A summary of some organizational management concepts applied to MakerDao

I forgot that there is academic work on organizational theory so I dived into that for a bit. Thanks for reminding me @makerman. Don’t expect any answers here, organization is a deep and difficult subject, we have a ton of work if we wish to apply the academic frameworks used to describe organizational management. Below is a crude summary of what I read about and how I understood it in the context of Maker. Below that are prose exploring the subject more. Shoutout LFW for some comments before. No worries if no one reads this I learned alot. The meta organization paper and discussion will probably end up being the most useful eventually. some of the other concepts are slightly dated IMO.

SUMMARY

Part I Organizational Management

  • Defined structure of labour tasks determines coordination modes. (so the concepts we use to divide work (risk, technical ect) partially determines the practical/optimal coordination mode
    • Modeling the interdependencies and uncertainties of the tasks undertaken can clarify which coordination mechanisms are most suitable (in the context of agreed goals of the organization). (if anyone knows of the theory/software/mathematical methods of doing this let me know how to find them.)
      • Interdependence of task can be viewed multiple ways
        • “Correlation of variables in task or environment space”
        • Workflow, how much does your task depends on the completion of other tasks? Possible mathematical representations with matrices
      • Uncertain tasks result from high unpredictability and complexity
    • In Maker, much of the flow of work goes through governance at some point in time. Looking at governance as a highly interdependent coordination problem
  • Coordination modes, ‘programming’ (impersonal) vs ‘feedback’ (personal or group) (basically what can/should we automate in terms of communication and when do modes break down into others? )
    • Increasing task uncertainty increases use of personal and group while decreases impersonal modes
    • On chain governance provides information always in an impersonal mode
  • M (multidivisional) vs U (unitary) form, the traditional structures of organizational management
    • M form considered “decentralized” A central organization can own others that operate autonomously
    • U form describes direct management with stronger specialization in each division
  • Work unit size another consideration
    • See Young GF, Scardovi L, Cavagna A, Giardina I, Leonard NE (2013) Starling Flock Networks Manage Uncertainty in Consensus at Low Cost. PLoS Comput Biol 9(1): e1002894. https://doi.org/10.1371/journal.pcbi.1002894 for this question explored in a slightly different context. Basically starlings pay attention to their closest 7 neighbors to manage consensus. Bio networks rich field to learn about systems and networks in general IMO

Part II Meta Organization

  • Meta organizations have a variation on formal authority in the system and are comprised of legal autonomous actors
    • Describing the boundaries of the organization helps understand the variation of formal authority in the organization
    • Internal stratification (another form of boundary) the other determinant of variation in formal authority
      • Lots of interesting work that could be done on exploring these two parameters
    • From what I read MakerDao fits under the meta organization approach, mainly for the lack of a formal authority in the employer-employee context
      • However we do have a formal authority (governance system) but it is different than traditional authority in a legal sense (not that we have any precedent for legal obligations of DAO stakeholders)
  • The micro structure of an organization can/should be analyzed in the wider context of its relationships/interactions with other groups
    • DAOs have not been academically analyzed (to my knowledge)
    • How are we gonna form relationships with other organizations?

I have a section with my thoughts after the first two, some points/questions:

  • Introducing an idea that mostly follows the path we are on: Divisional facilitators for the various critical domains. These facilitators are salaried and given a budget from governance. Rich would likely reside of “community/governance” domain and cyrus “risk” Nik “Oracles” ect.
    • The facilitators choose modes of compensation that best suits the task in their domain. As they should be experienced and connected in their field
    • The facilitators are responsible for recruitment of workers
    • Propose budgets for their domain
  • Net capital VS labor split of revenue interesting topic in Maker context
  • How do we measure task completion and facilitate coordination?
    • We need to incentivize both task completion and coordination
  • What are the critical domains of the system that need incentivization?
  • Controlled experimentation can give us a basis to start designing with confidence

CITED

Burton, R.M. & Obel, B. J Org Design (2018) 7: 5. https://doi.org/10.1186/s41469-018-0029-2

Gulati R, Puranam P, Tushman M (2012) Meta-organization design: rethinking design in interorganizational and community contexts. Strateg Manag J 33(6):571–586

Qian YY, Roland G, Xu CG. Coordination and experimentation in M-form and U-form organizations. J Polit Econ. 2006;114(2):366–402. doi: 10.1086/501170

Van De Ven, Andrew H., et al. “Determinants of Coordination Modes within Organizations.” American Sociological Review, vol. 41, no. 2, 1976, pp. 322–338. JSTOR, www.jstor.org/stable/2094477. Accessed 13 Jan. 2020.

Bustamante, Arturo Vargas. “U-Form vs. M-Form: How to Understand Decision Autonomy Under Healthcare Decentralization? Comment on “Decentralisation of Health Services in Fiji: A Decision Space Analysis”.” International journal of health policy and management vol. 5,9 561-563. 1 Sep. 2016, doi:10.15171/ijhpm.2016.73

Young GF, Scardovi L, Cavagna A, Giardina I, Leonard NE (2013) Starling Flock Networks Manage Uncertainty in Consensus at Low Cost. PLoS Comput Biol 9(1): e1002894. https://doi.org/10.1371/journal.pcbi.1002894

PART I ORGANIZATIONAL MANAGEMENT

“Organization design is a systematic approach to aligning structures, processes, leadership, culture, people, practices, and metrics to enable organizations to achieve their mission and strategy.”

“An organizational design must specify the fit between the structure of division of tasks in the organization with its coordination, or how to make these tasks work in concert.” Burton, R.M. & Obel, B. J Org Design (2018)"

We probably all intuitively understand those, but it gives as another good practical step for viewing the management problem. What divisions in labor do we wish to define to make the various actors in Maker work in concert. I recommend we determine the nature of the coordination between the groups completing tasks that we expect are required.

Task interdependence, uncertainty and worker unit size informs the coordination process (Van De Ven, Andrew H., et al). In other words, who(s) needs to talk to who(s) about what(s) in order to pursue a goal. The coordination occurs in three modes, impersonal, personal and group.

Starting with describing interdependence:

Interdependency can be defined as the correlation among the variables in the environmental space or task space. Simon (1996) examines interdependencies as the degree of divisibility or decomposability using a matrix representation of the connections. The more connected or dense the matrix, the more interdependent the tasks; and the sparser the matrix entries, the less connected and the more divisible tasks. (Burton, R.M. & Obel, B)

Tasks that can be done without input from other tasks are more divisible. Task interdependence above was defined as the correlation of variables in the environment or task space. Van De Ven, Andrew H., et al looks in terms of work flow; When one task depends on other completed tasks, the more interdependent a workflow is and more processing capacity is required.

In our situation, many results of tasks have to be relayed to maker token holders. In other words, to do their tasks (voting) maker token holders depend on the completion and transfer of information from other units like risk, community, or integrations. In other situations units obtain information from governance which changes their work. In both directions interdependencies increasingly flow through governance as Maker becomes more DAOlike.

Uncertainty also influences the coordination mode. The authors hypothesize that tasks which increase in uncertainty have more trouble being communicated through impersonal means. Increasing the use of personal and group modes (explored below). Description of task uncertainty included variability and difficulty (I perceived more as unpredictability/complexity) (Van De Ven, Andrew H., et al.). Complexity can imply more interdependence depending on the context of the task. The unpredictability of a task intuitively implies a more uncertain task.

Let’s look at descriptions of coordination by Van De Ven, Andrew H., et al. They viewed coordination in three modes, impersonal (programming), personal and group (feedback). Different situations require different modes of coordination to be effective. Task uncertainty and interdependence may influence which modes are most effective for certain tasks.

Impersonal coordination drive tasks through clear mechanisms which rely on pre established rules, procedures and standardized information systems. A common tread with impersonal coordination mechanisms is that departure from them is obvious. For makerdao I would consider some examples to be:smart contract logic, signaling procedures, blockchain consensus and governance (probably some others). MKR voters always communicate through impersonal means. The financial stats of the protocol are relatively certain and do well with impersonal communication as well.

Personal and group modes rely on human occupant(s) of a position to manage task adjustments (using vertical or horizontal communication),. Task adjustments could be changes in methods, tools, goals, and data initiated by a peer or through a vertical authority.

Alot of this is muddy when considering impersonal vs personal/group modes in an internet context, since at some points communication follows impersonal logic through software processing, but ultimately the information exchange between the two individuals is considered as a personal connection. Web meetings, forum posts, ect.

For thinking about our acse, we don’t want to suffocate a risk team by having strict impersonal procedures around dynamic risk analysis involving other groups, but we also don’t need to leave simpler, defined, unrelated tasks up to high subjectivity.

Work unit size is another primary variable in coordination. Van De Ven, Andrew H., et al admits incomplete understanding of the relationship between unit size and coordination. They do hypothesize an inverse relationship to task uncertainty and interdependence. That an increase in work unit size generally leads to a decrease in group, increase personal and strongly increase impersonal modes of coordination.’

We should consider ecological groups structures as well (as they evolve to thrive in dynamic uncertain circumstances). One paper by the name of Starling Flock Networks Manage Uncertainty in Consensus at Low Cost exploerses coordination in starling flocks (for those ignorant of starlings: https://www.youtube.com/watch?v=V4f_1_r80RY). Researchers tried to understand how coordination of these flocks occurs in chaotic uncertain environments. For starlings, paying attention to the 7 closest neighbors optimizes the balance between group cohesion and individual effort on multiple ranges of flock size, shape and thickness. Robustness here means resilience to noise in relation to a goal (which they had as flock direction)

“Although here we have focussed on the sensing strategy of interacting with m nearest neighbors, our methods can also be applied to networks resulting from any sensing strategy. For example, our methods could be used to evaluate the robustness to noise of zonal sensing strategies like those used in [5,8,17] as a function of a parameter such as zone size. Provided that a real or hypothesized sensing network can be constructed, its robustness can be calculated.”

Noise can make global consensus ( in this case flock direction, in ours, less contested votes? Agreed culture/goal direction?) impossible at a critical point. Determining what noise is considered in Maker and how to manage it seems important. The math used in this paper has more general implications beyond flocks of birds, but personally I don’t understand it.

M form vs U form.

The quality of the coordination depends on the quality of communication within/between organizations (Qian YY, Roland G, Xu CG. ). Two traditional constructs used to describe organizational structure are M form (multidivisional ) and U form (unitary form). M form describes situations where one parent company owning many other, but largely let them operate autonomously. “we define an M-form (multi-divisional form) organization as one that consists of “self-contained units” where complementary tasks are grouped together.” (Qian YY, Roland G, Xu CG. ). M form exhibits more flexibility for experimentation. The less specialized divisions in an M form structure better promote this. The authors also noted that when there is a low probability of success for innovation, small scale experimentation in M form is especially useful.

U form, higher managerial centralization and specialization of other units

Bustamante, Arturo Varga looking at M vs U form in healthcare decentralization:

“Under this model, the more decentralized organizational form (M-form) is superior if the benefits from flexibility exceed the costs of duplication and the more centralized organizational form (U-form) is superior if the savings from economies of scale outweigh the costly decision-making process from the center to the regions. Budgetary and financial autonomy and effective mechanisms to maintain local governments accountable for their spending behavior are key decision autonomy variables that could sway the cost-benefit analysis of healthcare decentralization.”

Another way of saying the above: M form can be seen as less efficient, but has better adaptive behaviour because of its semi autonomous divisions. U form, if grown to an economy of scale and not ran into the ground with bad decisions by central authority, presents superior profit.

Both Authors made similar points.

Maker already appears to have an informal M form structure. As I looked further some of the principles of meta organizations help define maker a bit clearer

PART II META ORGANIZATION

Primary source:
Gulati R, Puranam P, Tushman M (2012) Meta-organization design: rethinking design in interorganizational and community contexts. Strateg Manag J 33(6):571–586

“Every meta-organization must, by our definition, employ some substitutes for formal authority, but the precise manner in which those alternatives are generated and exercised exhibit systematic variations. We argue that patterns within this variation may be understood by considering two important dimensions of meta-organizations: the degree to which a meta-organization’s boundaries are open or closed and the degree of its internal stratification.”

Formal authority:

The foundation plays a large role in the functioning of the Maker protocol, which complicates this question right now. I am unaware of what exact control the executive foundation multisig can influence the protocol. Mkr voters hypothetical play the role of legitimate formal authority, but this goal requires a transition state, which is ongoing. A recent sign of progress with changes in contract ownership and mcd deployment as a whole.

Even with mkr holders as the “legitimate” formal authority, potentially non voting actors control certain aspects of the protocol. Oracle, legal, technical, risk and pre chain governance come to mind when considering fairly internal groups that can exert pointed influence on the protocol. Currently we move from a situation where a single vertical controls most of these groups. Interplay between voters who “own” the protocol and the labour groups that provide insight, energy, skill, is still emerging.

Boundaries:

“The meta-organization’s boundaries provide a basis for members’ identification with the collective and the collective’s differentiation from others. The essential aspects of such boundary arrangements include (1) who chooses members (2) criteria for membership (i.e., the attributes members possess and the degree of redundancy between them); and (3) duration and exclusivity of membership (i.e., whether members can belong to more than one meta-organization).”

So how do members identify with themselves and the collective? And how does the collective differentiate itself from others.What comprises the semi permeable membrane that regulates the flow of information into the system. 1-3 below match to the quote above:

  1. Anyone who can access the internet, read (english primarily for now, but communities with other languages exist, I’m unsure how active they are) and engage successfully with one aspect of the protocol can be a “member”. To a degree members self select, but the reception and acceptance of output by traditional members also influences who is encouraged to participate. Members also don’t necessarily know who the other members are due to scale and division in interest.

Someone could be a member without engaging with any actual people. (eg own crypto, buy mkr, mega whale, learn everything, vote and control)

The goal of making DAI a global, stable, open currency provides the broadest agreed goal

  1. Real funky question. I see a few ways of joining the community, and a continuum of rigidity for defining those methods. First, A rigid, objective process to join the community: Buy mkr and vote. Second, less rigid: participating in a discussion which eventually changes an aspect of the protocol through governance. All sorts of loose social and internet interactions also could lead to some engagement with the protocol. Any technical work directly on the protocol requires passage through governance, but secondary applications only requires interactions with the protocol through ethereum.

Journalistic work can influence the protocol, but not always. Depends on degree and goal of engagement from the author.

  1. Duration and exclusivity of membership. Knowledge, money, and access to technology provide the exclusivity. Duration could be any amount of time after the first interaction occurred. (Brings up an interesting question about rejoining after long periods of absences)

Successful capture of people’s attention is essential for a MKRless path to engagement and identification with the community.

How do we differentiate between more extreme polarities in engagement? Maker curious folk that have little to no understanding of the protocol vs the most active informed individuals? In the attention context, basic questions do not hold many people’s attention for very long (or not at all), concise useful summaries of things like the SCD shutdown debate do excite different individuals for a longer period of time.

Internal Stratification:

“Where stratification is present, the upper tiers enjoy more extensive decision-making rights, bear more responsibility for coordinating the activities of the lower tiers, and participate in the meta-organization’s design decisions.”

Maker does exhibit some stratification in every group. The clearest example is voting groups, voting power depends on the amount of mkr used to vote. They hold more responsibility because they stand to lose the most in an mkr dilution event or other failures. Other domain exhibit stratification as well. Cyrus as the interim risk head and Rich running many governance processes are two clear examples. These individuals clearly hold more responsibility/reputation through their employment with the foundation and ratification from governance, delineating them from others.

One interesting deviation from the quote above pertains to participation in the meta organization design changes. Recently, many changes to the protocol increasingly come from the community and not from mkr whales (well unless they keep their holdings secret as they participate) or the primary domain heads. The balance of power in structural changes is fairly unclear, partly because we don’t know which community members actively vote.

“Collectives that rely entirely on self selected membership may find it more difficult to fill competence gaps and to ensure coordination or task completion, since exit from the collective is as easy as entry”

I think most agree that relying completely on self selection for task completion would be unwise for Maker. We are attempting to stimulate the flow of attention that comes through the DAO to fill these competency gaps. There are some tasks that mkr voters likely prioritize over others. Coordination, completion and stimulation of those tasks through compensation is the basis for the organization structure.

SOME of my interpretations, thoughts descriptions of possible future work:
Compensation

Now for “fair” compensation, how do we fairly compensate individuals for work. Should these be measure by time? output? Responses? I see this coming down to: What can the DAO measure, how can the DAO measure it (impersonally or personally)? How to ensure/review accurate measurement (audits)? And finally, loose agreements on what measurements should earn what compensation.

The issue of measurement is extremely interesting to me. SourceCred seems to have a decent way of measuring and compensating activity in open places like GitHub or discourse. Their focus appears to be measuring the output and the effectiveness of the output to determine compensation. As a secondary compensation mechanism I see potential, but I don’t quite see how ongoing unpublished or private work could be compensated in this automated fashion. I completely support experimentations with this sort of system though.

Experimentation with what method of compensation works best within a domain should also be encouraged. Leaders within a domain likely have an intuitive sense of what methods of payment efficiently provoke useful work. Alot of the work done at Maker is done in google docs and meetings. To me it seems clear that to evaluate work done in a specific area, those interacting in that area seem best suited for determining the rate and mode for fair compensation. Compensation for a recurring weekly summary should probably look different than a comprehensive assessment of a companies fundamental value.

Right now we have one main head for each domain, Rich for governance, Nik for oracles, Cyrus for risk and someone for legal aha.

Here is my hypothetical scenario, Rich facilitating governance/community as an example

So an executive vote goes through which defines a contract between rich and makerdao. This contract specifies Rich’s roles and responsibilities to the protocol. Now, recruitment and compensation is his main responsibility. Along with a weekly salary for himself, he is granted a budget to allocate to contributions from various individuals or groups. These funds could be accessed through an IAM with access to the buffer or another treasury mechanism.

During negotiations for this contract Rich formed a rough chart of expenses for what he saw as necessary to support his domain, including communications (press, forum, summaries), onboarding coordination, meetings, new procedural documentation and coordination between domain teams to ensure accurate polls. He may design a recurring grant for various weekly contributions, lump sum payments for one time work like redesigning the discourse’s art or creating procedure, and a SourceCred like algorithm that incentivizes general community participation in discussion. Each month Rich publishes a brief report shows the DAI spent for whatever tasks. He may want to make an argument for increased budget due to an especially high need for collateral application coordination or may point out that he consistently does not use the whole monthly budget and that RISK may need the money more badly.

In this situation the way to scale would be to eventually design “Rich” as a group of “Richs” We also need to consider backup people (think understudies?) that can step in in case something happens to a domain facilitator. Employing some redundancy in personeel specialized in maker seems safer operationally, @makerman made that point. Maker will have to run 24/7 eventually right? To operate globally? Also why can’t we translate everything to everything? Has google not figured that out yet?

Also need to eventually consider what Capital vs Labour split looks like in maker. In the interview podcast SourceCred people mentioned it is the fundamental issue their trying to figure out on their end. They noted a general bias towards labour because of a feeling that they truly create the value in the product. Maybe not the time to discuss this, but understanding what revenues goes to what groups should help with determining a longer term balance.

A list of (expected) labours will help see what sort of coordination is required. Modeling the related interdependencies, uncertainties and unit sizes of the tasks they undertake can clarify what coordination mechanisms are most suitable (in the context of agreed goals of the organization). Let’s make a list of the various groups required to run the system and what task they do/the interdepencies of those tasks (INCOMPLETE):

Let’s look at uncertainties as tasks that change in nature and depend on many other variables. One example for thinking about uncertainty and modes of communication: Think about what work the integrations team does to add a collateral type, they read information from the blockchain and write code to make two other distinct pieces of code compatible. Code runs on logic so anyone can test the logic to make sure it does as designed. A community coordinator organizes all sorts of different people around a really complicated concept that is DAI. People do not submit to logic, so organizing people is a potentially more variable, task than logic based work

*Need to figure out how to visually represent

Fyi all these spheres commingle

Risk (ask Cyrus what ways contributions could be measured)

  • Makes tools to analyze data from the protocol
  • Makes tools to analyze data from outside the protocol (for collateral onboarding)
  • Presents the outputs of those tools and the context they were formed with

Data

  • Ensures feed information through a carefully designed informational network.
    • Maintains relationships with service providers
    • Updates/adjust protocol for handling price feeds
  • Vets price feeds for new collateral types
  • Risk operators must choose data sources, this may interact with oracles domain

Accountant (speculative)

  • Manage the DAOs debts and assets, also other speculative holdings.
    • Communicates in intervals to all

Voters

  • Voters pass governance polls then enact executive votes to change the protocol based on the governance polls
    • Governance polls are developed in informal settings, mainly the forum and governance calls
  • Active voters likely read, watch or talk to other people involved in the system to determine how they vote. Or they are involved deeply already
  • Affirms goals with votes
  • Votes on mechanisms/standards to derive other polls

Technical

  • Writes the code needed to implement change in the system

    • Adaptors for new collateral
  • Coordinate with other domain teams on new collateral types

  • any vote requiring code

  • Implementing MIP?

Communications

  • Information processed, made accessible and public
    • Internally based forum posts on the various inner workings of the protocol or its external relationships
    • Externally formed articles reporting on the protocol or its interactions
    • Help communications between domain groups
    • Communications between domain groups and the community
  • Tools to view information
    • Governance
    • Risk
    • Technical (just github right? Could make some automated collector of github info)

Business/social

  • Engagement at meetups and conferences
    • Acts as social actor to develop people’s interest and knowledge of Maker
  • More insight to non maker entity perceptions
  • “external “ communications

Bureaucratic

  • Similar to communications, but focused on the formal procedures related to governance

Legal

  • Idk
    • Idk

EXPERIMENTATION!!!

Burton, R.M. & Obel, B. highlight the need for observation and experimentation in the ‘science’ of organizational design. MakerDao has a fundamentally research focused background, in organization science some of Makers designs goes beyond the concepts described. To make adequate decisions mkr holders should incentivize academically formed research to gather data and models on the system. This basically describes the financial risk process most accept at maker, why is the social risk neglected? Not everything is game theory sillys.

Starting work on models could look like an exhaustive list of known and expected variables in implementing decentralized management of the protocol.

I would definitely appreciate tips on this one. If you read all of this holy shit. 0x67c39d3b2C87096dC700Da104D58d7abC0dc7f9C

4 Likes

Do you have a Gitcoin account i will send you a Kudo :slight_smile:

This worth a Kudo of course. Great work @Mitote

NIce post @Mitote

Off the top of my head comments:

First I always maintained working groups over 8 (i.e. meetings with more than 8 people) where the individuals were equally important basically got too large. In effect the time to achieve consensus grows somewhat like the square of the number of individuals in a group because of all the possible pair interactions (i.e. communication - usually to disagree). These tend to start at around 5 and can become extreme at > 10 (to the point that around 12-15 people there ALWAYS is at least one and possibly two people who simply will disagree vehemently regarding whatever consensus the rest of the group puts forward).

In the context of the above I found the starling 7 neighbors interesting because if you include the one looking at the neighbors it exactly matches my 8 derived from personal experience.

Second thought and I’m not sure if/how this applies to the above (because your post is pretty general regarding organizations) is the idea to divide organizational tasks into ‘routine’ - things that have to be done repeatedly (weely reports, polls, governance votes, etc), , vs. things that are one offs (smart contract upgrades/changes, etc.). Organizations tend to have a lot of routine tasks and focusing on creating a good structure to deal with these tasks efficiently often provides good focus and clarity on the basic organizational needs/productivity/communication costs and often can give guidance on a good organizational structure. The one-offs by their nature tend to consume more organizational communication time and can be difficult to streamline because they tend to have greater complexity and touch on many operational divisions for coordination and input. These are the most difficult tasks for organizations to deal with optimizing and will consume the greatest organizational time resources generally. Keeping these one-off organizational processes manageable so they DO NOT completely consume available organizational resources is a key element to organizational success (at least in my experience).

2 Likes

Thanks, I just made one @mitote1

Huh good to see that number match your experience. Yeah those routine tasks seem like a good place to nail down a process. They have low variability and complexity, so we should be able to make a solid defined impersonal structure to complete/communicate them.

I will send you one today @Mitote :slight_smile:

Some good ones I’m working through

https://www.researchgate.net/publication/264440930_Leadership_for_self-organisation_complexity_theory_and_communicative_action
“Leadership for self-organisation: complexity theory and communicative action”

https://journals.sagepub.com/doi/pdf/10.1177/0170840618783342
“Criticality: How Changes Preserve Stability in Self-Organizing Systems”