Governance is not a steering committee that listens to a pack and says 'keep going.' It is the set of decision rights and gates that decide who can spend, who can change scope, and who can stop the work — before the spend is already gone.
Decision rights, written down
Start with a short list of decisions that actually move money or date: start a project, change scope or budget beyond a threshold, accept a residual risk, slip a committed date, or close and hand over. For each, name the person (not the forum) who decides, who advises, and who must be informed. Forums that 'own' decisions own nothing.
Thresholds stop every change landing on the same desk. A PM can approve a two-day slip inside contingency. A sponsor approves a scope cut. The portfolio board approves a new workstream that eats someone else's capacity. If every decision escalates, you do not have governance. You have a queue.
Publish the rights where people will trip over them — the charter template, the change form, the first page of the steerco pack. A RACI buried in a 40-page framework is a rumour. People will invent their own rights by Friday.
Gates that can actually say no
A gate is a scheduled decision with evidence: is this still the right work, with a plan that fits the next phase, and an owner who can run it? Typical gates sit at idea, business case, delivery start, go-live, and close. The names matter less than the question: what would make us stop?
Write entry criteria in verbs and artefacts, not vibes. 'Risks owned, supplier contracted, operational owner named, benefits measure agreed.' If the gate can be passed with a confident presentation and no evidence, it is a ceremony. Ceremonies do not protect the portfolio.
Allow a conditional pass. 'Proceed, but dual-source the vendor by 1 May or we stop.' A binary yes/no with no conditions teaches people to hide problems until after the meeting. Conditions keep the gate honest without pretending every project is either perfect or dead.
Assurance, not surveillance
Assurance asks whether the project is telling the truth and whether the method fits the risk. It is a sample, a review, or a health check — not a second status meeting that re-asks every metric. If the PMO re-does the PM's job, you have hired two people to distrust each other.
Scale assurance to risk. A six-week internal tool does not need a three-line-of-defence review. A customer-facing release with a regulatory date does. Publishing one assurance path for everything is how small projects drown and large ones slip through on charm.
Record gate decisions in one place the next person can find. When a sponsor asks why you started a project that later failed, 'we discussed it' is not governance. A dated decision, the evidence, and the conditions are.
Summary & next steps
Governance is decision rights plus gates with evidence. Name the person, set thresholds, write entry criteria, and keep assurance proportional to risk.
Next: portfolio management — using those rights to select the right work, not merely to supervise whatever already got funded in a corridor.