RACI Matrix Template Explained

RACI is taught as four definitions to memorise, which is why most matrices are filled in politely and then ignored. The letters aren't descriptions of involvement. Each one is a different promise, and only one of them can stop you.

A regional healthcare provider built a RACI for its platform migration: nine activities, seven people. It took an hour, everyone agreed, and it went on the wiki. (Composite example: a regional healthcare provider drawn from patterns across several organisations, not a real one.)

Counted afterwards, it contained forty-one C's — an average of four and a half people to be consulted per activity.

Nobody was consulted. Not once, on any activity, in five months. And nothing went wrong as a result, which is the part worth sitting with: forty-one commitments were made, none were kept, and the project was fine. The C's weren't a plan. They were a way of not leaving anyone off a list.

A cell is a promise, and the four letters promise different things

Read as labels, the letters describe how involved someone is, on a scale from very to barely. Read that way they're almost interchangeable, which is exactly how they get assigned.

Read as obligations, they're not on a scale at all:

Letter	What that person owes	What they can do	Whose time it costs
R — Responsible	Does the work	Everything	Their own
A — Accountable	Answers for the outcome, right or wrong	Stop it	Their own
C — Consulted	Gives input before the decision	Change the outcome	Theirs, and yours — it's a wait
I — Informed	Nothing. Reads the result	Nothing	A moment

The two middle rows are the ones that carry weight. A can stop the work. C can change it. R and I can't do either — R executes a decision, I learns about one.

That reframing decides the arguments. "Should Security be C or I on the migration design?" isn't a question about how important Security is. It's: if Security objects to the design, does the design change? If yes, C. If no, I — and writing C is a promise you've already decided not to keep.

The test you can run: take any C in your matrix and ask what happens if that person says no. If the honest answer is "we'd proceed anyway", it's an I. Change the letter.

The letter everyone fills in wrong is C

R gets assigned correctly because someone has to do the work. A gets argued about, which means it gets attention. I is free. C is the one that gets over-assigned, because it's the polite letter — it signals respect, it costs the person filling in the matrix nothing, and it feels rude to demote someone to I in a room where they're sitting.

But C is the only letter that spends somebody else's calendar. Every C is a blocking dependency: the work waits until that person has looked. Four C's on an activity means four opportunities for it to sit for three days.

So the matrix drifts to one of two failure states, and the provider had the first:

• Nominal C's. Everyone is C, nobody is consulted, the matrix is decorative. It's harmless right up until the one activity where consultation actually mattered, and it doesn't happen either — because the habit of skipping C's is already established.

• Honoured C's. Everyone is C and everyone is consulted, so every activity carries four sequential waits. This is the version people mean when they say RACI is bureaucracy, and they're right.

The correction is a budget, not a rule: two C's per activity, and a third needs a reason said out loud. The provider's rebuilt matrix has fourteen C's across the same nine activities, and all fourteen have been used.

The test: count the C's in your matrix and divide by the number of activities. Above about two, you're either not honouring them or you're serialising the project.

Exactly one A, and the A must be able to stop it

The single-A rule is the one everybody knows and it usually survives — but it survives in letter, not in substance. What breaks is the second half: the person marked A must have the standing to halt the work.

An A who would be overruled if they said no isn't accountable. They're a name in a box for when it goes wrong, which is the thing accountability is supposed to prevent. It's the same failure as approval that ratifies rather than reviews — a signature that routes no information and moves blame.

Two practical corollaries:

• A and R are often the same person, and that's healthy. Mark it A,R. Splitting them by default creates a hand-off that exists only on paper.

• A should sit as close to the work as the risk allows. A senior sponsor accountable for every activity is a matrix with one real A and eight decorations. Delegating it downward is the same move that makes change authority work.

The test: for each A, ask whether that person could stop the activity tomorrow and have it actually stop. Any A where the answer is no is a placeholder.

The rebuilt matrix, and what changed in it

The provider's second version, after re-reading every cell as an obligation:

Activity	Eng Lead	PM	SRE	Security	Support	Sponsor
Define scope	C	A,R	I	I	I	C
Design approach	A,R	C	C	I	I	I
Build migration tools	A,R	I	I	I	I	I
Security review	C	I	I	A,R	I	I
Customer comms	I	A	I	I	R	I
Execute migration	A	I	R	I	I	I
Verify after	A,R	I	C	I	C	I
Go / no-go	C	C	C	C	I	A,R

Three things moved from the first version:

• Security went from C to I on design and build, and from I to A,R on its own review. It doesn't shape the approach; it can block the release. That's a stronger position than being one of five consultees, and it's the honest one.

• Support went from C to R on customer comms. They were writing the messages anyway. Marking them C described the meeting, not the work.

• Sponsor went from A on everything to A on the go/no-go only. The one decision they can genuinely make and stop.

Twenty-seven cells changed letter and not one person's actual involvement changed. The first matrix had been describing the org chart.

The test: for each row, ask who would be surprised if they weren't asked. That's your real C list, and it's usually shorter than the written one.

When a RACI is the wrong instrument

It's built for the case where responsibility genuinely spans groups. Outside that, it costs more than it routes:

• One team, clear ownership. A list of activities with one name each does the whole job. The C and I columns exist to manage people outside the room; if there's nobody outside the room, they're empty by definition.

• Ongoing operational work. RACI describes a project's shape at a moment. For steady-state, what you want is a service owner and an on-call rota, both of which answer "who" continuously rather than once.

• A single decision rather than a body of work. A one-off call with several stakeholders wants a decision record — who decided, who was consulted, what was rejected — not a matrix with one row.

• The team already has working agreements that hold. If people can already answer who's accountable for this without looking anything up, the matrix would only write down what's already true.

That last one is also the success condition. A matrix that's working stops being consulted, because the conversation that produced it did the job. A matrix that's ignored and whose questions can't be answered from memory is the one that failed.

The test: ask three people on the project who's accountable for the riskiest activity. Three matching answers without anyone opening the wiki means it worked. Three different answers means you have a document, not an agreement.

What to change this week

Don't rebuild the matrix. Audit one column: C.

Go cell by cell and ask what happens if that person says no. Demote every C that wouldn't change anything to I. Expect to remove most of them — the provider removed twenty-seven of forty-one — and expect the remaining ones to start getting honoured, because a short list is a list people believe.

Then check the A column against the stop test. An A who can't halt the work needs to be replaced by someone who can, or the activity needs to be split until somebody can.

The matrix isn't the deliverable. The moment somebody says "wait, I thought you were doing that" — during the hour it takes to build, rather than during the migration — is the deliverable. Every cell you argue about is one you won't discover later. The communication plan inherits the result directly: its owner column is your A and R columns, already agreed.