How should we measure the time an AI-assisted task takes?

Measure both active effort and elapsed time against an agreed finished result. Add the effort of everyone who finds, prepares, reviews and corrects the work. Separately record how long completion takes, including waiting. Include setup and ongoing support when assessing whether the approach is worthwhile.

What is the difference between handling time and elapsed time?

Handling time is the total time people actively spend on the task. Elapsed time runs from its agreed start to its accepted finish, including waiting. When colleagues work in parallel, add their active effort but count that period only once when measuring elapsed time.

How can we compare AI-assisted work fairly with our current process?

Use the same quality standard and comparable cases for each approach. Record differences in complexity, staff experience, corrections and abandoned attempts. Include a simpler alternative where one exists. Repeating an identical task can favour the second attempt, so account for what people have already learned.

How do we decide what a useful finished result looks like?

Name the recipient, intended use and conditions for accepting the work before drafting. A.I.S.I.R. is the five-stage framework at the centre of Thought Architecture: Awareness, Intent, Structure, Iteration, Reflection. That sequence helps frame the outcome before prompting and supports reflection on whether faster work was genuinely useful.

Where the time goes after AI writes the first draft

A draft can be quick to write and slow to get ready for use.

The colleague who checks it, the person who finds missing information and the person who chases a decision each hold part of the answer to one question: how much time did everyone spend getting this work finished?

Putting those accounts together gives leaders something useful to work with. It shows where effort has reduced, where it has moved and where the job still gets stuck.

Start with an agreed finish

Before comparing approaches, agree what counts as a completed piece of work.

For a hypothetical recurring summary, completion might mean that its facts have been checked, its important qualifications are clear and the person receiving it accepts it for the intended use. A polished draft waiting for those checks is still work in progress.

Keep that finish line consistent. If the AI-assisted version receives less scrutiny or leaves someone else to fill in the gaps, its completion time is being measured against a different standard.

Look across several pieces of work as well. A quick result on a straightforward example says little about how dependable the approach will be when information is incomplete or sources disagree. Consistent quality includes what happens on those difficult occasions.

Record corrections and returned work, including problems noticed after something was initially accepted. Include abandoned attempts that sent someone back to the usual process. These help explain the effort required to produce a result people can rely on.

Keep effort and waiting separate

There are two useful views of time.

Handling time is the time people actively spend on the work. Add it across everyone involved. That includes finding source material, preparing prompts, reviewing the output, making corrections and coordinating with colleagues. Count another person’s help even when it arrives as a quick conversation rather than a formal review.

Elapsed time runs from the agreed start to the accepted result. It includes the periods when nobody is working on the job because it is waiting for information, approval or attention. When people work in parallel, add their effort but count that elapsed period only once.

Those views can move in different directions. An approach may reduce active effort while the finished work still waits in the same approval queue. Another may reach completion sooner because a clearer hand-off removes a delay, even though the total effort changes little.

Both outcomes can matter. Recording them separately makes the improvement easier to describe accurately. It also helps locate the next useful change.

There is work around the tool to include too. Someone sets it up, prepares reusable instructions, helps colleagues learn and maintains the materials as things change. Separate that initial setup and training from the effort on each job, while keeping ongoing support and maintenance visible. Otherwise, the person quietly keeping the approach working can disappear from the account.

A small comparison worth doing

Choose a recurring, manageable piece of work with a clear recipient and an observable finish. Sketch its steps with the people who do them. A simple shared record is enough.

For each example, note when it started and finished, roughly how much active time each person spent, where it waited and what needed correcting. Brief estimates made close to the work can be useful if everyone uses the same approach and labels them as estimates. Avoid asking people to reconstruct their effort long afterwards.

Explain to the team what is being recorded and why. Keep the record focused on improving the workflow, with only the detail needed for that purpose.

Compare the current process with the AI-assisted approach. Include a credible simpler alternative where one exists, such as a clearer template, better source material or a shorter hand-off. Give each option the same quality requirement and include the work needed to set it up.

Use a mix that resembles the work the team actually receives. Include ordinary cases and awkward ones. Keep track of differences in complexity and the experience of the people involved, so an easy example handled by an experienced colleague does not carry the whole conclusion.

If the same example is attempted again, allow for the advantage of already knowing its contents. Similar cases can offer a fairer comparison. Record anything else that could explain the change, including clearer instructions introduced during the test.

Review the record together once there is enough to see where the effort goes. Look at the range of experiences as well as the typical result. A method that occasionally creates a large repair job may need more attention than its usual performance suggests.

The record should help a conversation, with enough detail to explain the result. It can stay small.

Follow the time that is released

When the same accepted work takes less total effort, staff time has been released. What happens to it deserves a separate look.

Usable capacity depends on whether that time can be applied to something worthwhile. Small gaps scattered through a day may be difficult to use for work requiring sustained attention. People may need different skills, access or a change in priorities before they can take on another task.

Demand matters too. Faster preparation will have a limited effect on completed output if there is little additional work to do, or if every item still waits for a busy reviewer. Adding more drafts to that queue may increase the backlog.

Before describing a capacity gain, identify the work that could use the released time and check whether it actually gets done. The benefit might also be more room for preparation, better follow-through or a more manageable working day. Those outcomes deserve to be described on their own terms.

Cash savings require evidence that spending has reduced after the added costs of the approach are included. Released staff time alone does not establish that result.

A useful place to begin is with the people who pick up the work after the draft. Ask them to walk through a completed example with you. Listen for the checking, chasing and small repairs that rarely appear in a quick account of time saved.

That is often where the next improvement becomes easier to see.

Where the time goes after AI writes the first draft

Start with an agreed finish

Keep effort and waiting separate

A small comparison worth doing

Follow the time that is released

Measuring the work around AI.

Bring practical AI thinking into your organisation.