Where do the hours in a function actually go?
Most of the hours in an operational function go to coordination rather than to the tasks themselves. Deciding what matters this week, chasing the input that has not arrived, carrying a result from one system to another, remembering what was promised, and noticing that something has quietly gone wrong.
This is why time-and-motion reasoning about AI so often disappoints. A team lists its tasks, finds that AI can do several of them faster, and calculates a saving that never materialises — because the listed tasks were never where the time was going.
Why does task automation leave throughput unchanged?
Task automation leaves throughput unchanged because it removes execution time while leaving the initiation and sequencing with a person. If drafting an article takes two hours and a tool reduces it to twenty minutes, the function still publishes only as often as someone decides what to write, gathers the inputs, and starts.
The bottleneck simply moves upstream and becomes less visible. The team now has capacity to produce more and no more capacity to decide what to produce, which shows up as a backlog of half-started work rather than as a throughput gain.
What does function ownership change?
Function ownership changes who decides what happens next, which is the step that was actually scarce. An AI that owns a function maintains its own view of the state of the world in its remit, decides what matters this week, does the work, and brings the result back for a decision.
That is why the unit of value differs so much. A tool that writes an article faster produces articles when asked. A Digital FTE that owns search visibility decides which questions are worth answering, writes and fact-checks, notices when a published page starts to decay, and proposes a refresh before anyone notices the traffic drop.
What has to be true for ownership to be safe?
Ownership is safe when three things exist together: a written remit, measurable targets, and an approval gate on anything that leaves the company. Remove any one and the arrangement becomes uncomfortable for good reason.
Without a remit the system will wander into work nobody asked for. Without targets there is no way to tell productivity from activity. Without an approval gate a mistake reaches a customer instead of a review queue. Together they make a broad scope safe, which is the opposite of the intuition that broad scope requires caution and therefore narrow tasks.
- A written remit: what is in scope, what is out, and what must escalate
- Measurable targets: the numbers the function is accountable for
- An approval gate: nothing outward-facing happens without a human decision
- An audit trail: who approved what, when, and on what evidence
How do you measure whether ownership actually moved?
Measure function ownership by where the manager's time goes, not by how many tasks were completed. Before ownership moves, the manager spends their time producing and coordinating. After it moves, they spend it deciding: approving, sending back, adjusting targets, and granting autonomy.
That shift is observable in a calendar and in an approval queue, and it is a better leading indicator than output volume. Output can rise temporarily from a burst of enthusiasm; the time shift only happens when the sequencing genuinely moved.
Which functions should move first?
Move the function where coordination overhead is highest relative to judgement, because that is where ownership pays back fastest. Typically that means work that is continuous, evidence-driven, and currently bottlenecked on one person's attention.
Content and search visibility are common first moves because the cadence is relentless and the evidence is public. Reconciliation is another, because the ledger is unambiguous and the exceptions are where the hours go. Start with one function and one property, record the baseline before you begin, and resist the urge to move three at once — the comparison is what proves the case.
What does ownership look like in the manager's day?
Ownership looks like a five-minute morning read rather than a queue of tasks to assign. The Digital FTE that owns reconciliation has already matched yesterday's transactions, surfaced the exceptions with their documents, and written a morning brief in the first person: what is waiting on the manager, what happened yesterday against targets, what it detected, and what it spent. The manager approves the proposals, decides the exceptions, and moves on to work that needs a person.
That is the shape of a function with an owner: the work happens on its own cadence, the manager's involvement is decision rather than initiation, and the record of every decision is kept as it is made. The monthly review completes it. The FTE writes its own report against targets, stating misses and unmeasured items, and the manager decides three things: whether the targets are right, whether the remit needs changing, and whether any class of action has earned more autonomy or needs less. A function with an owner is one where those three decisions are the manager's whole job.