Project management for creative agencies: the complete guide

Creative agencies rarely fail for lack of talent. They fail at the seams: the job that arrived as a passing message and never got written down, the third round of changes nobody counted, the project the client loved that lost money. Managing agency projects is mostly the work of closing those seams.

Why doesn't an agency run like a software team?

Both worlds use the same vocabulary — tasks, deliveries, deadlines — which is why agencies so often adopt tools and rituals built for product teams without asking whether they fit. The work has three differences that change everything.

Several clients run at the same time. A product team has one product and prioritises inside it: when something moves up, something else moves down, and one person makes the call. An agency has clients who never speak to each other and who each consider their own job the urgent one. There is no single queue. There are as many queues as clients, all competing for the same pairs of hands.

Creative work comes back. A feature ships and is done. A design route is presented, commented on, redone and presented again. That loop is not a defect in the process; it is the process. A system that only understands "to do, doing, done" has nowhere to put half the real work: the half sitting still, waiting on someone outside the team.

Profit is decided job by job. In an agency the margin is not measured at the end of the quarter, it is measured on every fixed-price piece of work. A project that burns twice the hours estimated cannot be rescued after the fact. It can only teach you something before you quote the next one the same way.

What that means day to day

• "Waiting on the client" has to be a visible state. Without it, stalled work hides inside work in progress and distorts the team's real load.
• Load is read per person, not per project. Every project can look comfortable while the designer who appears in five of them is completely blocked.
• Every task must be attributable to a client. Otherwise there is no way to know what serving each one actually costs.
• Logging hours is not bureaucracy. It is the data that turns a hunch about profitability into a conversation with numbers in it.

What are the phases of agency work?

There are five, and they work best treated as gates: until a job has settled the middle column, it does not move on. The third column is the one you will use most, because it names the symptom long before it shows up on an invoice.

What does a job need before it enters production?

Fixing intake costs little, and skipping it costs a lot. A job ready for production carries five things: the problem it solves, who it is for, how far the work goes, who has authority to approve it, and when it is genuinely needed — not the anxious date agreed on a phone call.

What breaks this phase is almost never a missing template. It is a missing place. Jobs arrive by email, by phone, by chat and in corridor conversations, and anything not written somewhere shared stops existing the moment the person who heard it goes on holiday. A single inbox — a list, a column, whatever, but one — solves more problems than any brief format.

Scope deserves its own line. "Design the campaign" is not scope, it is a title. Scope says how many pieces, in which formats, and what is excluded. The excluded part is by far the most useful thing in a brief, because it is what lets you say "that's a new job" without sounding defensive.

How that document gets written and agreed is a guide in itself, and it is here: creative brief.

How do you share out work across several clients at once?

Planning in an agency is two separate questions, and confusing them causes most schedule disasters. The first is what needs doing and in what order. The second is who actually has hands free. A plan that answers the first and ignores the second produces beautiful timelines nobody can meet.

For continuous flow — the work that arrives, moves and leaves on any given day — a status board works best, because work is visible as it moves and jams announce themselves when a column swells. That is the home ground of the Kanban method, above all its limit on work in progress: refusing to start anything new until something open is closed shortens each piece's delivery time, because fewer things are waiting at once, and it is the rule with the largest effect on delivery times.

For work with dependencies and committed dates — a campaign with a shoot, a website with content milestones — you need to see time horizontally: what depends on what, and what collapses if one piece slips. That is a Gantt chart, and it earns its keep precisely where the board runs out of answers.

Most agencies need both, over the same data: the board for daily work, the timeline for the client conversation. That is what Tasuki does with multiple views of one project, and it is the practical reason the two coexist rather than compete.

How do you track execution without filling the week with meetings?

A status meeting whose only purpose is having everyone say where they are is a management tool paid for at creative hourly rates. The goal is for status to be readable, not recited.

Three agreements get you there, and none of them is technical. First: status lives on the task, not in the head of the person doing it. Second: moving the card is part of finishing the work, the same way saving the file is. Third: if something is blocked, it gets marked as blocked and names who is blocking it — even when that is the client, and especially when saying so is awkward.

With that settled, tracking becomes two glances. What has not moved in too long, which is usually a block nobody declared. And who has more open cards than anyone can reasonably hold, which is the jam showing up before the delivery date does.

How do you get client sign-off without endless rounds?

Sign-off is one of the phases where agencies quietly lose margin, because extra rounds never feel like extra work. They feel like finishing properly. Three rules keep them contained.

Agree the number of rounds in writing and count them out loud. Not to refuse the fourth, but to be able to say "this is the fourth" so the budget conversation happens then, rather than at invoicing.

Know who approves. The problem is rarely the number of rounds. It is that round three brings in someone who has seen nothing so far and starts from the beginning. That person must be named in the brief and present at the first review.

Collect feedback in one place, against one version. Comments spread across three emails, a chat group and a phone call force the team to reconstruct the request before acting on it, and that reconstruction is where the mistake that causes the next round gets in. Keeping comments attached to the task, with a date and an author, is not tidiness for its own sake: it is what lets you answer "this was approved on the 12th" with a link instead of a memory.

How do you close a project so it teaches you something?

Closing is not archiving. A useful close compares estimated hours against actual, points to the phase where they went, and writes down what would be done differently. Five lines is plenty. What does not work is skipping them, because then the knowledge stays with the person who ran the project and leaves when they do.

That close depends on data that only exists if it was collected along the way: hours logged against tasks. Without them, the question of whether a client is profitable gets settled by impression, and impressions tend to favour the pleasant client over the one who leaves margin. With them you can work out what a job really cost and decide whether the next one gets quoted differently or declined outright. That calculation, step by step, is in project profitability for agencies.

Two records are worth keeping apart. Logging hours against tasks tells you what a project costs; it is an internal management tool, like the time tracking in any project system. Recording working hours is something else: it is an employment obligation with its own rules about what must be recorded and how long it must be kept, covered in working time records in Spain. Using one dataset for both jobs is a reliable way to end up with neither done properly.

When is formalising the process not worth it?

Process costs attention, and attention is the scarce resource in a small agency. In three situations, adding structure takes more than it gives.

When the team fits around one table and shares every project, the cost of keeping a system current can exceed the cost of turning your chair and asking. Structure starts paying when someone stops finding things out by proximity.

When the process is designed before the work exists. A board with twelve columns and four label taxonomies, built on day one, is abandoned by day three. Two or three states over real work, and expand only when something concrete hurts.

When a rule has no owner. An agreement nobody says out loud when it is broken is not an agreement, it is a document. One rule that actually holds — say, no work starts without a brief — beats a complete manual that everyone ignores.

Key points

• An agency runs several simultaneous queues, not one: read capacity per person, never per project.
• "Waiting on the client" must be visible, or stalled work gets counted as work in progress.
• Scope is defined as much by what it excludes as by what it includes.
• Review rounds are agreed in writing, counted out loud, and approved by someone named back at the brief.
• Without hours logged against tasks there is no conversation about profitability, only impressions.
• Logging hours to tasks and recording working time are different things with different purposes.
• Add process when a specific symptom hurts, never before: process designed in the abstract gets abandoned.