Every team measures how well it decides. Almost none measure how fast. And the second number is now the one that matters, because the part of the loop that used to be slow — the machine executing the decision — isn't slow anymore.
Performance Max reallocates budget in hours. AI Max rewrites match logic in real time. The optimizer moved to machine speed and left your organization exactly where it was: meetings, approvals, a weekly reporting cadence, a budget shift that waits four days for sign-off. Decision latency — the time between when a signal becomes available and when your organization commits to acting on it — is the gap between how fast your team decides and how fast the system underneath it now executes. It is the rate-of-learning gap, expressed as a number you can put on a wall.
The objection arrives immediately, and it's the right one: speed without accuracy is just faster mistakes. Correct. This is not an argument for deciding faster than you can decide well. Decision quality is table stakes — if your decisions are wrong, compressing the time to make them only gets you to the wrong answer sooner. Latency is the second-order metric. It's what separates two teams who already decide well. Once decision quality is good, speed is the only lever left — and it's the one nobody on the org chart is paid to pull.
This isn't a productivity essay. Productivity is about doing more things. This is about the structural cost of the time between knowing and acting, in a channel that now punishes that time more than it ever has.
The metric nobody owns
Walk the org chart. The marketing team is paid to ship campaigns. The finance team tracks cost. The agency is measured on ROAS. The analytics lead is measured on accuracy. Nobody — not one role — is paid to compress the time between "we should change this" and "the change is live." That gap is unowned, and unowned things default to slow.
This is why decision latency is the hidden multiplier on Cost Per Decision. Cost Per Decision establishes that every decision an organization makes carries a cost — the meetings, the analysis, the approvals, the execution. Latency is what that cost gets multiplied by. A decision that is correct but arrives two weeks late is not a correct decision. It's a stale one — made against a signal the market, and the auction, have already moved past. The cost isn't the decision. The cost is the wait.
And in a continuously-optimizing channel, the wait compounds. Every day a budget shift sits in committee is a day Performance Max bids against the old answer. Every week a creative kill waits for approval is a week of stale audience data feeding the model. The algorithm doesn't pause while you deliberate. It keeps spending — confidently, at machine speed — against the decision you already know is wrong but haven't yet been allowed to change.
Three places latency hides
Latency feels like one thing — "we're slow" — but it's three distinct failures, and they have different owners and different fixes. Name which one is yours before you try to compress it.
1. Approval latency. The decision is made. Everyone in the room agrees. It then waits — for a sign-off, a legal review, a budget holder who's traveling, a standing meeting that's three days out. The thinking is done; the authority to act is gated. This is the most common form in enterprises and the most embarrassing, because nothing is actually being decided during the wait. It's pure queue time. The fix is structural: pre-delegated authority below a dollar threshold, so a team can act on a clear signal without escalating.
2. Cadence latency. The signal exists, but your reporting rhythm only surfaces it on a schedule. If the data is clear on Tuesday but the weekly performance review is the following Monday, you've built six days of latency into the calendar before a human even sees the signal. Monthly business reviews are worse — they can bury a four-week-old signal under the ceremony of reviewing it. The fix isn't more meetings; it's moving the trigger from the calendar to the signal — alerting on the condition, not the date.
3. Ownership latency. No one is structurally accountable for the gap, so it expands to fill whatever slack the org allows. This is the deepest form, because it's not a process problem you can patch — it's an org-design problem. Until someone's job is to lower decision latency, the other two forms regenerate faster than you can fix them. The fix is the hardest and the highest-leverage: make latency a named, owned, measured metric, the way uptime is owned in engineering.
The reason this matters more now than it did three years ago: AI compressed the execution side of every one of these loops to near zero, which means the human side is now the entire latency. When the machine took a week to retrain and you took a week to decide, your latency was hidden inside the system's. Now the system answers in hours and yours is the only delay left. The bottleneck didn't disappear. It relocated — from the platform to your org chart. That's the same lesson as architecture beating platform: the durable constraint was never the tool. It was how you're built to use it.
"The bottleneck didn't disappear. It relocated — from the platform to your org chart."
What the number actually says
Latency isn't an abstraction; it's arithmetic. Three inputs: how many decisions your org makes a week, how many days the average one takes from signal-clear to live, and what a day of delay costs you — wasted spend, missed opportunity, or downstream rework. Multiply them and you get the figure that belongs on the wall: the annual cost of the gap between how fast you decide and how fast the system executes. Most enterprise marketing orgs run 80–250 decisions a week at 5–15 days each. The annual number that falls out of that is rarely small, and it is never zero. (I built a decision-latency calculator to run your own three numbers, and a set of benchmarks so you can see where your sector and size actually sit.)
A word on those benchmarks, because honesty is part of the credibility here: the current bands — mid-market decisions resolving in roughly 5–9 days, enterprise in 12–21, the slowest regulated approval chains past 30 — are a directional v1 synthesis from public reporting on marketing approval cycles, not yet primary Uncommon Move data. Treat them as order-of-magnitude, not gospel. Primary data collection is underway. If you want to see your own number against a real distribution rather than a synthesized one, that's the dataset being built — and the point of publishing the directional version now is to be honest about what's measured and what isn't.
What the number is not is a measure of money you're losing today. It's the size of a structural gap. If your marketing team takes more than 48 hours to reallocate budget against a fresh signal, you are not running an AI-driven marketing organization. You're running a legacy one with AI tools bolted on. The dollar figure is just that sentence, priced.
Why this is a P&L problem, not an ops problem
It would be easy to file decision latency under "operational efficiency" and hand it to a project manager. That's a mistake, and it's the mistake that keeps the metric unowned. Latency isn't an ops detail. It's a margin lever, and it sits directly on the translation chain from a marketing decision to enterprise value.
Here's the connection. Every day of latency is a day of unspent learning. The faster an organization closes the loop between signal and action, the more times it gets to learn per quarter — and rate of learning is the one competitive advantage in this discipline that compounds rather than depreciates. A team that decides in three days runs through roughly twice as many learning cycles per year as a team that decides in six. Two years of that gap and the faster team isn't twice as good; it's structurally ahead in a way the slower team can no longer catch, because it's making better decisions and more of them, against fresher signal, with two years more accumulated learning. Rate-of-learning gaps in this discipline have historically been the gaps that do not close.
That's why this is the number a CFO should want on the wall next to Cost Per Decision. One tells you what a decision costs to make. The other tells you what it costs to wait. Together they describe the actual economics of how your marketing organization thinks — and in a market where the machine has already gone to machine speed, the organization's thinking speed is the last variable left under your control.
Lower your decision latency and nothing on the dashboard changes next week. The campaigns look the same. The ROAS looks the same. What changes is the slope: more learning cycles per quarter, fresher signal behind every bid, a compounding lead that doesn't show up as a better number this month but shows up as a different company in two years. The optimizer already moved to machine speed. The only question left on the table is whether your organization is fast enough to keep deciding what it should do — before the answer it's executing is already stale.