institutional design

group sizes

Some important group size thresholds (many of these are approximate):

0: an automated process
1: a single person; no communications difficulties, no time spent in meetings, no potentially ruinous power struggles, no politics
2: two people can easily and efficiently have deep conversations. Problem: as this is an even number, if the two people are equals there is a danger of deadlock in case of disagreement.
3: three people can still have deep conversations, but less efficiently than 2 people. If all three are equals, deadlock is not a danger, but politics begins, as a faction of two people can cut out the third
4-8: informal social mechanisms can still be used to govern conversation, but deep conversations become inefficient/difficult if everyone has things to say; there is a lot of waiting for other people to speak and then by the time it's your turn, the conversation has moved on from the topic you wanted to address; informal meetings easily drift off track. This range is often recommended as the optimal size for hands-on committees, boards, etc, with 7 being a commonly given number (but note that some people think that eight is a bad size for committees). This is the range of the 'two pizza team' rule and similar suggestions [1] [2] [3]. A manager with more than 4 reports probably should be a full-time manager (as opposed to someone who manages but also does hands-on work themselves) [4]. This is also the range of most commonly-recommended and observed managerial span of control [5] [6]. This is also about the range in which a small number of competing firms tend to be considered an [7] (which suggests that this may be the upper range in which mutual trust is possible in the context of a typical Prisoner's Dilemma-like situation).
9-14: Transition between small group and mid-sized group. Typically, some sort of formal mechanism or hierarchical power structure is needed to moderate conversations. Sometimes can work the same way as 4-8, but that's pushing it.
15-50: Can no longer be considered a 'small group'. This range contains many legal thresholds for 'small companies' and small groups (eg 15, 25, 35, 50) [8] [9]. This also contains the range that is sometimes considered to be unworkable as a traditional managerial span of control, but possibly workable for situations in which the manager only needs to make themselves available to subordinates, rather than actively manage their projects [10] [11]. Some boards of trustees of non-profit institutions that want to have large boards for the sake of fundraising are this size (or even a little larger).
51-150: larger than a small group, but still smaller than [12]. At the larger end, these groups are getting too large for everyone to know each other (although the more outgoing people will know everyone). These groups are still small enough that if they wanted to, they could have an all-hands meeting in one place, and (assuming that most people won't want to speak) to give everyone at the meeting who wants to speak briefly a chance to do so once or twice.
151-1000: In this range, having substantial unscripted back-and-forth in an all-hands meeting becomes infeasible, and indeed it may be hard to even find a space to fit everyone (but not impossible, even in ancient times eg ancient Athens might have had a quorum of 6000 [13]). This range contains the higher end of estimates of [14] (eg 250), and the larger legal thresholds for 'small business' and medium business (eg 500, 1000) [15] [16]
1001+: a large group, by anyone's definition

In addition to ordinary communications difficulties and the possibility of power struggles and politics, as groups grow larger:

people have less time (per associate) to get to know each other and to keep up with the details of conflicts that don't directly involve them. This can lead to situations where other people are judged by the masses based on kneejerk reactions rather than a careful consideration of each situation's individual nuances. This can lead to an unwillingness of individuals to express their true opinions, leading to political correctness in conversation, groupthink, the Abilene paradox
people may begin to bear loyalty to subgroups rather than the entire group.
even if there is voting or other mechanisms for every person to influence decisions, the chance that any one person (especially a person not in a leadership role) can change the decision grows smaller, leading to feelings of disempowerment, skepticism about group decisions, and a lack of care when exercising influence over the group's decision (since it's unlikely that one's influence will tip the decision, other considerations, such as signalling social affiliation to one's peers, become relatively more influential)
each individual has less influence, proportionately, on group performance. This leads to a reduced impact of equity incentives, and an increase in a "that's not my job" attitude and a consequent decrease in collaboration and group adaptability.
some argue that rigid/impartial adherence to comprehensive written policies (as opposed to subjective, personal consideration of the unique factors of each case by the leadership) becomes more valuable
as the group becomes larger, a random pair of individuals will know less about each other, and therefore trust each other less. This can lead to transitions from a regime in which people feel responsible to look out for each other's best interests, to a regime of favor-trading, and further to a regime of transactional/one-off interactions governed by explicit (often written) contract-type agreements.
relationships between people are always valuable, but as group size grops weak connections (acquaintances who encounter each other repeatedly, or even acquaintences who have only met once or a few times) become more important (and divisive)
people don't have time to keep up with group affairs
the group becomes more 'inwardly' focused, and a significant proportion of individual's resources and ambitions are oriented towards competing for status and power within the group, as opposed to achieving the group's goals

making decisions

decisions must get made, and if possible, made quickly. For motivation, having a command-and-control process that makes decisions quickly is preferably to an inclusive, consensus-building process that cannot come to a decision (this doesn't mean that i recommend decisions be made in a non-inclusive manner, it means that if you choose to make them in an inclusive manner you must still ensure that they are made).

meetings

meetings suck and if too much time is spent in meetings, it's the fault of the person who called/runs the meetings.

don't expect other people to hold back and not talk for long at the meeting just because they wish less time was spent in meetings. It's your job to enforce the agenda and time limits.

try to isolate big topics that could lead to long discussions (such as strategy) into their own meetings; you don't want these taking all the time and preventing other work from getting done.

If you are designing or architecting something, watch out for using meetings with more than a few people as a forum to bounce your own ideas off of others. How many is too many? 3 people, including you, is definitely not too many; i'm not sure how much is 'too many' but it's somewhere between 4 and 8 people. It's good to get feedback, but meetings are not the most efficient way to do this, because everyone must sit quietly while one person talks, so the more people there are, the more people are wasting their time sitting quietly waiting for others to finish. This is annoying to them. In addition, having too many people prevents deep discusssion. In addition, having too many people leads to 'design by committee', leading to bloated compromise designs instead of coherent, simple ones. Instead, consider having a series of one-on-ones or small meetings where you present your current ideas to others and hear their ideas.

Send a written summary of anything important decision taken at a meeting (this can sometimes reveal that not everyone interpreted the meeting outcome in the same way, which could be disasterous if unnoticed).

top-down strategy, bottom-up priorities and tactics

One institutional design that is often recommended is for overall company vision and strategy to be set in a top-down manner, to clearly communicate to each team what the vision and strategy is and what their role is in this big picture, and then to let each team determine for itself what is important to do, and how to do it.

This fits in with Goleman's "authoritative" leadership style.

don't make effectiveness report to efficiency

"avoid placing efficiency-based functions such as operations or quality control over effectiveness-based functions such as R&D, strategy, and training" -- Sisney

don't make long-term report to short-term

"avoid giving short-range functions like Sales, Operations, and Engineering power over long-range functions like Marketing, R&D, and People Development" -- Sisney

Centralize compliance

Centralize any functions that cause a risk to the company as a whole if they fail; e.g. legal, accounting, HR, risk management.

Centralize standardization

Centralize functions that must be standardized to form a basis for managing the company.

For example, if there are key metrics that are supposed to be used to manage the company, but different parts of the company use procedures and/or software for computing these metrics, then their numbers may not be direcly comparable. Instead, the function of producing these numbers should be centralized.

Centralize for efficiency?

I get the sense that large companies often centralize additional functions for the sake of efficiency. This saves money but at the cost of reducing the autonomy of parts of the organization. This isn't recommended for startups because at a small scale, the dollar value of these efficiency gains isn't very much.

Have metrics

Have metrics for as many parts of the company as possible, e.g. revenue, costs, efficiency, etc.

To take this to the extreme, you could even give each 7-person team their own metric (Amazon apparently does this: see http://blog.jasoncrawford.org/two-pizza-teams ).

beware: people will try to meet the metrics rather than to do good

Divisional vs. functional

Consider a company with multiple products or even multiple product lines. One strategy is to decentralize as much as possible. This is called a 'divisional' strategy. Each division has every function within itself except for a few centralized functions. For example, each division does its own marketing and its own engineering. The key property of a divisional organization is that a meaningful P&L (profit-and-loss) metric is calculated for each division, and this metric is what the division is accountable for. In other words, the divisional strategy is to push P&L down as low as possible. You might say, this isn't the real P&L because what about those remaining central functions? My answer is, the costs of those central functions scale sub-linearly with revenue (i bet they are log-scaling). Just be sure that any costs that scale linearly with revenue are in the division.

Another strategy is called 'functional'. Each function is a separate organizational unit, and each product involves interactions between all of these units. This strategy is more centralized and dependent upon the executive leadership.

Ownership of tasks

Accountability for a task should usually be 'owned' by a single role. This is particularly important for occasional tasks that no one else within the company is waiting on, as these may otherwise be forgotten about (examples: paying taxes; testing backups). Make sure that these tasks are not orphaned during reorganizations or when a person leaves.

RACI model

The RACI model stands for Responsible, Accountable, Consulted, Informed. The idea is that each person involved in a task or decision is in one of those 4 role types.

"Responsible" is someone who is actually doing the work. "Accountable" is the "boss" who is accountable for the work being done on time and correctly; the Accountable person delegates the task to the Responsible people, and signs off/approves when the work is done. "Consulted" is someone whose opinion is consulted (two-way communication). "Informed" is someone who is informed about the outcome (one-way communication).

There should always be exactly one (no more, no less) Accountable person per task/decision. In some cases the same person can hold both the Responsible and Accountable role types, but otherwise no person should be in more than one of these.

A matrix diagram can be created where the rows are tasks, the columns are roles, and in each entry is one of the letters RACI, or a blank space.

Some models add an additional role type indicating someone else who can also veto the result (in addition to the Accountable person), and/or an additional role type indicating someone who supports the work on the task, helping the Responsibles.

[17]

Roles vs. people

todo

Communications media

Phone

Email

IM

Microblogging

Wiki

todo

Compliance functions

todo the need for eg legal to report directly to CEO and for accounting to be centralized across divisions

Power struggles

todo

Misc institutional ideas

Apple's "top 100" meeting; not just the top-ranking individuals by title, but also a selection of the 'best' individual contributors (see http://fortune.com/2011/05/09/inside-apple/ )

startup case studies

todo

big company case studies

Apple, as a big company

An executive who has worked at Apple and Microsoft describes the differences this way: “Microsoft tries to find pockets of unrealized revenue and then figures out what to make. Apple is just the opposite: It thinks of great products, then sells them. Prototypes and demos always come before spreadsheets.” ... Specialization is the norm at Apple, and as a result, Apple employees aren’t exposed to functions outside their area of expertise. Jennifer Bailey, the executive who runs Apple‘s online store, for example, has no authority over the photographs on the site. Photographic images are handled companywide by Apple‘s graphic arts department. Apple‘s powerful retail chief, Ron Johnson, doesn’t control the inventory in his stores. Tim Cook, whose background is in supply-chain management, handles inventory across the company. (Johnson has plenty left to do, including site selection, in-store service, and store layout.)

Jobs sees such specialization as a process of having best-in-class employees in every role, and he has no patience for building managers for the sake of managing. “Steve would say the general manager structure is bullshit,” says Mike Janes, the former Apple executive. “It creates fiefdoms.” Instead, rising stars are invited to attend executive team meetings as guests to expose them to the decision-making process. It is the polar opposite of the General Electric–like notion of creating well-rounded executives. " -- http://fortune.com/2011/05/09/inside-apple/

Some Apple senior management, as of 2011, from https://fortunedotcom.files.wordpress.com/2011/05/insideapple-05.png?w=1024:

The CEO and their direct reports:

CEO
- COO
- SVP, mac hardware engineering
- SVP, operations
- SVP, general counsel
- SVP, CFO
- SVP, marketing
- SVP, IOS software
- SVP, retail
- SVP, industrial design
- VP, communications
- VP, HR
- VP, mobile advertising
- VP, mac software engineering
- VP, internet services
- VP, creative director

VPs who report to someone on the above list:

COO: