· Zenous Team · 4 min read
Decision Latency: The Delivery Metric That Should Replace Status Green
Programs do not slip because work is slow. They slip because decisions are slow. How to measure decision latency, what good looks like, and why it is the first number AI actually moves.
Ask a slipping program what went wrong and you will hear about vendors, scope, and integration surprises. Trace the timeline instead and a different pattern shows up: the vendor issue was known for six weeks before anyone decided what to do about it. The work was not slow. The deciding was.
Decision latency is the time from a question being raised to a defensible decision being made and recorded. It is the single number we would keep if a sponsor allowed only one, because it predicts slippage earlier than any RAG status and it cannot be gamed by optimistic reporting. A program can call itself green for months. It cannot hide forty-day-old open decisions.
How to measure it
You need three timestamps per decision, and most programs already have them scattered across minutes, logs, and tickets:
- Raised: the date the question first appears in a report, a risk log, or a steering pack.
- Decided: the date an accountable person commits to an option in writing.
- Recorded: the date the decision lands where the team can act on it.
Latency is raised-to-decided. The decided-to-recorded gap is a bonus signal: when it is large, the program has a communication problem on top of a governance one.
Start with the last ten decisions, not a new tracking system. An analyst with the steering packs can reconstruct them in an afternoon. Median them. That number is your baseline, and it is usually a surprise: healthy programs run 5 to 10 business days; programs in trouble run 30 plus, and the worst decisions never close at all. They just stop being asked.
Why it beats status reporting
Three properties make decision latency the better executive metric.
It is a leading indicator. Scope variance and budget burn tell you about damage already done. A rising decision backlog tells you about damage scheduled for next quarter. The six-week vendor decision becomes a three-month replan; you could see it at week two.
It measures the sponsor too. RAG status only ever grades the team. Decision latency grades the whole system, including the steering committee that let a decision age past its third meeting. In our PMO audits the longest latencies almost always sit above the program, not inside it. That finding changes behavior faster than any process recommendation, because it is the first time the governance layer has seen its own number.
It is cheap to keep honest. A decision either has a written owner, option, and date, or it does not. There is no amber.
What good looks like
Set bands the way you would for any control: median latency under 10 business days, no open decision older than 30, and a visible list of the top five oldest decisions in every steering pack. That last one does most of the work. Naming the oldest open decision, with its age in days and the name of the person it is waiting on, is the cheapest governance intervention we know. Nobody wants to be the name next to the number 47.
Two failure modes to design against. First, fake closure: decisions “made” verbally that reopen every month. That is why the recorded timestamp matters. Second, latency laundering: splitting one hard decision into five small ones that each close fast while the real question ages. The countermeasure is to track the original question, not the fragments.
Where AI actually helps
Decision latency is also the first delivery number that agentic AI moves in a way a CFO can verify. Not because agents make decisions. They should not. But most latency is not deliberation time; it is assembly time: the three weeks spent gathering the options, the impact data, and the one-pager before the decision maker ever sees the question. Agent workflows compress that assembly from weeks to days. The AI-native PMO piece shows the operating pattern; the effect on this metric is the point of the pattern.
A program that instruments decision latency before adopting agents gets a before-and-after number that survives audit scrutiny. A program that adopts agents without it gets a faster pipeline into the same stalled steering committee.
Start Monday
Reconstruct your last ten decisions and compute the median. Add the five-oldest-decisions table to the next steering pack. If the median is over 20 days, the fix is rarely more process; it is usually decision rights nobody has written down, and that is a governance conversation. A delivery audit maps it in 7-14 days, or book a consultation and bring the ten timestamps. They are usually enough to see the shape of the problem.
