Skip to content
Interface language: English
Imperium / Signal

The team starts everything and finishes little: the cost of too much work in progress

Too many started items hide queues and dependencies. Compare arrivals, completed work and waiting time without a magic task limit.

Article language: English

A seated woman beside a glass sphere with dark reflections.
In this article

On Monday, twelve items are in progress. By Friday, the team finishes four and starts seven more. Nobody has been idle, the chat has been busy, and the open list has grown to fifteen. In this fictional example, the team now has three additional commitments to revisit.

Starting work is easy to see: someone creates a file, replies to a customer or opens a task card. Finishing is often less visible. It means gathering comments, checking the result and handing it over in a form the recipient can use. If a company rewards motion alone, it may accumulate more and more started work while losing sight of when anything will actually leave the team.

First define what is really in progress

The May 2025 Kanban Guide defines work in progress, or WIP, as work started but not finished. It calls for explicit start and finish boundaries and control of work between them. WIP therefore includes more than what someone is working on at this precise minute.

Your team needs to name those boundaries in ordinary language. Does an item start when a commitment is accepted, when materials arrive, or only with the first action? When is it really finished: after preparation, checking, sending, or agreed acceptance? Different answers can suit different tasks. The problem arises when people use different definitions while a report pretends to show one shared measure.

Imagine a fictional studio preparing exhibition materials. A designer has made a display panel, but the file is waiting for its dimensions to be checked. From the designer's perspective, the creative work is done. From the customer's perspective, the panel is not yet ready for production. The team should see both states instead of leaving the item in the checker's private memory.

A request nobody has accepted also needs a visible place. It need not be mixed with started work, however. An article about an internal service catalogue (in Polish) helps explain how a company receives different requests. A message arriving should not silently commit the team to any deadline.

Items awaiting an outside response do not disappear. Waiting for the customer affects the team's future load, especially when many items may return at once. The state can be shown separately, but the team must retain sight of the accepted commitment and the conditions for its return. The guide to work awaiting a customer (in Polish) considers what is missing and how work resumes.

At the first review, record why each item remains open. A panel awaiting data, one ready for checking and one requiring correction need different actions. A single “in progress” status hides that difference. The review need not immediately produce an elaborate board. It should identify the next feasible step for each selected item.

Consider the oldest item. Is it old because work is continuing, or because nobody clarified a missing condition? If it has waited weeks for a decision, adding someone to create new panels may not help that case. The team needs an answer suited to the type of blockage, not a blanket diagnosis that everyone has too much work.

Starting more can hide where work stops

In our studio, five panels await the same person who checks file compliance. The rest of the team starts new items because they have time at their own stage. Locally this looks sensible: everyone is using their time. Across the process, however, intermediate results keep accumulating for the same specialist skill.

A question about a new app is premature. First find out why work does not move on. Does checking require missing data? Does the checker have other commitments? Do similar corrections recur because the team understands the finish condition differently? Each answer points to a different change.

Naming a bottleneck is not an accusation against a person. If work reaches one stage faster than it can leave, examine the arrangement of tasks and available skills. Opening more items does not explain that arrangement. It may, however, increase the number of people asking for an update on the same day.

Distinguish choosing important initiatives from managing current work. The guide to project priorities with a limited team (in Polish) explores how to choose between initiatives. Here the question is how accepted items reach completion. Even well-chosen projects can stall if all are started at once.

As a trial, the studio can review the five waiting panels before starting a sixth. Perhaps two only need data confirmed, one has returned for changes and two are ready for checking. This remains a fictional example, not a promise of those results. It shows that helping work reach the next stage can mean something other than assigning everyone new tasks.

The review should also retain items no longer worth continuing. Cancellation needs its own status and reason. Do not present it as a finished result, but do not preserve an empty commitment merely because work once began. Recipients and participants need to know what was closed, what is still waiting and who can decide the remaining scope.

Item count, completion rate and time tell different stories

In their 2008 chapter on Little's Law, John D. C. Little and Stephen C. Graves describe a relationship between the average number of items in a system, the average arrival rate and the average time spent in it, under assumptions of stable operation over a longer period. The relationship concerns averages and defined assumptions, not the deadline for one item.

For a team, that is a reason to compare several quantities instead of judging by the number of busy people. How many items were accepted? How many were finished? How many remain open, and for how long? A short period can show the direction of a queue's change, but one division is not enough to promise a customer an exact day. To understand separately what activities people perform, consider transparent work sampling. The proportion of observations spent checking does not reveal how long a particular panel waits in a queue. The questions need separate records. The guide to work sampling (in Polish) explains how to record the kinds of work observed.

Aeneia conceptual photography

The Monday example balances simply only because we assumed no cancellations or reopenings. A real register must distinguish them. An item closed because it was abandoned is not a result delivered to a recipient. An item reopened after an error calls for a different reading from a new request. Counting must preserve the meaning of the work.

Do not treat a panel, a catalogue and an entire exhibition stand as identical units. If scope varies greatly, say so and identify groups suited to the question. You do not need a large points system. You do need to know what a number means before using it to judge a change.

Limiting starts is an experiment with conditions

A company can select one type of item and, for an agreed period, limit new starts while concentrating on moving open work to its next agreed state. No one number suits every team. The starting point should reflect tasks, skills and real dependencies.

In the panel studio, the experiment may concern the stage before checking. The team agrees what must be ready before another item reaches the checker and how it will clarify missing information in open cases. It needs an output that can actually move forward and information about what is missing to prepare it.

Before the trial, agree what happens with an urgent item. If every request gets an exception, the limit is decoration. If no route exists for a genuinely urgent situation, people may start bypassing the register. The rule should support an explicit choice: what do we accept now, and what does that mean for work already started?

A promised deadline between sales and delivery connects incoming work with a commitment to the customer. For the conditions behind a delivery promise, consult the guide to deadline promises between sales and delivery (in Polish). The person accepting an item needs to see actual availability and conditions for starting. They cannot rely solely on the team having somehow managed before.

Examine the trial from the requester's side too. If the team limits started work, the requester should understand what happens to the next request. Will they receive queue information, a request for missing data or a proposal for a different scope? An internal limit without a clear response may simply move pressure into private messages. The queue still exists, but is harder to see.

After the agreed period, compare finished results, open items and their waiting times. Do not declare success merely because fewer items were started. If the team finishes less necessary work or blocks appropriate requests, investigate and change the trial. The rule should help produce useful results, not protect a chosen metric from reality.

Suppose two panels move to the next stage after the trial and the number of remaining open items falls. That is not a full assessment. Check whether comparable work was actually done or merely moved to an invisible list. If the team improved its status records, the result may mean better information rather than faster work. Both may be useful, but they need separate names.

If a new urgent job then arrives, the rule should permit a conscious choice. Which work gives way? Who is affected? Who communicates the change? A colour on an item does not free time or skills. In our studio, identify the actual effect on preparing and checking the other panels instead of adding an urgent item to a schedule that no longer reflects accepted work.

At the end of the experiment, listen to recipients too. Are they receiving material in a state they can use? Have corrections simply moved beyond the measured team's boundary? If the customer's check reveals the same defects, moving a card into “done” earlier does not describe the whole effect. The measurement boundary may remain local, but its limits should be visible in the interpretation.

The studio's next decision might be to admit only materials with confirmed dimensions to checking. If that missing information was what stopped the selected cases, test the effect of this particular change on subsequent work.

Do not add interruptions to an open queue

Every status question can trigger another return to an item. Interruptions in team work should therefore be considered alongside the number of open items. Limiting formal starts is not enough if chat still forces everyone to reconstruct several contexts at once. The guide to interruptions at work (in Polish) considers how people return to a task.

Automation can produce intermediate work too. If a tool prepares drafts faster and each needs the same human check, the number of files can grow without a comparable increase in finished results. That is a hypothesis to test in a particular workflow. An automation exception queue helps reveal some of the work returned to people. The guide to automation exception queues (in Polish) describes how such cases are handled.

During a seasonal peak in enquiries, incoming demand can change substantially. Reassess the conditions for accepting and finishing work. A fixed rule without observation can lose its value just as a complete absence of rules can. For changes in demand and capacity, the guide to seasonal peaks in enquiries (in Polish) offers a related perspective.

For the draft-producing tool, count a working output separately from a result after the required check. If a report treats each new file as completed, the team may miss growing load on reviewers. If it measures only the final handoff, it must still be able to find the started materials. The two views answer different questions and together make the work's path traceable.

Aeneia Audit can help examine whether AI tools move work toward a useful result or simply increase intermediate outputs. Start with a specific part of the work, its data and a verifiable distinction between starting and finishing.

Aeneia / Aeneia editorial

Editorial article. Source links are included in the text.

About Aeneia

Brands we have worked with

  • L’Oréal
  • Garnier
  • Netflix
  • Carrefour
  • Warner Bros. Discovery
  • Biedronka
  • HBO Max
  • Wolt
  • Żabka
  • COSRX
  • Embassy of China in Poland
  • Mixa
Aeneia / Initium

From knowledge to a decision

Tell us what you want to put in order. We will agree on where to start.

Contact Aeneia

We partner with and use solutions from:

  • Google
  • OpenAI
  • Microsoft
  • Stripe
  • Anthropic
  • Linear