Time to First Draft: The Workflow Metric That Shows AI Is Working

Time to first draft is the elapsed time from a request being ready to a usable draft reaching the reviewer. It is one of the clearest signs that AI is working, because systems record it automatically and it moves quickly. On its own it can mislead, so pair it with time to approved and the light-edit rate.

What is time to first draft?

Time to first draft measures how long a workflow takes to turn a complete request into a draft good enough to review. The clock starts when the inputs are ready and stops when the draft reaches the reviewer. It is measured per workflow, from system timestamps, and reported as a median.

The definition needs two firm edges. The start is not when someone first mentioned the job; it is when the brief, data or request was complete. Likewise, the end is not when AI produced text; it is when a draft fit to review reached the person who approves it. Both edges should be events a system records, such as a ticket moving column.

Two related measures sit beside it:

Why is time to first draft a useful AI metric?

It is frequent, recorded by systems and sensitive to change. A team producing weekly reports or proposals generates dozens of data points a month, so an improvement shows within weeks. Because timestamps do the measuring, it avoids the self-reporting problem that makes "hours saved" so unreliable.

Self-reports are weak evidence. In METR's 2025 trial, experienced developers believed AI had made them faster when it had made them slower. Timestamps do not have that problem. They record what happened, not how it felt. The article on why time saved is not ROI goes further into the gap.

There is also good evidence that drafting is where AI helps most. According to MIT research by Shakked Noy and Whitney Zhang, published in Science in July 2023, access to ChatGPT cut the time 453 professionals took on writing tasks by 40%. Output quality, judged by independent evaluators, rose by 18%. The authors' working paper adds that AI moved effort away from rough drafting and towards ideas and editing. That shift is exactly what time to first draft captures.

How do you measure time to first draft?

Pick one recurring workflow, define the start and end events, and record both as timestamps in the tool the work already moves through. Take four weeks of baseline before changing anything. Report the median per workflow each month, because a few unusual jobs distort an average.

A practical setup:

1. Name the workflow. For instance, "monthly client report" or "proposal from brief".
2. Define the start event. Brief marked complete, data loaded, or request approved for work.
3. Define the end event. Draft moved to the review column, or sent to the approver.
4. Use existing tools. A project board, ticket system or shared tracker records both moves.
5. Record a baseline. Four weeks before AI, same people, same work types.
6. Report the median monthly. Split by work type if jobs vary a lot in size.

Keep work types apart. A two-page update and a 20-page proposal should not share a median. If a workflow handles both, track them separately or the number swings with the mix.

Why isn't a faster first draft enough?

Because speed and quality can move in opposite directions. A workflow can produce drafts in minutes that take hours to fix, which lowers time to first draft and raises time to approved. Track the two together, plus the light-edit rate. If only the first number improves, the work moved to the reviewer.

Research shows AI can make work worse as well as faster. According to BCG's 2023 experiment with 758 of its consultants, those using GPT-4 performed 40% better on creative product tasks. On a business problem-solving task outside the tool's strengths, however, they performed 23% worse than colleagues without it.

So read the three measures as a set:

• All three improve: AI is working.
• Only the first draft gets faster: the rework moved to the reviewer.
• Approval speeds up on its own: review got easier, often because drafts are more consistent.

What makes time to first draft fall?

Context, method and complete inputs. AI drafts faster and better when it can read the business's approved facts, voice and past examples, and when the method for the job is written down once rather than retyped. Clever prompts help less than most people expect. Missing context is the usual cause of slow, heavily edited drafts.

Many teams sense the gap. According to Atlassian's 2026 survey of US knowledge workers, only 12% said their AI understands their organisation like a tenured senior colleague. 45% said it was more like a newly hired junior employee, at best. A junior needs briefing every time, and so does an AI without shared context.

That is why the biggest gains come from fixing what the AI reads, not how it is asked. The guide to context engineering explains how to supply it.

What should you do with the number?

Review it monthly per workflow against its own baseline, never against another team's. When time to first draft falls, check that time to approved falls too. If the review queue grows, add reviewer time or tighten the brief. Keep a named person approving everything sent or published, however fast the draft arrives.

Fast drafting moves the bottleneck. Once a draft takes minutes, the approver becomes the slowest step, and pressure builds to wave work through. Resist it. The point of a faster first draft is more time for judgement, not less.

Skein tracks time to first draft for the workflows it builds, beside measures such as pipeline and cost per lead; the outcomes page lists what is measured. That suits teams producing recurring written work, such as proposals, reports, articles and briefs. A team whose work is mostly live conversation, such as support calls, will learn more from other measures.

Frequently asked questions

What does time to first draft mean?

It is the time from a request being complete to a draft fit for review reaching the person who approves it. It is measured per workflow, from timestamps in the tools the work moves through, and reported as a monthly median against a baseline recorded before AI was used.

How do you calculate time to first draft?

Record a start timestamp when the brief or request is complete and an end timestamp when a reviewable draft reaches the approver. Subtract one from the other for each job, then take the median for the month. Keep different work types separate so the mix does not distort it.

Is time to first draft a good measure of AI productivity?

It is a good leading measure for recurring written work, because systems record it and it moves quickly. It is not enough alone. Pair it with time to approved and the share of drafts approved with light edits, so you can see whether rework has simply moved to the reviewer.

What is a good time to first draft?

There is no universal figure, because a two-page update and a 20-page proposal differ hugely. Compare each workflow with its own baseline from before AI. A good result is a lower median for first draft and for approved work, with more drafts needing only light edits.

How can you reduce time to first draft?

Give AI the business's approved facts, voice and past examples, write the method for the job down once so it is reused, and make sure requests arrive complete. These usually beat better prompting. Then watch the review step, because it becomes the bottleneck once drafting is fast.

Related reading