Skip to content
Guide

The Employee Productivity Guide: Measuring Task Work Without Distorting It

For task-based work, the only measurements worth trusting are ones derived from events the system already records — when work was raised, accepted, started and completed. Everything else is either a proxy for activity, which rewards the wrong behaviour, or a score, which hides its own formula. The most useful thing most managers discover is that the largest slice of elapsed time is queueing, and no amount of pressure on people shortens a queue.

In this article

Start by ruling out the measurements that mislead

Nearly every productivity metric that damages a team has the same flaw: it measures activity, which people can produce on demand, rather than outcomes, which they cannot.

  • Task count. Rewards whoever splits their work into the most items. Within a month you will have more tasks and the same output.
  • Hours logged. Measures presence and the diligence of timesheet completion, which are not correlated with either effort or results.
  • Message and activity volume. Measures how much someone communicates about work, which is often inversely related to doing it.
  • Composite scores. A single number with a hidden formula cannot be argued with, cannot be learned from, and will be gamed as soon as its inputs are guessed.
  • Anything a person can inflate in five minutes. If they can, eventually they will, and you will have taught them to.

The real cost

The damage from a bad metric is not that it is inaccurate. It is that people start optimising for it, and the behaviour it produces is usually worse than the behaviour it replaced.

Four measurements that come from real events

If work is recorded as tasks, four durations already exist in the data without anyone logging anything. They are subtractions between two timestamps, and each answers a different question.

The four durations, and what each one tells you
MeasurementCalculationWhat it tells you
Time to acceptAccepted − createdHow long work waits before anyone takes it on. Usually a routing problem.
Time to startStarted − createdWhen work actually begins. Close to acceptance when accepting starts the task.
Completion timeCompleted − startedHow long the work takes once someone is on it. The closest thing to effort.
TurnaroundCompleted − createdWhat the requester experienced. The number a customer would recognise.

The pair worth looking at first is completion time against turnaround. A team with a 90-minute completion time and a six-hour turnaround does not have a speed problem — it has four and a half hours of queue. Pushing people harder cannot recover time nobody was working.

These are exactly the four figures TaskIt computes, from stored task timestamps, with nothing estimated or scored. A duration is only averaged when both ends exist, so an incomplete record shrinks the sample rather than counting as zero — which matters more than it sounds, because the alternative flatters whoever has the messiest data.

The counts worth keeping, and the one rate

Alongside durations, a small number of counts describe a person’s or a team’s situation honestly. Completion rate is the only ratio worth watching, and it needs a floor.

  • Assigned, accepted, started, completed. Four counts that together show where work is accumulating.
  • Pending and in progress. What someone is currently holding. This is the number to look at before adding to it.
  • Overdue and blocked. Blocked is the more useful of the two: it is a deliberate statement with a reason, not just a date that passed.
  • Completion rate. Completed against assigned. Only meaningful above a minimum sample — someone who finished their only task should not outrank someone who finished eighteen of twenty.

Be careful with counts across people. They are only comparable when tasks are roughly comparable in size, which they usually are not. A count is a description of a person’s week, not a league table position.

Know who the number is actually about

Every per-person figure depends on an attribution rule, and every attribution rule has a blind spot. Knowing which one you have is the difference between a fair conversation and an unfair one.

The common rule — and the one TaskIt uses — is to attribute a task to its current assignee. That has one very good property: because a task has at most one current assignee, per-person counts summed across the company can never exceed the company total. Reports that break that invariant are the ones that eventually produce a figure nobody believes.

The cost is a blind spot. Someone who did 80% of a job and then handed it over earns nothing in the per-person figures, and someone who joined a collaborative task at the end earns the completion. That is not a flaw to be argued away — it is a limitation to be compensated for by reading handover history and collaboration reports alongside the numbers.

A practical rule

Never open a performance conversation with a per-person average without first checking whether that person’s work changes hands a lot. If it does, the average is describing the queue, not them.

Separate queueing from effort before you do anything else

Most apparent productivity problems in small and mid-sized teams are queueing problems, and they have completely different fixes.

  1. Measure time to accept. If it is hours, work is waiting to be picked up. Nobody is being slow; the work is not reaching anyone who is free.
  2. Compare completion time with turnaround. A large gap is queue. A small gap with a long completion time is genuinely the work itself.
  3. Check the workload distribution. One person holding eleven tasks while two hold three explains most overdue lists without any reference to effort.
  4. Count blocked tasks and read the reasons. Repeated external blockers — waiting for a customer, waiting for approval, material not available — are process problems wearing a performance costume.

The fixes for queueing are structural: route work to a department queue rather than to a named person, notify the channel people actually read, and rebalance a queue you can now see. None of them require a conversation about anyone’s performance.

How to use the figures with a person

The purpose of a measurement is to find out what is happening, not to prove a point. If the person cannot see the records behind the number, do not use the number.

Show the working

A figure someone can trace back to specific tasks invites a correction; a score cannot be corrected, only resented. "Your average turnaround is eleven days" is an accusation. "These four tasks account for most of it — two were blocked waiting on the supplier" is a conversation.

Ask about the outliers, not the average

Averages hide the interesting cases. The three slowest tasks in a month usually contain the actual finding, and it is usually structural — an approval that took a week, a customer who went quiet, a dependency nobody owned.

Never measure what you have not explained

If people do not know that acceptance time is being looked at, some of them will not accept tasks promptly and will be measured badly for a behaviour nobody asked them to change. Say what you are looking at, and why, before you look at it.

Do not automate the judgement

A ranking is a starting point for a question, not an output. The person at the bottom of a completion-rate ranking is sometimes the one being handed everything difficult, and no formula will tell you that.

Measurement is not monitoring

There is a clear line between recording what happened to the work and recording what a person did at their desk. Crossing it costs more trust than any metric returns.

Screen capture, keystroke logging, idle-time detection and location tracking all measure presence rather than outcomes, and they change the relationship between a manager and a team in ways that are difficult to reverse. They also tend to produce their own distortions — people learn to look busy, which is a skill of no value to anyone.

Task-derived measurement is different in kind: the events being recorded are things people did to the work — accepted it, updated it, handed it over, finished it — and they are recorded because the work needed recording anyway. TaskIt deliberately has no timer, timesheet, effort log or activity monitor of any kind, which is a design choice rather than a gap.

A monthly rhythm that fits in half an hour

Productivity work fails when it is an initiative. It works when it is a short, repeated look at four things.

  1. Overdue and blocked. Start with what is currently wrong, and read the blocked reasons rather than counting them.
  2. Workload. Who is holding what. Rebalance before adding anything new.
  3. Time to accept, per department. The single most actionable figure most teams have, and the one that responds fastest to a structural change.
  4. The three slowest tasks of the month. Not to attribute them — to find out what they had in common.

Make it arrive

A scheduled report that turns up weekly is worth more than a dashboard nobody opens. The rhythm is the point; the figures are just what it looks at.

Frequently asked questions

There is no single one. The useful set is four durations derived from recorded events — time to accept, time to start, completion time and turnaround — read alongside counts of pending, overdue and blocked work. Completion time is the closest thing to effort; turnaround is what the requester experienced.

Completion time measures from when work started to when it finished. Turnaround measures from when it was raised. The gap between them is queue — time nobody was working on it — and it is usually the largest and most fixable part of elapsed time.

Only when the tasks are genuinely comparable in size, which is rarely true. A count describes a person’s week; it is not a league table. Comparisons are safer within one repeated kind of work than across a mixed workload.

No, and the two are different in kind. Task-derived measurement records what happened to the work — accepted, progressed, handed over, completed — using events that were recorded anyway. Screen capture, keystroke logging and idle detection measure presence, and tend to teach people to look busy.

A few weeks of real work at minimum, and more for anything you intend to say about an individual. Averages that only include tasks where both timestamps exist are honest about small samples, but a small sample is still a small sample.

Get figures you can show people

A month of real work is enough for the four durations to tell you whether your problem is queue or effort.