How to count the pairs on your team

Step one: fix the period

Pick a period and write it down before you look at anything. One quarter is the usual choice: long enough that a normal piece of work fits inside it, short enough that the answer describes the team as it is rather than as it was.

Fixing it first matters more than which length you choose. If the period is decided while you are counting, it will stretch to include the collaborations you remember and shrink around the gaps, and the number will be a record of your optimism.

Step two: write down the people

Everybody in the group you are measuring, in any order, as roles rather than names if the result is going to be shown to anybody. This site uses roles throughout for that reason: a grid with names on it is a document about individuals, and it will be read as one however it was meant.

Include part-time and shared members, and make a note next to them. A designer who is with you one day a week is on the list, and the expectation for their row should be different.

Step three: decide what counts, once

This is the only hard part, and the quality of the whole exercise depends on deciding it before you start rather than pair by pair.

The bar used here: one piece of work, inside the period, that needed both of them and would have come out differently if either had not been there. That includes a review that changed something, an afternoon spent on one problem, a document neither could have written alone, a handover that required a conversation.

It excludes being in the same meeting, being on the same thread, being in the same channel, being assigned to the same project without interacting, and liking each other.

You may set a different bar. A high bar — a shared deliverable with both names on it — gives a smaller, harder number. A low bar — any substantive exchange about work — gives a larger, softer one. Either is fine. What is not fine is a bar that moves while you are counting, and it will move unless it is written down.

Step four: go through the pairs

Not the people: the pairs. For eight people that is twenty-eight questions, and asking them in order is what stops you from missing the quiet ones. Going person by person produces a list of each person's most memorable collaborations and silently drops the rest.

Mark each pair as worked, met only, or neither. The middle category is worth keeping even though it does not count, because it tells you later which absent pairs are nearly pairs.

Step five: write it down with the date

The count is not interesting on its own. Six out of twenty-eight means nothing without something to compare it to, and the only comparison worth having is the same team three months later. The first count is the baseline and its value is entirely in the future.

The two ways it goes wrong

• It is done from memory by one person, who is usually the manager, and therefore the person present at every pair that includes them. A manager counting alone will reliably overcount their own rows and undercount the rest, because the pairs that did not involve them are exactly the ones they did not see. The fix is to ask: send each person their own row and let them correct it.
• It becomes a performance measure. The moment the count is reported upward as a number to improve, people will produce pairs to report rather than pairs that were needed, and within two quarters the grid will be full and mean nothing. The count is only informative while it is a description. There is a whole note later in this part about keeping it that way.

What to do with the middle category

Rows marked "met but never worked" do not count toward anything, and they are the most useful thing the exercise produces after the empty rows.

They are a ranked list of the cheapest improvements available. Each one is a pair where both people already know the other exists and is reasonable, and the only missing ingredient is a reason. Converting one costs a single small task; converting a pair who have never met costs the task plus everything that has to happen before anybody will propose it.

So when the count is finished and somebody asks what to do first, the answer is usually written on the grid already.

What the panel above shows

The grid is the team counted here, with the line that was discarded while making it. It was discarded by the sign rather than by me: the same role had been written twice, once from a conversation and once from a calendar, and the second copy carried a pair the first did not. Had it been drawn, the count would have read seven working pairs instead of six, and nobody would have had any way to tell which number was right.