RACI and Decision Rights for Startups
RACI has a deserved reputation as corporate theatre — a matrix nobody reads, produced by a process nobody enjoyed, describing a company that no longer exists. That reputation is why most startups have no decision rights at all, which is considerably worse. Undefined decision ownership is the single largest source of friction in a growing company: it produces slow decisions, recurring escalations, and the specific category of conflict where two capable people argue about who had the right to decide rather than about what to do. This page is the lightest version that actually works.
What’s inside
Why RACI earned its reputation, and what survives
The four roles that matter
Building the decision inventory
The one-page format
Keeping it live rather than archaeology
The founder’s part
Where Blomma fits
Why RACI earned its reputation, and what survives
The classic version assigns four roles — responsible, accountable, consulted, informed — to every task in a project. Two things go wrong.
It is applied to tasks rather than decisions. Mapping who is responsible for each activity produces an enormous document, most of which is obvious, and it does not answer the question that actually causes friction: when we disagree, who decides?
And the distinction between responsible and accountable is genuinely confusing. Most people cannot reliably say which is which, which means the artefact requires a glossary, which means it does not get used.
What survives is the underlying insight, which is sound: for anything contested, someone should decide, some people should be asked first, and others should be told after. Strip it to that, apply it to decisions rather than tasks, and it becomes one of the highest-return documents a growing company can produce.
The test of whether you have done it right: someone can settle a live dispute by opening the page, in under thirty seconds, without asking what a word means.
The four roles that matter
For each decision, four things. Keep it to four — more and you have built a governance document.
Decides.One name. Not a committee, not “the leadership team,” not two names. If you cannot assign one name, that decision belongs in the escalation rule instead, deliberately and explicitly. This is where most attempts fail: founders write “product and engineering” and have documented the absence of a decision right.
Consulted.Who must be asked before, with enough notice to give a real view. This role does most of the relationship work in the document. Being consulted properly is what makes not deciding tolerable, and most of what people experience as being cut out is a missing consultation rather than a lost decision.
Informed.Who hears afterwards, and how quickly. Cheap to specify, and it prevents the particular injury of learning something significant from a third party.
Threshold.What size or type escalates. Spend under fifty thousand is one person’s call; above it goes up. Thresholds are what make single-name ownership acceptable, because they bound the downside.
That is the whole framework. No matrix, no glossary, no distinction between responsible and accountable.
Building the decision inventory
The substance is the list of decisions, and it should come from your own history rather than a template.
Two prompts produce most of it.What were the last ten things that caused friction?Those are your undefined decisions, identified by evidence.What are the ten significant decisions coming in the next six months?Those are where the next friction is coming from.
A representative set for a company of fifty to two hundred: pricing and discounting; roadmap priority for the quarter; build, buy, or partner; senior hires; firing a senior person; compensation bands and offers; spend above a threshold; accepting a customer requiring custom work; architecture and platform choices; tooling; brand and public positioning; legal or regulatory positions; and headcount allocation between teams.
Aim for fifteen to twenty-five decisions. Fewer and you have missed the contested ones; more and you are documenting tasks again.
Then add the two categories founders reliably omit.Decisions about people— who decides a promotion, who decides a role change, who decides someone should leave. Anddecisions about the decisions— who can change this document.
The one-page format
Literally one page, one line per decision. Something like:Pricing changes under 10% — Decides: VP Sales. Consulted: Product, Finance. Informed: exec team. Above 10%: CEO.
No preamble, no principles section, no explanation of the framework. Anyone should be able to find a decision and read its owner in seconds.
Three things to include beyond the table. Adefault rulefor anything not listed — usually “the closest named owner decides, and tells their lead.” Without this, every unlisted decision escalates and the document creates the problem it was meant to solve. Anescalation rulefor genuine deadlock. And areview date.
Then send it to everyone affected, not just the leaders. A document your organisation has not seen does not change behaviour, because people continue routing decisions to whoever they expect will give them the answer they want.
Keeping it live rather than archaeology
The difference between a useful decision-rights document and the RACI matrix in everyone’s memory is entirely about whether it gets used. Three habits.
Open it in the room.When a decision comes up and the routing is unclear, pull up the page during the meeting. That is not bureaucracy; it is the entire return on the exercise, and doing it visibly a few times teaches everyone the document is real.
Update it when it is wrong.Every time you hit a decision that is not covered or is assigned badly, add or fix the line that day. A document that gets amended is alive; one that gets recreated annually is theatre.
Review it at each stage change.Roles that fit at fifty people do not fit at a hundred and fifty. Twice a year, or after any reorg.
One anti-habit: do not expand it. The pressure to add more decisions, more roles, and more nuance is constant and it is how you end up with the thing you were trying to avoid. If it exceeds one page, something should come off.
The founder’s part
The document lives or dies on one behaviour, and it is yours.
The first time someone makes a call inside their rights that you think is wrong, you have a choice. Let it stand, and the document becomes real. Override it, and you have demonstrated that the actual rule is that you decide — and everyone updates accordingly, permanently. One override costs you the whole page.
That does not mean never intervening. It means the threshold for intervention should be written down in advance, while you are calm, and it should be something like “irreversible or seriously costly.” Almost always the thing in front of you is neither.
And if you do need to override, do it explicitly and explain the reasoning — “this one is going to bind us for three years, so I’m taking it, and here’s why” — rather than quietly re-deciding. An acknowledged exception preserves the rule. An unacknowledged one replaces it.
Where Blomma fits
The reason most companies do not have this document is not that the format is hard. It is that writing it forces you to answer questions you have been leaving ambiguous, several of which are uncomfortable: whether a particular executive actually holds the decisions their title implies, whether you are willing to release pricing, whether the CEO title carries the final call and what that means for your cofounder.
Blomma is an always-on AI career coach with no stake in your company. Use it to build the inventory from your real friction history rather than a generic list. Use it to check each line for the most common defect — a decision assigned to two names, which is no assignment at all. Use it to work out your genuine intervention threshold and write it down before you need it. And use it in the moment that decides whether any of this holds: the first time someone decides something inside their rights and you disagree. Where the situation warrants a human who has done this at your stage, bring one in..
Decision rights do not remove disagreement, and they are not supposed to. They remove the second argument — the one about who had the right to decide — which is the one that damages relationships rather than improving decisions. One page, fifteen lines, and most of the friction in a growing company goes away.