The Product Manager’s Atlas
Product Execution

RACI and Decision Rights

4 min readΒ·871 words
product-executionracidecision-rightsgovernancedisagree-and-commit

Most of what gets called an alignment problem is really a decision-rights problem wearing a disguise. The team is not misaligned on the goal. They are unclear on who gets to make the call. So the decision gets made by everyone, which is slow and political, or by no one, which is drift. This note is about fixing that, and it is the companion to the role split in The 10-50-90 Execution Framework.

#RACI in one table

The classic model assigns four roles to every meaningful decision or deliverable.

RoleMeaningThe trap
ResponsibleDoes the workFew traps here; the work just needs an owner
AccountableOwns the outcome; the single deciderMore than one "A" = no real owner
ConsultedInput is sought before the callConsulting too many people = paralysis
InformedTold after the callConfusing this with Consulted causes most friction
β–²There is exactly one Accountable

The most common RACI failure is two people who both think they are accountable, or a committee that is collectively accountable, which means no one is. If you cannot name the single person who owns a decision, you do not have decision rights. You have a future argument scheduled for the worst possible moment.

#The model I actually use: Owner / Partner / Consulted

Full RACI can get bureaucratic. For day-to-day product work I run a lighter three-role split, the same one baked into The 10-50-90 Execution Framework:

  • Owner. Accountable for the decision and its outcome. Makes the final call.
  • Partner. Co-creates the work and owns the decisions in their domain. Not a passenger, but a peer with real authority in their lane.
  • Consulted. Their input must be genuinely weighed, but they do not decide.

The reason this beats generic RACI for a product team is the domain split it implies, straight from how I run delivery:

Decision domainOwner
Product success & business viabilityPM
Usability & the user experienceDesign (User Experience Design)
Technical feasibility & implementationEngineering

A PM who tries to own usability is overstepping; an engineer who tries to own scope-vs-value is overstepping. Healthy teams know whose call is whose. This is the PM/Design/Eng Venn expressed as decision rights.

#Disagree and commit

✦The most useful two words in product

You will lose decisions, including ones you owned the input on. The mark of a professional is disagree and commit: argue hard before the call, then once it is made, execute it as if it were your own. Re-litigating a settled decision in the hallway is corrosive. It signals to the team that decisions are not real, which makes every future decision slower.

Disagree-and-commit only works if the disagreement was genuinely heard first. If you commit without ever having voiced the disagreement, that is not commitment. It is silence, and silence in the room becomes sabotage in the work.

#Escalation: a tool, not a failure

New PMs treat escalation as admitting defeat. I coach the opposite. Fast, clean escalation is a skill, and the APM-level Product Delivery bar explicitly rewards knowing when to escalate. The rule I use: escalate when a decision is blocked, above your pay grade, or cross-team and stuck, and escalate with a recommendation, not just a problem. "We're stuck between A and B, here's the trade-off, I recommend A, I need you to break the tie" is leadership. "Help, we can't agree" is not.

For how much senior leadership to pull in, I lean on Curtis Stanier's The Initiative Pyramid: a few high-risk initiatives warrant leaders being consulted; many low-risk ones only warrant them being informed. Match the escalation weight to the risk.

β„ΉDecision rights live in the spec

The cleanest place to record who owns what is the spec itself. A one-line "Decisions & owners" section pre-empts most mid-project ownership fights, because you settled them when everyone was calm instead of when the deadline was on fire.

#How AI is changing it

AI does not change who should decide. Accountability is a human and organizational property, not a technical one. What it changes is the inputs. With analysis and first-draft options generated in minutes, the bottleneck shifts decisively from gathering information to exercising judgment on it, and Reforge's framing is that AI multiplies the burden of choice rather than reducing it. In agentic products this gets sharper. Deciding the human-in-the-loop checkpoints, meaning which calls a system may make on its own and which require a person, is itself a decision-rights problem you have to design. See Designing for Trust and Probabilistic Systems and Competencies AI Commoditizes vs Elevates.

#Continue Reading