Last time we analyzed the macro side of complexity, considering its implications in terms of interdependencies with other businesses. As one might expect, these complex features do not exist exclusively outside the walls of a corporation. Commonalities abound, and in many ways the adaptation strategies for macro complexity scale down well into the micro. There are however some defining features unique to the local level which demand specific approaches in order to manage the risk associated with these complex structures.
From Many, One
Chief among the aforementioned similarities (and likely the first that comes to mind for many) is the nature of a business as a unicity of disparate elements. Indeed, the subject of organizations engaged in negotiation with other entities led to the development of the two-level game model, in which negotiation not only occurs at the top level between parties but also between stakeholders within a group. In this way a number of strategic approaches translate across domains, with a key differentiating factor being that within a party negotiation must conclude with ratifying a decision. This is because, for a party to function in the top level negotiation, a group must arrive at concrete proposals with which to engage the counterparty. Negotiation breakdown cannot occur at the small scale, as that would entail decision paralysis and ultimate corporate death. And while parties may intend to present a self-consistent image (often in the form of a single executive), often in action the various contributing voices mesh into anything but. This is far from the only problem of cooperative parties; as we shall see, inconsistency through cognitive biases aggregate into decision noise which further compounds the problems complex entities face.
Divide and Conquer
It is essential to establish why bias from stakeholders can be a problem for negotiation. From a structural standpoint, the members of a complex network of stakeholders are intentionally biased, as each serves to act as a voice for a particular sub-goal the business is trying to achieve. For example, we differentiate between CEO, CIO, CFO, etc with the intention that each individually has a strong foundation in a particular axis of decision-making. This is in large part an adaptation to complexity, as the level of sophistication with which most businesses operate presents information of such density and on so many topics it’s unreasonable for a single individual to rationalize from all of them. But increasing the number of actors in adaptation to complexity can also raise the noise floor for their aggregate decision-making process, resulting in inconsistency and increasing the margin of error.
First a little background on error. In Noise, Kahneman & co lay out two fundamental principles that contribute to poor judgement. Bias, the more commonly known, indicates a tendency to misjudge by a specific degree in a particular direction. It indicates a consistent rationality error; a tendency to overvalue some information types and/or to undervalue others. Noise is a measure of all other error types which are fundamentally random in nature. Occasional forgetfulness and mood effects are two examples of noise sources, as their origins are a kind of “static” in the environment or individual which causes variation in how judgements are made. As we have already observed, stakeholders operate on a bias principle. The intent of a business is to sum the various bias sources together, such that the resulting decision made should eliminate those biases. In fact this is exactly what is observed; if equal weight is given to all sources of bias, then net bias disappears. So what is the problem? While potentially eliminating individual bias from the net process, we fail to measure the impact combining these biases together has on the other half of the error problem: noise.
Out of the Frying Pan
To demonstrate the impact of individual bias on net noise, we begin with a normal distribution of error. The following is a judgement curve, with the area beneath the curve containing all (equally likely) decisions in a single unique judgement.
If we take the center pole as the (theoretical) set of “best possible” / most balanced decisions to make given all available data, then in this case we observe a bias towards a correct decision, with the distribution curve indicating the noise floor. If a random decision point within the curve is selected, the bias leads to a higher probability of decisions made near the “best possible”, with a degree of variance due to judgement noise. However, if that line were further to the left or right, the curve’s bias would cause a tendency away from the ideal. And since we can’t observe the best possible decision directly (otherwise we’d always choose it), we strive to avoid biasing in organizations to make rational decisions.
Now, let’s consider what occurs when a decision is made by a collective, with each stakeholder in the group contributing their particular bias towards a sub-goal for the group that they specialize in analyzing. A finance manager might prioritize fiscal stability, whereas a technology officer may emphasize the integration of new tools. And as each individual only has expertise in a given area, their recommendations while correct on one axis may fail to properly account for other input variables. Such a case of stakeholder bias might look like this:
Each point on the chart represents a decision recommendation from one of these contributors. Visually we start to see the problem with individual bias; there is no guarantee that summing these judgements will make the net decision spread more or less accurate than it would be otherwise. Particular judgement recommendations which lie outside the curve pull it wider and taller; judgements within it pull it tighter. So while in theory biased tendency when aggregated can reduce net bias impact on the sum decision, they can also increase judgement noise by expanding the space under the decision curve. And in practice correctly balancing the importance of each stakeholder’s values against each other to remove bias requires knowledge of where that “ideal decision line” is, but that is a fundamentally unknowable factor. Trying to extract the noise from judgements by creating field specialists, then attempting to remove their individual bias by summing their contributions can often serve to recreate the same effects we were trying to avoid in the first place: human decision error.
Not Done Yet
Of course, bias summation isn’t the only thing that occurs when parties build complex stakeholder networks. We have to consider the potential for increased specialization to create error in some areas while eliminating it in others. As we know, framing effects cause different, specific types of rationality errors. Hyper-specialization is a perfect environment for a focus on instrumental goals without the proper context necessary to give them a correct weighting versus other sub-goals when justifying a decision of this or that. While the aggregation process is intended to eliminate these bias sources, the fact that it is unclear to any of the contributors which goals should be prioritized in a given situation means that sometimes the loudest voice wins. Hardly an unbiased result.
There are countless other issues with increasing the stakeholder count. Personality effects can cause conflict which brings emotion into what ostensibly ought to be a purely rational process. Individuals who are considering moving companies may over-value short term gains versus long-term corporate development. When specialists find they have to choose between their own focus area and another’s (e.g. financial stability versus risky growth potential), people have to compromise. But this causes another problem: antagonistic relationships necessarily generate a discount factor such that all parties’ respective utility functions are eventually left with some value “still on the table”. And outside these abstractions, in the “real world” a compromise between two good decisions can easily be itself a much worse decision than either of the two alone due to interaction effects. On and on the problems compound, with no end in sight. Make an executive decision and trust your own noise/bias balance? Let the group of experts decide, potentially introducing even more noise through their own individually increased bias?
Paradise {Not} Lost
The key to unraveling this puzzle lies in a pure rational agent: AI. An engine can’t tell you what you ought to value, but it can be built with neither bias nor noise and therefore indicate the best possible course of action once you’ve decided what the goals are. In the context of stakeholder groups this means a system which can merge the expertise of a diverse collective (all of whom make decisions within their domains better than anyone else, even algorithms) and combine their respective analysis so that their collective wisdom is directed toward solution generation without fusing their error in other domains (which occurs when making a company decision recommendation) and thus generating judgement noise. AI can also calculate interaction effects through economic modeling, improving the balancing when considering multiple contributors’ differing strategic recommendations. This is not to give the impression of naive positivism: machine learning isn’t “solving the game” of economics. But with the backing of silicon, groups can merge their values and various educated judgements in a way that bypasses many of the typical humanistic problems plaguing rationality. Intellext offers the dream of many a science fiction story; human augmentation. Make better, faster, more informed decisions with AI. Be you, but stronger.
About Intellext™
Intellext is an AI startup that is revolutionizing the way contracts are negotiated, accelerating time to close, and improving deal terms. Intellext’s Intelligent Negotiation Platform™ eliminates the complexities of contract redlines and stakeholder collaboration and optimizes deal terms by applying machine learning during the negotiation process.