, adjust the liquid inventory, not the measurement. The control action does not add or subtract level, it adds or subtracts inventory, and level is the process response. Similarly, to control temperature in a house, adjust the heat inventory, not degrees Fahrenheit.

If you want to adjust some CV, first identify the inventory that causes the process measurement to change. You don’t increase or decrease the CV; control adds or removes the inventory that it quantifies.

Maybe the inventory is capricious or ever-changing direction that frustrates workers.  Moral-boosting meetings do not reduce the inventory. 

Basic 4: Control-related categorization of process dynamics

In a stable SISO process, the rate of inventory leaving is proportional to the inventory. Consider house temperature. The hotter the house, the faster the heat diffuses to the environment.  Temperature does not leave or enter, heat does, but temperature is the state variable, the measure of inventory. In a human process this might relate to training or messaging that fills the mental inventory of the desired way to do things. But over time, expediencies, shortcuts, forgetfulness, outside influences, etc., could cause the inventory to leak out or lack potency, altering the desired behavior. Control teaches us to keep adding inventory at the rate it leaks out.

Consider some managerial examples:

The measure of safety could be the number of reportable incidents. You don’t add or remove incidents, you add to the inventory of employee concern/knowledge/awareness about safe practices. By training, posters, performance feedback, displays, etc. you continually pump in concern/knowledge/awareness to influence the employees. The inventory of concern trickles out as new employees replace old ones, and as the desire for convenience or employee appraisal based on productivity progressively overrides their attention to safety. The inventory of concern for safety is an open-loop stable process.

Employee training for many other desired behaviors could be similarly mapped to an open loop stable process. Consider desired customer relations, workplace harassment, eliminating bias in hiring and promotion, employee morale, report writing style, resource waste, etc. I think that the level of employee focus on these issues can be enhanced by training, but without adding to the inventory, it will relax to some nominal hemostasias level.

Get your subscription to Control's tri-weekly newsletter.

Customer brand allegiance and local community relations could also be considered open loop stable, which can be manipulated by advertising and corporate support of local benevolent organizations. If the public image score drops, add support to the local Y, or library, or pet adoption program, or high school football program; and make sure the support is publicized.

In an integrating process the control volume does not lose inventory. Consider filling a bucket.  Level is the state variable, which is changed by adding or removing liquid. In a human process, this might relate to an inventory of perceived injustices or affronts. Grudge builds up until it overflows. Control teaches us to manage the inventory accumulation. Also, happiness and satisfaction are contagious.

In an unstable SISO process, the rate of inventory accumulation is proportional to the inventory.  Consider placing two rabbits on an island. The two-rabbit inventory self-generates until it is limited by fuel supply. In a human process, anger generates anger and compounds. Control teaches us to quench the growth before it explodes. 

Delay in a process (deadtime) is caused by either of three things. One is transport delay, such as fluid flow from here to there, or carrying a sample from the process to the lab, or writing a report and having it validated and approved at each of several stages upward before becoming visible to the decision maker, or the time after a decree to have it sequentially disseminated and forms and procedures revised before it is implemented in operations. A second is laboratory analysis, such as waiting 15 minutes to get the results of a COVID-19 test, or 30 minutes for an on-line analyzer to complete the cycle, or the time it takes for a SPC assignable cause to be detected then analyzed before action. The third is an apparent delay due to a series of lags, such as tanks in series, or when a decision must be sequentially adjusted for details.  In a human process there may be several such sources of delay. Control teaches us to push cautiously and wait to see the effect before additionally adjusting the MV. Better yet, redesign the process to reduce the delay.

Inverse action means that the process first responds in the wrong direction, then changes and moves the other way.  As an example, add cold water to a boiling pot. Adding inventory should increase level.  But the first action is to stop the boiling, collapse the bubbles which lower level.  Then as heat is added and boiling resumes, the bubbles cause the level to rise above the before-addition level. In a human process, initial push-back against an imposition may come before compliance and the desired outcome. Control teaches us to push cautiously and wait to see the effect before adjusting the MV.

Multivariable processes usually have several desirable response variables and several inputs that can be manipulated. Some response variables are primary, meaning of direct importance to the bottom line or purpose for the process. Others are secondary, they do not affect the corporate performance but are essential to keep the process operating. For instance, a distillation column has two primary variables, top and bottom composition, and two associated MVs boil-up and reflux. However, pressure control of the column, and level control in both the reboiler and condensate receiver are essential to maintain operability. In a car, speed may be the primary controlled variable related to its purpose, but engine temperature, oil pressure, and gasoline level are all essential secondary variables. Whether primary or secondary to the process role, it is essential to control all these variables. Control also teaches us that if independent controllers are observing one variable and making changes to fix their myopic view, they upset other loops. There needs to be coordinated compensating action to prevent one controller from upsetting the others, or one controller needs to be tuned to be very gentle so that the upsets are minimal. 

In managerial processes, it is very easy for one manager to push too hard to make the human process fulfill his/her job function or ambition, and undermine others’ objectives, causing others to overreact and return the upset. Control teaches us that the interaction between all the primary and secondary variables needs to be considered.

Constraints may be of many types. They may be on the CV, the MV, on auxiliary variables, or on rates of change. They may be imposed by safety, contracts, product specifications, regulations, laws, corporate traditions, managerial comfort, public perception, or human biases. A constraint violation might immediately happen, or the process may evolve over time to violate a constraint in the future because of today’s MV action or disturbances. In a management process this could be the consequence of letting debt compound and block cash flow in the future. Control teaches us to forecast an approach to constraints and to act now that prevents a future constraint violation, or to use override action on the primary controller when its ability is overridden because of a secondary variable.

Noise confounds seeing the true value of the process variable. Don’t react to noise. Doing so is just tampering, increases cost and adds variability.  For example, there are normal annual fluctuations in university enrollment. The first thing that a new University President observed was a drop in enrollment, obviously caused by the incompetence of the prior President. To be a hero, the new President initiated many changes and programs to increase enrollment. They siphoned funding sources from other projects, weakening then. But enrollment increased. The new initiatives remain. So does the normal variation.