What is Kanban and how to apply it in a creative agency

The costliest misunderstanding about Kanban is treating it as a planning method. It does not decide what to do or when to deliver it: it shows you where already-started work has stopped and forces you to unblock it before taking on more. That is the whole idea, and it is harder to live with than it sounds.

Where did Kanban come from?

Kanban was invented on Toyota's factory floor, inside the production system Taiichi Ohno built from the 1950s onwards. The Japanese word means "card" or "visible signal", and that is literally what it was: a physical card travelling with a batch of parts that, once it came back empty to the previous station, authorized the next batch.

The part that matters is the direction of that authorization. In a push system, every station makes as much as it can and hands the result to the next one, which finds it waiting. In a pull system, the one the card creates, a station only produces when the next station asks for material. Work stops piling up between stations and the pace is set by whoever receives, not by whoever produces.

Ohno said the idea came from supermarkets: the shelf is restocked when a customer takes the product off it, not before and not just in case. A full warehouse looks like wealth and is really money standing still plus problems nobody has noticed yet.

What changes when you move it to knowledge work?

Two things, and both explain why Kanban behaves differently in an agency than on a production line.

The first is that the inventory becomes invisible. In a factory, half-finished work is stacked boxes: they take up floor space and anyone walking past sees them. In an agency, half-finished work lives in open tabs, in a downloads folder, in an email thread with the client and in three different people's heads. Nobody trips over the fourteen started jobs, so nobody counts them. The board exists to make that invisible inventory visible.

The second is that cards are not identical parts. Retouching a banner and rebuilding a brand identity both fit on a card and have nothing in common. That does not break the method, but it does mean reading metrics carefully: an average cycle time drawn from wildly different card sizes says very little, while the age of one specific card, how many days it has sat still, says a lot.

David J. Anderson formalized that translation into knowledge work. His method rests on four change-management principles (start with what you do now, agree to pursue incremental improvement, respect current processes and roles, and encourage leadership at every level) and six core practices. This section covers the four practices that most change an agency's day-to-day; the other two —close feedback loops and improve collaboratively— show up in the last setup step. If you are building the agency's whole way of working, read how project management works in a creative agency first: Kanban is one part of that, not the whole of it.

Which practices actually matter?

Visualize the work

Everything the team has on, in one place and in one format. The working rule is blunt: if somebody is working on something that is not on the board, the board is lying, and a board that lies gets abandoned within a fortnight.

That includes the work nobody wants to add: the last-minute fixes, the favour for the sales lead, the "it is only ten minutes" that happens five times a week. That work exists, it burns the same hours as everything else, and it is exactly the work that vanishes from estimates.

Limit work in progress

A maximum number of cards per column, written on the column itself. When the column is full, nothing new enters until something leaves. It is the principle almost everyone skips first, and the only one that genuinely changes how a team behaves.

Manage the flow

Look at the board to see where work stops, not to see who is busy. Those are different questions and they lead to opposite decisions: the first one leads to helping clear a blockage, the second to handing out more work.

Make policies explicit

Write down what it actually means for a card to move from one column to the next. Who is allowed to move it, what has to be attached, who signs it off. Without that, everyone applies their own judgement and the same card changes column for different reasons depending on who touches it.

Why does limiting work in progress help when it looks like a brake?

It feels wrong because it conflates two things: being busy and delivering. A team with fifteen open cards is unmistakably busy and can go weeks without closing a single one. Clients do not perceive work in progress, only what is finished.

The relationship is arithmetic, not ideological. How long a card takes to cross the board depends on how many cards are inside it divided by the rate at which the team closes them. If twenty cards are open and the team closes five a week, a new card takes four weeks on average to come out, however urgent it is declared to be. Cutting open work is the direct lever on lead time, and it does not require anyone to work faster.

There are two quieter effects. One is the cost of switching: every jump means rebuilding context, and in creative work that context is expensive to rebuild. The other is that problems surface sooner. With work capped, a jam in review blocks the intake column within a day instead of staying buried for a month under layers of newly started work.

How do you design a board for an agency?

Columns have to mirror the real process a job goes through, not the agency's org chart.

The usual mistake is columns by department: Design, Copy, Account, Development. It looks tidy and it fails, because a card that needs copy and design at the same time has nowhere to sit, and because the board stops telling you where the work stands and starts telling you who is holding it. A board built on states reads at a glance: intake, brief, in progress, internal review, with the client, ready to ship.

The intake column deserves its own attention. It is where everything a client asks for lands before it is properly defined, and where you decide whether a job enters with a brief that holds up or enters half-formed for somebody to reconstruct along the way. An explicit policy on that column prevents a good share of rework.

Why does "In review" turn into a warehouse?

Because it is a waiting column dressed up as a working column. While a card is in progress, somebody is doing it. While it is in review, usually nobody is doing anything: it is waiting for someone to find a gap. Because the name sounds like activity, nobody treats it as what it is, a queue, and the queue grows until nobody remembers what is inside it.

Three fixes work, and they combine well:

• Split waiting from working: a "Waiting for review" column with a low cap and a separate "Reviewing" one. Splitting it shows immediately whether the problem is that nobody starts reviewing or that reviewing takes longer than expected.
• Put the reviewer's name on the card. A card waiting for "someone" waits indefinitely.
• Watch card age, not just card position. A card that has sat in review for twelve days is a signal even when the column is not full.

Client swimlanes or work-type swimlanes?

Swimlanes are the horizontal rows of a board, and in an agency the dilemma is nearly always the same one.

The practical answer is to use labels for clients and reserve swimlanes for the one thing that genuinely has to break the reading of the board, which is nearly always urgent work with a committed date. One lane per client in an agency with a long client list turns the board into a spreadsheet. Tools such as the Tasuki Kanban board let you drag cards between lists and label them by client, which is the combination that holds up best as the account list grows.

Kanban or Scrum?

Both organize iterative work, but they start from different assumptions about how much the job can change halfway through.

Kanban wins in most agencies for a very concrete reason: a two-week sprint assumes scope will not move for two weeks, and a client calling on Tuesday with a campaign change breaks that assumption most weeks. Scrum fits when there is a long project with agreed scope and a team that is not simultaneously on five other accounts: a bespoke build, an internal platform, a product of your own.

They can also be mixed, and often are: Scrum's review and retrospective cadence laid over a Kanban flow. What does not work is keeping Scrum's ceremonies without its scope commitment, which means holding the meetings without getting back the stability that justifies them.

When is Kanban not enough?

Kanban shows you now. It does not show you the calendar, and some agency work lives entirely in the calendar.

It falls short in three situations:

• Hard dates with consequences. A launch with media already bought, an event, a delivery with a contractual penalty. The board says the card is in progress; it does not say whether it lands on the 14th.
• Chained dependencies. You cannot shoot until the script is approved and the studio is booked, and the studio has to be booked weeks ahead. Kanban has no way to express that one task cannot start until another finishes.
• Contractual milestones and phased invoicing. When payment depends on a milestone, someone has to be able to see that milestone on a timeline.

That is what the Gantt chart is for: it represents exactly what Kanban lacks, which is duration, dates and dependencies. The combination that works in an agency is using both for different jobs, the Gantt view for the commitment to the client and the board for the team's week, rather than forcing either one to do the other's work.

Kanban also says nothing about whether a project makes money. A board can look immaculate while the account loses money, because flow and margin are separate things: margin needs logged hours and a sale price, which is the subject of project profitability.

How do you start without stopping work in flight?

Kanban does not require reorganizing anything on day one. It goes on top of the process you already have.

1. Draw the real process, the one you actually run, ugly columns included. The ideal process gets designed later, with evidence.
2. Put everything that is open onto the board, no exceptions. The shock of that first snapshot is half the value of the method.
3. Let a week pass without changing anything else and watch where cards pile up.
4. Cap only the column that jams, and set the cap below today's count.
5. Write on each column what it takes to leave it.
6. Change one thing at a time and wait long enough to know whether it helped.

Key points

• Kanban is a pull system: work enters when there is room, not when someone pushes it.
• Visualizing achieves little without capping. The work-in-progress limit is the principle that changes team behaviour.
• A card's lead time is open work divided by the closing rate, so cutting open work shortens delivery without working faster.
• Columns describe the process, never the org chart.
• "In review" is a queue: split it from working, name the reviewer and watch card age.
• Labels for clients, swimlanes only for what genuinely has to break the reading of the board.
• With hard dates, dependencies or contractual milestones, Kanban needs a Gantt beside it.