Question 2: Where Is the Variety Mismatch?
The short version
Every act of control involves a regulator and a system being regulated. For the regulator to work, it must be able to match (or cover) every state the system can be in with a corresponding response. When the system can produce more distinct situations than the regulator has responses to, control fails. Not sometimes. Every time those unmatched states appear.
The deficit is structural: a mathematical property of the relationship between two systems, independent of effort or intelligence. Ashby’s Law of Requisite Variety states it precisely:
Only variety can absorb variety.
The variety of the regulator must be greater than or equal to the variety of the system being regulated. When it is not, the deficit is the exact measure of the regulator’s uncontrollable residue: the states it cannot respond to, the situations it will always handle poorly.
The second diagnostic question is: where, specifically, is the variety mismatch, and which side of the equation is wrong?
Why this question is non-obvious
The standard assumption in management, policy, and governance is that complexity can be overcome through effort, intelligence, or authority. If the system is out of control, the response is: try harder, hire more people, issue stronger directives, apply more pressure.
Ashby’s Law says this assumption is false, not in practice, but in principle. It is a mathematical result, not a management preference.
Variety is Ashby’s formal name for complexity. It is defined simply: the number of distinct states a system can be in. A traffic light has four states (red, amber, green, off): its variety is four. A patient presenting to a hospital can be in an enormous number of distinct clinical states. A financial market can produce combinations of instrument values, correlations, and liquidity conditions that no human analyst has ever enumerated. A city’s daily traffic patterns are a space of states that defies complete description.
The critical point: you cannot know, in advance, which of those states the system will be in at any given moment. The regulator has to be able to respond to whichever state appears, and if it encounters a state it has no response to, it fails to regulate.
This is why more effort does not solve a structural variety deficit. A hospital administrator working twice as hard does not double the variety of responses available to them. A financial regulator issuing more guidance does not increase the number of distinct instrument configurations they can handle. Effort is a scalar on existing responses: it does not add new ones.
What does add new responses is deliberately building requisite variety, and that is a design question, not a motivation question.
Mapping the variety equation
Any control relationship has three elements that must each be examined:
The system’s variety: how many distinct states can the system being regulated be in, and how fast does it move between them?
The regulator’s variety: how many distinct responses does the regulator have available, and how quickly can it deploy them?
The channel and transducer: is the connection between regulator and system wide enough to carry the necessary signals? And at every boundary crossing, is the information being correctly translated into a form the receiver can act on?
All three must hold. A regulator with sufficient variety that receives no signal about the system’s state cannot regulate it. A regulator with a clear signal but no matching response cannot regulate it. A regulator with both variety and signal, but whose communication channel is saturated, cannot regulate it.
Most variety analyses focus only on the first two (the blocks of variety) and miss the channel and transducer failures that are often the actual point of breakdown.
The two responses to a variety deficit
When the system’s variety exceeds the regulator’s variety, there are exactly two ways to rebalance the equation:
Amplify the regulator’s variety: increase the number of distinct responses available to the controller.
Delegation is the most powerful amplifier. A manager who passes decisions down to local units is not abdicating control: they are amplifying their regulatory variety by adding the response repertoires of every unit they delegate to. Tools, sensors, real-time data, expert systems, and specialist teams are all amplifiers. Police with fast cars, radios, and coordinated dispatch have more regulatory variety than police on foot with no communication. A doctor with a diagnostic scanner has more clinical variety than a doctor relying on visual examination alone.
Attenuate the system’s variety: reduce the number of distinct states the regulator has to respond to.
Filters, standards, eligibility criteria, simplified interfaces, and rules are all attenuators. Building regulations reduce the variety of structural configurations that need to be individually assessed. Job specifications reduce the variety of applicants a hiring process has to evaluate. A menu reduces the variety of meal requests a kitchen has to handle. Attenuators are not failures of control: they are essential tools for making control feasible. The question is whether they attenuate the right variety, and how much.
Every management and governance technique that has ever existed is one of these two things, applied to one of the connections in the control relationship. This is not an overstatement. Ashby and Beer offered it as a challenge (name a technique that is neither) and it has not been answered.
The diagnostic implication: when you encounter a failing control system, find the four connections (regulator to system, system to regulator, system to environment, regulator to environment) and ask, for each: is the variety being amplified or attenuated here? Is the balance right? If not, which specific amplifier is undersized, or which attenuator is missing or misplaced?
Applying the question
Four diagnostic moves, in sequence:
1. Count the states on the system side. This does not require an exact number: it requires an order-of-magnitude sense of the system’s variety relative to the regulator’s variety. Is the system producing one or two unexpected states per year, or one per week, or continuously? Are the unexpected states novel (the regulator has never seen them) or just numerous (the regulator has seen them but cannot process them all)?
Novel unexpected states indicate a structural variety deficit: the regulator’s response repertoire is too narrow. Numerous expected states indicate a throughput problem: the regulator has the right responses but cannot deploy them fast enough. These require different interventions.
2. Count the states on the regulator side. How many distinct responses does the regulator have? This is often surprising: organisations that appear complex frequently have a very small number of actual response types: approve, reject, escalate, defer. The variety of their internal deliberation is large; the variety of their outputs is small.
When the output variety is low relative to input variety, the regulator is compressing information rather than regulating it. It is producing average responses to non-average situations. In stable environments, this is efficient. In volatile environments, it is the source of systematic failure at the tails: the situations that needed a non-average response.
3. Check the channel. Is information about the system’s state actually reaching the regulator? Not in principle, but in practice. Many control failures are not variety deficits at all: the regulator has the right responses, the system is in a manageable state, but the signal is not getting through.
Common channel failures: reporting cycles that are too slow (monthly reports in a daily-scale system), transducers that translate raw data into the wrong format (operational data converted into financial summaries before reaching the manager who needs operational intelligence), channels that are saturated (the manager receives more reports than they can process, so they stop reading them), and political filters that edit signal before it reaches the regulator.
Beer’s steel company example: years of rich sales data accumulated on punch cards. No one built the transducer to convert that data into management intelligence. The cards sat on shelves. All that variety (all that information about the system’s state) was present but could not reach the regulator. The control failure was not a variety deficit; it was a transducer failure.
4. Identify who is absorbing the residual variety. When a regulator cannot absorb a system’s variety, the unabsorbed variety does not disappear: it falls somewhere. Find where it lands. In organisations, unabsorbed variety typically falls on front-line workers: the nurse who cannot get the system to respond to an unusual clinical situation improvises; the customer service agent who cannot get a decision handles the complaint themselves; the local official who cannot get a policy clarification makes a judgment call.
These are not failures of discipline. They are the system doing what is necessary to maintain homeostasis when the formal control loop has broken down. The improvisation tells you exactly where the variety deficit is. The people absorbing residual variety are functioning as informal regulators, which means the actual control architecture is different from the official one.
The micromanagement trap
The micromanagement case is worth examining carefully because it appears, superficially, to be an attempt to increase regulatory variety, and it produces the opposite.
A manager who is alarmed by apparent loss of control responds by increasing their involvement in decisions. They attend more meetings, require more approval steps, review more outputs. This looks like amplifying the regulator’s variety. But it has two destructive consequences:
First, it saturates the manager’s channel. They are now processing more signals than they can respond to distinctly. Their actual response variety decreases: everything gets the same reply because they no longer have time to give differentiated ones.
Second, and more importantly, it reduces the variety of the units being managed. When a team knows every decision will be reviewed and reversed, they stop developing their own judgment. They learn to escalate everything and decide nothing. Their response repertoire atrophies. The total variety available to the system (the combined variety of manager and team) is smaller after micromanagement than before it.
The manager has attempted to increase control by concentrating variety, but concentration of variety is not the same as requisite variety. The system now has one controller with insufficient variety attempting to regulate a system it cannot match: exactly the deficit the manager was trying to avoid, but worse, because the team’s adaptive capacity has been destroyed in the process.
The cybernetic correction: control variety must be distributed to match the variety of the system being regulated. Delegation is not a loss of control: it is the mechanism by which requisite variety is achieved. The manager’s role is to set the policies within which local variety operates, monitor for instability, and intervene at the level of policy, not at the level of individual decisions.
The failure mode this question reveals
In financial regulation: The variety of financial instruments grew exponentially through the 1990s and 2000s. Derivatives, structured products, and synthetic instruments created combinations of exposure, correlation, and leverage that no previous regulatory framework had responses for. The regulatory bodies retained the response repertoire they had developed for simpler instruments. The Ashby violation was not hidden: it was visible to anyone who counted the states. The 2008 crisis was the system visiting states the regulator had no response to.
In healthcare: A GP consultation of ten minutes has a fixed variety. A patient’s clinical situation (especially a patient with multiple chronic conditions, social complexity, and mental health needs) has a variety orders of magnitude larger. The general practice system is a regulator with structurally insufficient variety for a significant fraction of its patient population. The patients who fall through the cracks are not bad luck. They are the unabsorbed variety.
In school governance: A standardised curriculum and assessment system has a fixed variety: it produces responses to a predefined set of student performances. The actual variety of students’ learning needs, contexts, and trajectories vastly exceeds this. The students who fail in standardised systems are not, primarily, failing to learn. They are producing variety that the system has no response to. The system cannot help them because it has no response for their state.
In AI governance: The behavioral variety of a large language model or a recommendation algorithm (the number of distinct outputs it can produce in distinct contexts) is effectively unlimited from the perspective of any current regulatory body. The regulatory response repertoire (approve/reject, fine, require audit) has a variety of perhaps five or six distinct actions. The Ashby violation is extreme. Every AI governance framework operating with this response repertoire will fail to regulate the systems it is responsible for, not because of lack of effort or expertise, but because the variety equation is wrong by multiple orders of magnitude.
In organisational change management: A change programme that issues communications, runs workshops, and monitors survey scores has a fixed response repertoire. The variety of staff responses to change (the combinations of concerns, resistance types, informal influence patterns, and emotional states) is vastly larger. The programme reliably absorbs the responses it was designed for (overt resistance, expressed concerns, visible non-compliance) and fails to regulate the ones it wasn’t (covert resistance, informal counter-narratives, passive non-engagement, loyalty to previous leadership). These are not rogue actors. They are the residual variety the programme cannot absorb.
What the answer tells you
Once you have located the variety mismatch (and identified whether it is a variety deficit, a channel failure, or a transducer problem), you have three intervention points:
1. If the gap is structural and unbridgeable at the current level: attenuate the system’s variety. Some systems are genuinely too complex to regulate with the resources available to a given regulator. The correct response is not to try harder but to reduce what needs to be regulated: by simplification, by scope reduction, by restructuring what reaches the regulator at all.
This is not defeat. Every viable control system uses attenuation: the question is whether it attenuates the right things. A well-designed filter removes variety that is irrelevant or low-stakes before it reaches the regulator, while ensuring that high-stakes variety reaches the regulator with full fidelity.
2. If the gap is bridgeable: amplify regulatory variety through the right mechanisms. Identify the specific response categories that are absent from the regulator’s repertoire and add them: through delegation, specialist capacity, real-time sensing, or tool design. Each amplifier should target the specific variety that is currently unabsorbed.
Generic capacity additions (more staff, bigger budget) rarely achieve this because they add response volume, not response variety. The question is not how many responses the regulator can produce but how many distinct types of response it can produce.
3. If the channel or transducer is the failure point: fix the signal path before fixing the variety blocks. If the regulator has sufficient variety but is not receiving the signal that tells it which response to deploy, fixing the variety blocks will not help. Find the point where signal is lost (too slow, wrong format, over-filtered, saturated) and restore it. This is often the cheapest and fastest intervention, and the most overlooked.
Before moving to Question 3
Question 2 tells you whether the control relationship between regulator and system is structurally sound, whether the variety equation is balanced, and if not, where the deficit lies.
But it does not tell you which specific function within the organisation or institution is responsible for managing that variety at each level. An organisation may have adequate variety at one level of its hierarchy while being severely deficient at another. A government department may have requisite variety for its operational decisions but no variety at all for strategic adaptation.
Locating the failure at the level of the whole organisation does not tell you which architectural component is broken. That is the subject of Question 3.
Reference
The theoretical foundation for this question is Ashby’s Law of Requisite Variety, developed by W. Ross Ashby.
Primary sources:
- Design for a Brain (1952)
- An Introduction to Cybernetics (1956): Chapter 11 contains the formal proof
The law is a theorem, not an empirical generalisation. It follows from the mathematics of information theory and does not depend on domain. Its practical application to management was developed by Stafford Beer throughout Brain of the Firm (1972) and The Heart of Enterprise (1979), and demonstrated operationally in Project Cybersyn (1971–1973).