Process improvement in an agency: measure one workflow, cut waiting time and rework

Summary

A job is often worked on quickly and then sits around for a long time. You only see where the time goes once you measure a single workflow for four weeks: lead time, waiting time between steps, rework loops and the number of open jobs. Little’s Law links three of these numbers in one simple calculation¹². Then you change exactly one thing, as a test with a criterion set in advance, and measure again⁵⁸. At the end you will find a measurement sheet and a four-week plan.

How long does a finished draft sit with you?

Ask yourself one question before you read on: how long does a finished draft sit with you before someone approves it? A day? A week? Most teams cannot say. People know how many hours went into a job, because time tracking asks for it. How many days the job was on its way, nobody writes down.

Yet there is a calculation that needs only three numbers. John Little proved it in 1961: the average number of items in a system equals the arrival rate times the average time an item spends in the system, L = λW¹. Fifty years later he put it like this: “Little’s Law says that the average number of items in a queuing system, denoted L, equals the average arrival rate of items to the system, λ, multiplied by the average waiting time of an item in the system, W.”²

A worked example with made-up numbers: your team has 12 jobs open at the same time on average, and 3 are finished each week, roughly as many as come in. Then a job is on its way for 12 divided by 3, so 4 weeks on average. You do not have to track a single job for this. What counts as the “system” is up to you, Little leaves that open on purpose¹. In this article it is one workflow: from the moment a job comes in until it is signed off.

Working time is not lead time

Working time is the time someone actually works on the job. Lead time is the time from start to finish, including all the waiting in between. The two numbers are often far apart, and the gap opens up between the steps.

Another worked example, also made up: a monthly social media package with twelve posts. The concept takes three hours, copy and graphics six, the internal review one, the corrections after client feedback two. That is twelve hours, a good one and a half working days. The calendar looks different. The brief waits two days before anyone starts. The graphics wait a day for the copy. The internal review comes two days later. The package sits with the client for five working days, corrections start a day after that, and the second approval takes another three days. That adds up to about 16 working days of lead time for one and a half days of work.

How large the share of waiting really is depends on the team. A case study at a software company for horticultural farms calculated it from the data in its development tools: there, 72 percent of total cycle time was waiting time¹¹. The authors name batching, prioritisation and people not being available as the main drivers¹¹. That is a single company in software development, and only the abstract of the study was available for this article. It gives you no general rate for agencies. That is why you measure yourself.

Pick one workflow

Start narrow. One workflow is enough, and it should meet three conditions: it happens often, it bothers many people, and it can be counted. A workflow that runs twice a year gives you no numbers in four weeks. A workflow that annoys only one person will not find allies.

In many agencies, client corrections meet all three conditions: from the first draft to sign-off. These loops run every week, involve creative, account management and the client, and each one has a date. The German agency association GWA writes in its study on agency-client relationships that ideally agencies and clients work together so well “that they do not burn money in endless and pointless loops” (our translation)¹³. How many loops there actually are, the study does not say. Most teams are missing exactly that number for themselves, too.

If things already get stuck earlier, at the brief or at project kick-off, read the article on the operations bottleneck first. This article picks up after that.

Agencies are not alone in this. Any commissioned work with approvals has the same shape: quotes in mechanical engineering, invoice checks, HR cases, the monthly close. A case study at a waste treatment association of several municipalities in the Porto region shows what was possible there: the monthly accounting close went from 8 to 4.5 days, payroll processing from 10 to 5 days¹². The team used several lean methods at once and the authors link the results mainly to value stream mapping¹². There is no comparison with a group that changed nothing.

Measure four numbers, no new tool

For four weeks of measuring you need four numbers:

1. Lead time: days from arrival to sign-off, per job.
2. Waiting time between steps: days in which the job was ready for the next step and nobody worked on it.
3. Rework loops: how often a job went back to an earlier step.
4. Jobs open at the same time: counted once a day, at the same time of day.

You do not need a new tool for this. A tally on the board is enough, or two date columns in the task list you already use: “start” and “end” for each step. Waiting time is then the gap between the end of one step and the start of the next.

The first, second and fourth numbers are linked by Little’s Law. The 1961 proof assumed stable conditions, strict stationarity in technical terms¹. In his look back in 2011, Little shows that over a limited observation period the relationship holds without that assumption and is even numerically exact². One catch remains: jobs that were already open at the start or are still open at the end only count with their time inside the period². Little’s own summary: “we are in the measurement business, not the forecasting business.”² The formula tells you how things were. What to change, it does not tell you².

If you want to bring numbers like these into a fixed rhythm for the whole agency, you will find that in the article on efficient agency management. Here it stays with one workflow.

One change, as a test

After the first week of measuring, one step usually stands out: jobs pile up there, and the waiting time is longest there. At that step you change exactly one thing. With two changes you will not know afterwards which one worked.

The approach is old. W. Edwards Deming describes the cycle like this: plan a change or a test, carry it out, “preferably on a small scale”, study the results, and then adopt the change, abandon it or run through the cycle again⁸. Deming called it PDSA, with “Study” instead of “Check”, and considered the common form PDCA a corruption⁷. ISO 9001 in its sixth edition, published on 16 September 2026, sticks with PDCA and states that the cycle can be applied to all processes⁵⁶. For your test the name does not matter. What matters is that you set the criterion in advance, for example: “Average waiting time before client approval drops from five to three days.”

Three changes that work well to start with:

• Fixed approval windows: the client approves on Tuesdays and Thursdays, and someone does the internal review every morning. This targets waiting time directly.
• Acceptance criteria before the start: before the first draft, it is clear how the client will recognise “done”. This targets rework loops.
• A limit on jobs open at the same time: the Official Kanban Guide describes this limit as “the maximum number of work items allowed at a time”, which can be set per work state, per person or for a whole system⁹. The Kanban Guide by ProKanban.org does not use the term; it only requires that members of a Kanban system “explicitly control the number of work items in a workflow from started to finished”¹⁰. More on this in the article on Kanban.

Little’s Law gives you the reason for the limit. In the worked example above, 12 jobs were open and 3 finished per week, so 4 weeks of lead time. If throughput stays at 3 per week and only 6 are open, the calculation gives 2 weeks. Whether throughput really stays the same, only measuring will show. In the software case study, cycle times fell by about 30 days after projects were cut smaller¹¹.

The limit is often justified with the cost of switching between tasks. The research on that is narrower than it sounds. Rubinstein, Meyer and Evans measured switching costs in the lab, for simple tasks, in fractions of a second: in their first experiment a switch cost 975 milliseconds on average, more with complex rules, less with a cue for the next task⁴. Mark, Gudith and Klocke had 48 people answer emails. In the rounds with interruptions they actually finished faster, not counting the interruption itself, at 20.3 or 20.6 instead of 22.8 minutes on average³. But after only 20 minutes they reported significantly more stress, frustration, workload, effort and pressure³. In the lab, then, interruptions cost nerves more than minutes. The stronger reason for the limit is still the calculation.

You will find more methods against waste in the article on lean. For the test, one is enough.

Write down what applies now

After week 4 you compare the numbers with week 1. If the change met the criterion, it becomes the rule. If it did not, you drop it or start a new test, just as in Deming’s last step⁸. ISO 9001 defines continual improvement as a “recurring activity to enhance performance”⁵. After the first test, the step with the next longest waiting time is up.

Only document what has proven itself. One page is enough: the workflow in steps, the new rule, the date it applies from, and one person who owns it. Who decides at which step and who is only informed, you settle with a RACI matrix, for example using the RACI Builder. For the page itself there is the Process Documentation Kickstart playbook.

Work aid: workflow measurement sheet

One sheet per job. For each step, note who does it, when it starts and when it ends. Waiting time is the number of working days between the end of the previous step and the start of this one. The rows below are examples from the monthly package.

In addition, once a day at the same time, a tally: how many jobs in this workflow are open right now?

Four weeks is short. If the workflow runs only a few times a week, extend both measuring phases rather than turning three jobs into a rule.