Systems Thinking In People Analytics – Or: Exploring The Art Of Ripening An Avocado

By Peter Romero and Andreas Kyprianou

You’re in the mood for avocado on toast – who isn’t nowadays right? No problem, you went shopping two days ago – surely those ‘ripen at home’ avocados are ready by now, since you left them in a nice dark cupboard. You reach for it full of excitement and anticipation, only to find a rock hard fruit unsuitable for human consumption. ‘Beans on toast it is again’ you sigh. If only you were familiar with principles of systems thinking.?

As any avocado lover worth their salt will tell you, the only right way to ripen an avocado is to put it in a paper bag together with apples, bananas or kiwis. The reason- fruits and vegetables need a gas called ethylene in order to ripen faster and the three fruits mentioned above are the biggest producers of this. The paper bag in turn traps the ethylene gas and speeds up the ripening process.?

What does any of these have to do with People Analytics? The avocado lover’s fallacy is a classic illustration of systems thinking – a tool-set and philosophy based on the fact that the world is made of numerous complex relationships that interact with each other in a variety of ways. Some of them are rather unexpected, if not outright counterintuitive. Linear thinking, which proposes that one cause has only one effect,?is the opposite of systems thinking. For some problems, linear thinking is more than enough. However, organisations are complex, integrated systems, and therefore for most problems that People Analytics functions face, linear thinking is ineffective, if not downright false.

In this article we’re going to explore the systems thinking philosophy and provide the tools that you can adopt as a systems thinker when it comes to some of the most common HR problems we come across.?

The Systems Thinking Philosophy

The most basic concept of systems thinking is that a system is a collection of interacting elements. Whether you are aware of it or not, you are an element of many systems yourself, as well. For example, you are a part of your family, your organisation, your society, or our planet. You yourself, are a complex biological system, composed of many smaller subsystems like organs and cells. And, the elements of a system interact. In the simplest case, the interaction between two things in a system is directional – one thing has influence on another thing.?

Since all parts of a system are interrelated, changes that are seemingly narrow in scope, can set off a domino effect that reaches far wider than ever anticipated. A decision you make today could have unintended consequences in some other part of the system now, or at some point in the future. To fully understand the consequences of all possible actions, it is therefore important to understand these influences and general behaviours of systems.?

This is where systems thinking diverges from linear thinking. The latter tends to focus on visible events only and react to these events. For example, a key employee hands-in her resignation letter. The leadership panics, and makes her a counter-offer in a futile attempt to keep her (which one should never accept, as a side-note). Most companies are stuck on that level of linear thinking, that only deals with symptoms, yet ignores both trends and deeper potential root causes. Systems thinking on the other hand, encourages as a first step to look for patterns and events below the surface of the apparently obvious. Maybe, multiple key employees of various departments have resigned in the recent six months, of which the majority were female? This could be indicative of an underlying cultural root cause: a fault line between the genders in the organisation, maybe based on unfair promotion or compensation practices.?

Systems Thinking in People Analytics

With the advent of People Analytics, the maturity curve became popular and omnipresent. Not only does it represent the direction in which People Analytics leads organisations in the digital agenda, but it also represents a deep human desire to know about and predict the future.?

Figure 1- The people analytics curve

The main selling point of this curve has always been to move organisations from the relatively mundane descriptive and diagnostic analytics to the exciting world of predictive and prescriptive analytics, decorated with buzzwords like AI and Machine Learning (here is a pro-tip: the next time some business exec uses those, ask for the underlying math, and get some popcorn).?

System thinking teaches us that this curve is practically wrong, since it over-simplifies reality, and, worse, leads to wrong conclusions. There are two reasons for this:?

• First - in many cases, the ‘why’ is much more important for the organisation than the ‘what’. Think about the most basic example we discussed earlier - what is more important, understanding who will leave the business, or understanding why people leave your business and fixing the underlying symptoms?
• Second - predictive analytics is powerful, because predictions are based on past evidence without the burden of having to understand why things happen. Yet herein lie also two big risks. First, all predictive algorithms are based on historical data, and therefore carry the bias of past events, as well. A famous example is Amazon’s gender-biased recruitment tool that they had to scrap, since it was trained on data that contained almost no female examples. The result was that everything that indicated that the candidate could be female, led to an instant exclusion. Second, predictive analytics has problems with sudden and unexpected events. They are based on an underlying assumption that the future will not radically diverge from the past. But exogenous shocks happen, for example the sudden emergence of COVID-19, which influenced the entire world economic system; the post-pandemic era will be radically different from the past and the present both economically and socially. And with that, future behaviors will be different than past behaviors - think about social distancing or checking into locations. Evidence from the past will not reliably predict the future. Therefore, it is time to ask how reliable, how trustworthy, and how risk-burdened future predictive models will be.

This is exactly where systems thinking comes in. It offers a wide array of diagnostic tools for better understanding the reasons for why things happen, gauging the implications of responding or not, and deciding about optimal responses. In essence, it helps us to invert the maturity curve, understand root causes, and from there, create much more powerful predictive models. Those, in turn, will inform much more adequately and precisely the prescriptive actions you need to take.?

Figure 2- The ‘inverted’ curve

Systems Thinking Tools?

There is an array of tools in the systems thinking toolbox, from dynamic and structural thinking like causal loops and structure-behaviour, to computer based tools like management flight simulators, learning laboratories, and psychometric tests. We will explore most of these in our series of articles. However, for this introductory article, we will start with two tools that form the foundation of systems thinking: causal loops and system archetypes.?

Tool 1: Causal Loops

Causal loops are essentially a visual representation of systems behaviour. It is a form of cause and effect modelling that allows you to visualise and model all parts of a relevant system, illustrating the interactions and influences we described before. The easiest way to understand causal loops is to explain them on a basic workforce example like employee productivity.??

Figure 3 – Reinforcing loop

The oval shapes in figure 3 are called ‘nodes’ and represent system statuses of various kinds, for example managerial tools like performance bonuses or resulting outcomes like employee productivity.Nodes are linked to each other by connectors that indicate their relationship, either via a ‘+’, or via a ‘-’. A ‘+ connector’, is sometimes also called ‘same direction connector’, since it indicates that the values of two nodes move in the same direction when a change occurs. For example, figure 3 symbolises that when performance bonuses increase, employee productivity will increase, as well. When performance bonuses decrease, employee productivity will decrease, as well. On the other hand, when employee productivity increases, performance bonuses will increase. And, when employee productivity decreases, performance bonuses will decrease, as well. This closed connection is called a loop.

Each loop can be reinforcing or balancing, which is marked by an ‘R’ or a ‘B’ – in a reinforcing loop such as the one in figure 3, change in one direction is compounded by more change in the same direction. In the balancing loop in figure 4 on the other hand, counters change in one direction with change in the opposite direction: as workload increases, so does the capacity gap. This, in turn, increases hiring and workforce capacity, which then regulates down the capacity gap.??

Figure 4 – Balancing loop

The beauty of causal loops is that once you start making the links between the loops, you will start discovering side-effects, influential forces, and connections not visible from the start. For example, the hidden link between productivity and workforce capacity, which is uncovered in figure 5. As employee productivity increases, workforce capacity increases, as well, which results in a smaller capacity gap, which, in turn, leads to less hiring.?

Figure 5 –?Uncovering Hidden Links

As a People Analytics leader, your message to the CEO would have been as simple as compelling in this case: pay less people better. Of course, the situation is never that simple, but we hope this example has illustrated the benefits of causal loops for People Analytics practice, especially during iterations of modelling and reporting.

Tool 2: System Archetypes

Once you have understood causal loops, creating them is straightforward. However, finding meaning in these loops is much more challenging. This is where system archetypes come in. These are eight ‘archetypal’ system structures and behavioural patterns that keep on recurring across a variety of settings and organisations. Of these, we will introduce two today: ‘Fixes That Fail’ and ‘Success To The Successful’. As all system archetypes, these are amazing starting points for People Analytics practitioners to drill down to the root of issues.?

Fixes That Fail

‘Fixes That Fail’ is one of the most common occurring archetypes and refers to the situation, where a fix to a problem has an unintended consequence, which is not immediately obvious, but eventually exacerbates the original problem. A diversity example is depicted in figure 6, based on a staggered causal loop.?

Figure 6 –?Systems Archetype 1: ‘Fixes That Fail’

It starts with the problem symptom of poor gender diversity. The immediate and short term fix that most People Analytics practitioners face, are instructions from leadership to simply hire more women. While this may reduce the problem symptom, it will lead to much worse problems. If hiring is not conducted carefully, the original situation could worsen - women, who are hired in a rush, could be unsuitable for their roles, become disillusioned and move on. In return, this could lead to cultural bias against female hires in parts of the organisation. We saw this pattern especially in organisations with strong hierarchies, and in countries with rather conservative cultures.

A first remedy to break this archetypal cycle is to acknowledge it exists, and to commit solving the underlying issues. In case of gender diversity it does not mean that keep on hiring more women is wrong - quite the opposite is true - however, the quality of recruitment and the communication of recruitment difficulty needs to be increased, while solutions for the real root cause of the problem are explored. Beyond hiring, this could be to sponsor initiatives that encourage women to enter certain careers, for example in STEM. It could also mean to ask the People Analytics leader to partner-up with leadership and HR in re-defining corporate culture. This is where holistic data from your People Analytics function comes in and merges with qualitative discussions, for example exploring attrition patterns, mobility or promotions data, as well as development opportunities. In short, this enables a real detailed discussion of hiring practices at the senior strategy table.

Success To The Successful

Let’s assume, people data shows a mid-career dent of female leadership careers, whereas the male leadership careers begin to flourish exactly at that point. This is where the next systems archetype might be useful for analysing the situation and finding solutions to it - ‘Success To The Successful’. It describes a situation, when an individual or group receives more resources and starts becoming more successful through that, which perpetuates a cycle of success that cannot be caught-up with by others. Such situations are very similar to how the most successful soccer teams attract the best players because of their initial success, which subsequently leads to even more glory, but we digress. Figure 7 uses chained causal loops as an example of a leadership development program gone wrong.?

Figure 7 –?Systems Archetype 2:?‘Success To The Successful’

Let us assume John and Tracey started their careers at the same company, displayed the same potential, had the same successes, and it took them the same time to progress to the same level. However, while Tracey was on maternity leave, John got the opportunity to attend a Future Leaders’ development program, where he convinced several senior executives of his qualities. After Tracey’s return, both apply for the same promotion and John is chosen on the back of his participation in the program. Five years down the line, Tracey is considered to be a steady performer, while John is celebrated as the next superstar.?

As before, the first step to break this archetypal cycle is to acknowledge it and to analyze the root causes more deeply. For example, People Analytics practitioners should explore in this case the measurement system behind performance evaluation. Are all factors taken into account, for example maternity leaves? Is there any bias in the decision making process, for example through nepotism of the senior executives who got to know John and have a couple of drinks with him? Is the element of chaos and luck completely discounted, or does the decision making allow for that? We experienced these examples in a frightening range of organizations and cultures, no matter how progressed these were. Some had very obvious and flawed, not to say unfair, High Potential programs, while others hid their bias well with progressive equal opportunity measures like sponsoring egg freezing, yet ultimately displaying the same bias. We believe that diversity and inclusion is a prime topic for People Analytics to put a focus on, especially when it is practiced on the paper.

Applying Systems Thinking For Predictive Models

Finally, the process of systems thinking is very important for creation of predictive models in the People Analytics practice. Think about figures 1 and 2 again. Once we understand the root causes, we have to work our way back up to create potential future scenarios based on the actions prescribed through the systems thinking analysis. This is maybe the starkest contrast to the classical, and frankly technically outdated approach in organizational behavior and management science. Their approach is to look at isolated issues, come up with a couple of predictors, assume these were independent and identically distributed, and then find an artificial experimental setting, where a series of conditions is kept equal, and just one is changed. While good on paper, this approach often fails in practice, since it oversimplifies the problem. The solution is to not artificially constrain the number of parameters. Even a decade old machine learning approach - which is considered technological stone age in AI research - would do better in most cases.?

We recommend creating as many hypotheses as possible based on conversation in the field and theories from literature, formulate them into synthetic variables, and then inject them as additional parameters to machine learning models. The resulting predictions will show the most likely root causes to problems, which can conveniently be clustered into actionable versus non-actionable, and thus represent new hypotheses. These should be discussed with local stakeholders, and then be used as additional parameters for new models. You should consider this as a trust-building, communication task, rather than creating an opaque model using esoteric language and then going back to the business and telling them what to do.?

Figure 8 –?Systems Thinking in Machine Learning

This is especially important, when one has a plethora of data points of varying granularity, for example daily measures, yet wants to predict success on an annual basis. During the process of communication, one may instead re-formulate the outcome in a more behavioural sense, using the competency model of the organisation. This breakdown further helps you to explore the mental model behind the data generating process, and thereby identifying the minimal cultural units it results in, for example, some organisations identify activity as performance, which leads to mindless actions, since these are considered to be beneficial for success. Once identified, one can then re-formulate success, and integrate discussion time into it, as well. Here is a hint for the geeks amongst you: in a way, these causal loops divert from classical research models that are composed of causal assumptions, moderation-, and mediation-effects, by preferring local optimisations for specific contexts, thus breaking down one corporate culture in many localised sub-cultures with a common bias. But that’s material for later articles.

Becoming A System Thinker

We hope that the above tools and philosophy are helpful for you in your own journey on becoming a system thinker.?

1. Try and think holistically instead of breaking everything down into components.?

Thinking holistically can be a problem in HR, where we view things quite linearly. Most employee journeys are mapped out as a straight path from hiring to leaving, and many of the processes are quite linear, for example annual or bi-annual performance reviews, regular engagement surveys, tenure based adaptations of seniority levels. However, with the advent of People Analytics, the ability to have a better view across all departments demands a different approach to thinking, in systems and in interactions rather than in top-down processes. Foster that thinking in every aspect of your People Analytics journey.

2. Step away from data.?

Don’t start with data but with the questions. You need to be fully dedicated to finding out what the data generating process behind things is; why things happen. Start exploring those questions with colleagues, peers, and domain subject experts using open discussions to uncover shared knowledge, beliefs, and experiences. Develop hypotheses about why things happen and represent those hypotheses as potential causal models. Use these for creating preliminary predictive models by turning hypotheses into synthetic data, which is fed into machine learning models that predict outcomes of interest. Then, use these preliminary models to have further discussions, create new causal models and further preliminary predictive models to?verify, validate, quantify, and refine the assertions of your causal model. Lastly, see everything from the perspective of your client. It’s known that Jeff Bezos has one empty chair in every meeting that represents the voice of the customer. Put the benefit of your internal customers brutally in the centre of all your decisions, even if that means that you have to disappoint stakeholders or start from scratch. Prefer those causal models that are most actionable, demand the least effort, yield the most results, and are - aligned with Occam’s razor - most simple.

3. Finally- trust in the chaos.?

Not everything can be explained and not everything needs to be either. Trust your instincts, don’t try to control everything, and chime in the collective flow of conversation with stakeholders you initiated. Your ultimate role as head of People Analytics must be to facilitate, not to enforce.?

Now go out and enjoy a couple of tasty and well-deserved avocado toasts.