AI saves your agency time when a comparable client job takes less total effort and meets the same quality standard. Count preparation, checking, revisions and handover alongside drafting. Then record what your team did with the time returned.
You can test this on one repeated job over fourteen days. A simple record will show where the gain survives and where someone else absorbs it.
Why does faster drafting leave the team just as busy?
A faster first draft can move effort into checking and corrections. An account lead may spend the saved hour fixing claims, restoring the client's voice or chasing missing evidence.
Workday's January 2026 survey found that nearly 40% of reported AI time savings went into rework. It surveyed 3,200 active AI users at organisations with annual revenue above US$100 million.
That is a warning about what to count, not a forecast for your agency. Your own job records need to establish the result.
Financial benefits can also lag behind faster work. PwC's January 2026 survey of 4,454 chief executives found that 56% reported no significant financial benefit from AI. Its global sample covered 95 countries and territories, rather than small agencies alone.
Both sources were checked on 13 October 2026. Neither establishes what your team should save.
Which agency job should you measure first?
Choose a repeated job with a clear finish and an available comparison. A weekly campaign summary, a proposal section or a batch of approved social copy can work.
Define the unit before starting. For example: one client report, from a clean data export to an internally approved file ready to send. Keep its scope, inputs and quality requirements stable.
Use several recent comparable jobs to establish the baseline: your record of how the work runs before the change. If reliable records are missing, measure the existing process first. Do not reconstruct precise minutes from memory.
Record how many jobs you compared and who did them. Separate an experienced analyst's work from a new starter's first attempt. A change in client, brief or reporting period may explain a difference that AI did not cause.
The UK Government's AI impact evaluation framework, checked on 13 October 2026, distinguishes efficiency, quality and real-world performance. It also cautions against converting time savings straight into cash savings. The scorecard below is Viravesso's small-agency adaptation, not a government-prescribed test.
Copy this record for each completed job
Use the same definition and quality standard before and during the test. Record a count beside any percentage.
| Measure | What to record | Why it matters |
|---|---|---|
| Completed jobs | Number reaching the agreed finish | Avoid comparing different workloads |
| Preparation and drafting | Total active minutes across everyone involved | Include gathering inputs and writing prompts |
| Checking and revisions | Active minutes spent reviewing and correcting | Count the effort after generation |
| Handover and approval | Active minutes plus number of approval rounds | Include work transferred to another person |
| Elapsed time | Time from agreed start to agreed finish | Separate waiting from working |
| Quality | Accepted first time, major rewrites and errors | Keep the same acceptance criteria |
| Tool and setup cost | Fees, usage charges and one-off setup time | Make the full investment visible |
| Returned capacity | What the team did with the time | Distinguish useful capacity from assumed revenue |
How do you run the fourteen-day test?
Keep a short record after each completed job. Use a timer or the team's existing time records, then add the reviewer's effort before closing the entry.
During the first working week, use the agreed AI workflow on comparable jobs. Keep the same source pack, approval standard and named reviewer. Count unsuccessful attempts and the time spent returning to the previous method.
During the second week, repeat it with the same rules. Log any change to the prompt, tool or process. If you change several things together, you can assess the combined workflow but cannot isolate AI's contribution.
Fourteen days is a practical first checkpoint. It is not proof of a lasting gain. The Government framework treats short studies as indicative; a monthly report may need several reporting cycles before a useful comparison exists.
Tell the team what the test measures and why. Use it to improve the process, rather than rank colleagues by how fast they prompt. A person who catches a false claim is doing necessary work.
Agree the release standard in advance using the agency AI review gate. A time saving only counts alongside the quality needed for that job.
What does a useful result look like?
A useful result shows fewer total working minutes per accepted job, with quality holding steady. Keep elapsed delivery time beside it: less effort and faster delivery are different outcomes.
The following client-report example is hypothetical. Both versions use the same approved data and finish at the same internal release decision.
A faster draft can produce a smaller whole-job gain
Hypothetical minutes for one comparable client report. These are illustrative figures, not measured agency results.
| Stage | Previous process | AI-assisted process |
|---|---|---|
| Prepare approved source data | 20 minutes | 20 minutes |
| Draft commentary | 60 minutes | 20 minutes |
| Check and revise | 25 minutes | 40 minutes |
| Handover and approval | 15 minutes | 15 minutes |
| Total active effort | 120 minutes | 95 minutes |
| Net time returned | Reference point | 25 minutes |
The draft saves forty minutes in this example. Extra checking consumes fifteen, leaving twenty-five minutes across the whole job. Count each minute once: checking is already inside the ninety-five-minute total.
If four comparable reports each returned twenty-five minutes, the team would regain one hundred minutes. That becomes usable capacity only when someone can use it. Four scattered gaps may not fit another client job.
Record major rewrites and first-pass acceptance alongside the totals. Three accepted reports out of four is more informative than an unexplained 75%. A tiny sample can change sharply after one difficult brief.
When does saved time become a financial return?
Saved time produces a financial return when it changes a real cost or contribution to profit. An unused hour does not automatically become an extra billable hour.
Record what happened: less paid overtime, fewer external production costs or extra accepted work. Better research and a calmer deadline can be worthwhile outcomes too. Describe them honestly without assigning invented revenue.
For a financial comparison, use benefits and costs from the same period. Include subscription fees, usage charges, setup and training, plus the human effort needed to run and check the workflow. Allocate shared tool costs once using a method you can explain.
Keep the first month's setup cost visible. You can also show the recurring result after setup, but label the difference. Do not remove awkward costs until the number becomes attractive.
If you calculate a percentage return, divide net financial benefit by the investment cost and multiply by one hundred. Net benefit means the financial gain after the investment cost. Avoid counting the same labour saving both as lower job cost and extra revenue.
The agency fee discussion is a separate decision. Your records help you understand delivery economics; the agreement with the client determines what you charge.
Decide what to do with the workflow
Review quality and the sample before deciding. No universal time-saving percentage qualifies a workflow.
| Observed result | Decision | Next action |
|---|---|---|
| Less effort, steady quality, useful capacity | Keep testing | Check that the result repeats |
| Faster drafting, more corrections | Change one stage | Improve the source pack or review handover |
| Same effort, stronger accepted work | Assess the quality benefit | Record the improvement without claiming time saved |
| More effort or weaker quality | Pause this use | Return to the previous method and inspect the cause |
| Too few comparable jobs | Extend the observation | Wait for another relevant work cycle |
What should you do this week?
Choose one repeated client job and name its finish. Record the existing effort, agree the quality standard and give one person responsibility for collecting the test results.
At the fourteen-day checkpoint, review the whole job with the people who do it. Keep the workflow if the evidence supports it. Change one weak stage or stop the use when it adds effort without sufficient benefit.
ViraOps Workflow Playbooks help you finish agency jobs with a defined sequence, prompts and expected outputs. Pick the relevant playbook, then use the scorecard to test whether the process earns its place in your team.
Finish a repeatable agency job with a clear process. ViraOps Workflow Playbooks give you the stages, prompts and expected outputs to run it.
