RACI Matrix

The question that comes up most often in work spanning several teams is who actually decides this. Without an answer to it, a meeting can end without leaving any decision behind at all.

Pagedeck's pricing screen redesign went that way. Three meetings in, no design had been settled on. Noa would put up an option, Sofia would say it didn't match the terms she had promised in sales calls, Theo would raise the implementation cost, and Grace, passing through, would mention a different direction. All four of them had something worth saying and none of them was the final approver, so every meeting started the discussion over from the beginning.

The problem wasn't that there were many opinions but that the weight of each opinion was unrecorded. Nobody could tell whether Sofia's point was a hard constraint or a preference, or whether Grace's remark had been an instruction or just an idea. A RACI matrix is where that weight gets recorded in advance. Under decision making, the rules for setting who decides and how, it belongs to the role-definition group, the one that attaches labels to people.

One piece of work can be filled in within an hour, in five steps.

1. Break the work into rows of activities. One row reading "pricing screen redesign" settles nothing. Split it into approving the design, approving the copy, approving the start of development, and setting the launch date, and the answer differs from row to row: Mia decides the design and Grace decides the launch date.
2. Agree on what the four labels mean. Responsible is whoever does the work, accountable is whoever owns the outcome and gives final approval, consulted is whoever has to be asked before the decision, informed is whoever only needs to know the result. Without writing those four definitions down somewhere, somebody will treat consulted and informed as the same thing.
3. Fill each row in. There's one rule: accountable is exactly one person. Put two names there and the two look at each other while nobody decides. Responsible, by contrast, can be several people at once.
4. Confirm it with the people you labeled. Somebody who doesn't know they're in the consulted column doesn't offer an opinion, and then reopens the decision afterward. Filling the table in alone and then sharing the file guarantees this happens.
5. Open it at the moment things stall. This table isn't wall decoration; it's the document you open when a decision has stopped moving. When a meeting halts at the same point twice, find that row and check who is named accountable on it.

Not trying to fill in every cell is the technique that makes it usable. Twelve rows of activities against six people is seventy-two cells, and a table with all of them filled goes unread. Most of the cells should be empty, and a row carrying more than three labels is a signal that the activity can still be split further.

Two kinds of place fit: work spanning three or more teams, and repeating procedures such as incident response or launch checks. Repeating procedures pay the effort back most, since one drawing gets reused exactly as it is the next time.

The limits are clear.

1. Drawn once, it goes stale fast. People change roles and teams split, and then the names in the table are wrong. A wrong table is worse than no table at all, because people use it to go and ask the wrong person.
2. More consulted names make decisions slower. Five in the consulted column means meeting five people before deciding. It's tempting to put every stakeholder — anybody affected by the outcome or able to affect it — into that column, but only people whose own work actually changes belong there; the rest move down to informed.
3. It's too coarse an instrument for a single decision. This table divides how a piece of work gets run, so for something like a single budget approval, where the decision itself is the subject, the columns don't sit well on it.
4. A table cannot substitute for actual authority. Writing one name as accountable means nothing if that person isn't in a position to overrule the CEO. Something like disagree and commit, where objections are raised before the decision and everybody executes after it, has to exist alongside for the table to function.

"This row lists two accountable names. It has to come down to one before anything gets decided."

Split accountability means that nobody ends up deciding.

"Am I consulted or informed here? If I'm consulted, pull me in when the design exists."

Reading your own label to work out at what point you're involved.

"Let's move them down to informed. This decision doesn't change anything on their side."

Applying the rule for keeping the consulted column small.

At Pagedeck, Mia makes one of these tables per redesign. What she does after making it isn't sharing the file but confirming it directly with the four people named in it. Once Sofia knew she was in the consulted column, she started sending the sales terms over before any design existed, and the meeting that used to run three times ran once instead.

The biggest change from before the table wasn't meeting length but what happened afterward. Decisions simply stopped getting reopened after the fact.

The most common arrangement puts accountable and responsible on the same person, since it looks natural for whoever does the work to own the result. But joining them means the person doing the work is approving their own work, and the review step disappears. Split apart, the approver brings different information than the doer. With Noa producing the design and Mia approving it, Mia is also weighing this quarter's priorities and consistency with other screens. One person holding both labels loses that second view entirely.

The other frequent belief is that drawing the table organizes the collaboration, because attaching labels feels like the end of the confusion. But all this table records is who decides, never why they decided that way. With no record of the reasoning, the same item arriving three months later gets discussed from scratch. So a record of decisions gets kept alongside the table itself. Bikeshedding — hard items passing quickly while trivial ones absorb the time — isn't stopped by the table either; that belongs to a separate rule about how items get weighted.