Atelier

One campaign. Three sides. One source of truth.

A creator campaign management concept for agencies coordinating work across brands and creators — designed around shared campaign state, role-specific workflows, and the decisions that need attention now.

B2B SaaS

Complex Systems

The Problem

A single campaign can move between an agency coordinating the work, a brand commissioning and reviewing it, and several creators negotiating, producing and delivering against it.

Each side needs different information, different actions, and different levels of control.

I started Atelier with a product hypothesis:

As campaigns, creators, approvals, negotiations and deadlines multiply, maintaining a shared understanding of what is happening becomes harder.

I didn't conduct primary research with agency teams before designing this concept, so I treated that as a hypothesis to design against — not a validated finding.

The question became:

How do you give three sides of a campaign what they need without creating three disconnected versions of the truth?

That question shaped the architecture before it shaped the interface.

The product bet

Existing creator-management platforms already address parts of this workflow. I wasn't trying to invent the category; I was exploring a specific product architecture for coordinating agency, brand and creator work through one shared system.

The agency became the system of record.

I considered treating Atelier as a marketplace connecting brands and creators. That would have made discovery the center of the product.

But discovery wasn't the problem I wanted the product to organize itself around.

The agency sits between both sides of the campaign: maintaining its creator roster, coordinating campaign progress, responding to negotiations, tracking deliverables and approvals, and maintaining visibility across work.

So I made one foundational decision:

The agency would operate the system. Brands and creators would participate in it.

That changed Atelier from three parallel experiences into one shared operational model with role-scoped interfaces.

Agency
Coordinates stakeholders, owns the roster, operates campaigns and maintains visibility into campaign performance.
Brand
Discovers eligible creators, reviews work, makes campaign decisions and evaluates performance.
Creator
Acts on their own collaborations, deliverables and next steps.

The shared campaign stays consistent. The amount of product each role sees does not.

One system,
three depths

Sharing a campaign doesn't mean sharing an interface.

Role-based products can easily become a permissions exercise: design one interface, then hide a few buttons depending on who's logged in.

I wanted the distinction to run deeper.

The Agency needs operational depth: campaign oversight, creator management, negotiations, approvals, deadlines, compliance, payments and performance.
The Brand needs decision surfaces: discover eligible creators, shortlist, create and review campaigns, approve work, and understand campaign performance.
The Creator needs focus: what am I working on, what needs me next, and what am I waiting on?

So each role gets its own navigation and hierarchy while operating on the same campaign model. The final prototype routes each role into genuinely different product surfaces rather than three copies of the same dashboard.

The amount of interface should follow the amount of responsibility.

Modelling the workflow

A campaign has a lifecycle. A creator has a position inside it.

I modeled campaign work across 14 stages, from initial discovery through negotiation, creation, review, publishing, payment and completion.

For navigation and reporting, those stages collapse into six broader operational phases:

CASTING → NEGOTIATING → CREATING → REVIEW → PUBLISHING → PAYMENT

That structure works neatly when a campaign has one creator.

Then I modeled a more realistic case.

What happens when one campaign has three?

Priya Kapoor
Content Creation · 5 days left
Diego Reyes
Negotiation · 9 days left
Ines Costa
Approval · 2 days left

There is no honest single stage for that campaign.

The model had to change

One campaign isn't always in one stage.

Forcing Priya, Diego and Ines (our 3 hypothetical creators) into a single campaign status would make two-thirds of the picture wrong. So I moved lifecycle state down to the creator assignment.

Each assignment can independently hold its stage, deadline, deliverables, negotiation and attention state.

The campaign then derives the information needed at portfolio level from those assignments.

Each assignment can independently hold its stage, deadline, deliverables, negotiation and attention state.

The campaign then derives the information needed at portfolio level from those assignments.

The roll-up answers “where should I look?” while the workspace preserves what is actually happening creator by creator. The final implementation derives these campaign-level values from the underlying assignments rather than replacing their individual states.

Inside the workspace, selecting a creator changes the lifecycle and work shown beneath it.

I stopped asking, “What stage is this campaign in?” and started asking, “What is happening inside this campaign?”

Priya can be creating content

while Diego is negotiating

and Ines is waiting on approval

The system had to know what needed attention

“Needs attention” couldn't become another status someone had to maintain.

My first dashboard organized lifecycle, upcoming work, money and attention with roughly equal weight. It technically worked, but it made the user understand the state of the system before showing them what required action.

The better question was: What needs me right now?

So I rebuilt the hierarchy around attention — and if the dashboard was going to prioritize action, the system needed to infer attention from the work itself.

Atelier derives attention from assignment state, including:

Those conditions are prioritized so the most consequential blocker can roll up to the campaign level.

Structured workflows feed the same model. A creator countering campaign terms doesn't just change the negotiation record — it becomes an agency action until someone responds. The final attention logic explicitly accounts for negotiation turn, review state, overdue work and deadline risk.

That allows the interface to surface something actionable:

Diego Reyes countered your offer — your response is needed.

rather than simply:

Renewal Ritual Launch — Negotiation.

The dashboard translates underlying workflow state into:

What happened → who owns it → what needs to happen next.

A chart only earns a place if it changes what someone does next.

The dashboard translates underlying workflow state into:

What happened → who owns it → what needs to happen next.

A chart only earns a place if it changes what someone does next.

Visibility before discovery

Some product rules have to exist before the search box.

Creator discovery introduced a different kind of problem.

The obvious flow would be:

Search roster → filter → rank results

But an agency roster isn't automatically a catalogue every brand should be able to browse.

Creators in Atelier can be:

Discoverable
Selected brands
Unavailable
Restricted

So I treated visibility as an architectural rule rather than another discovery filter.

But an agency roster isn't automatically a catalogue every brand should be able to browse.

Creators in Atelier can be:

Discoverable
Selected brands
Unavailable
Restricted

So I treated visibility as an architectural rule rather than another discovery filter.

For a Brand user, Atelier first constructs the pool of creators that brand is eligible to access. Only then can browsing, search, filters or AI matching operate on it.

Browse and AI Match operate on the same pool

An earlier version of Brand Discovery made the distinction between Browse and AI Match less explicit than it needed to be.

Browse lets a Brand manually explore creators the Agency has already made available to it. It isn't an open marketplace.
AI Match doesn't unlock another creator network. It ranks creators from that same eligible pool based on their fit with the campaign's requirements.

So the distinction isn't who Atelier can access. It's how the Brand evaluates the creators it can already access.

In both cases, Agency-controlled visibility is resolved first.

AI can help prioritize an eligible pool. It can't expand it.

The order becomes:

AGENCY ROSTER

↓

VISIBILITY RULES

↓

BRAND-ELIGIBLE CREATOR POOL

↓

BROWSE · FILTER · SEARCH · AI MATCH

A creator who shouldn't be visible to a brand shouldn't enter its ranking system in the first place.

Visibility isn't a filter on discovery. It's a boundary around discovery.

Browse and AI Match operate on the same pool

An earlier version of Brand Discovery made the distinction between Browse and AI Match less explicit than it needed to be.

Browse lets a Brand manually explore creators the Agency has already made available to it. It isn't an open marketplace.
AI Match doesn't unlock another creator network. It ranks creators from that same eligible pool based on their fit with the campaign's requirements.

So the distinction isn't who Atelier can access. It's how the Brand evaluates the creators it can already access.

In both cases, Agency-controlled visibility is resolved first.

AI can help prioritize an eligible pool. It can't expand it.

The order becomes:

AGENCY ROSTER

↓

VISIBILITY RULES

↓

BRAND-ELIGIBLE CREATOR POOL

↓

BROWSE · FILTER · SEARCH · AI MATCH

A creator who shouldn't be visible to a brand shouldn't enter its ranking system in the first place.

Visibility isn't a filter on discovery. It's a boundary around discovery.

Closing the Loop

Publishing shouldn't be where the campaign stops being useful.

My original model focused on getting a campaign from discovery through execution. But once content was published, Atelier became much less useful.

That exposed a gap:

If campaign outcomes can't inform the next campaign, the system records work without learning from it.

So I added a Performance layer for both Agency and Brand, combining campaign-level metrics with creator-level performance.

Rather than reducing performance to a single success score, Atelier keeps the evidence visible — making it easier to see what performed, who contributed, and what might inform the next campaign.

The same data serves different decisions: the Agency can use it for campaign oversight and future casting, while the Brand can use it to understand outcomes and inform future campaign and creator decisions.

Rather than reducing performance to a single success score, Atelier keeps the evidence visible — making it easier to see what performed, who contributed, and what might inform the next campaign.

Attribution needed boundaries too.

Not every change observed during a campaign can honestly be attributed to it.

Atelier distinguishes tracked campaign activity from broader contextual signals. For example:

Attributed visits → tracked campaign links

Site traffic during campaign period: +14% → observational

The second may be useful context. It isn't proof that the campaign caused the increase.

The interface shouldn't imply more certainty than the underlying data can support.

DISCOVER → EVALUATE → NEGOTIATE → EXECUTE → REVIEW → PUBLISH → MEASURE → LEARN

This closes the loop: campaign outcomes can become context for future creator evaluation and planning.

AI at decision
points

I didn't want AI to become another place to go.

I prototyped AI assistance where the workflow contains translation or evaluation work — not as a chatbot sitting beside the product.

Turn an unstructured brief into something usable

A Brand can start with an unstructured campaign brief.

The prototype extracts structured requirements including objective, platform, deadline, deliverables, talking points, CTA and hashtags.

But those fields remain editable, and the campaign isn't created until the user reviews and confirms them.

The prototype extracts structured requirements including objective, platform, deadline, deliverables, talking points, CTA and hashtags.

But those fields remain editable, and the campaign isn't created until the user reviews and confirms them.

Use those requirements to narrow discovery

Campaign context can then inform creator discovery. Atelier first applies the visibility boundary, then search and filters narrow that eligible pool further.

AI matching operates inside that pool rather than gaining access to creators the Brand wasn't allowed to discover in the first place.

The prototype makes that narrowing visible — from Agency roster, to Brand-eligible pool, to campaign fit — before presenting ranked matches.

Matches expose reasons and cautions, not just a score.

The user still chooses who to evaluate, shortlist or request.

Use those requirements to narrow discovery

Campaign context can then inform creator discovery. Atelier first applies the visibility boundary, then search and filters narrow that eligible pool further.

AI matching operates inside that pool rather than gaining access to creators the Brand wasn't allowed to discover in the first place.

The prototype makes that narrowing visible — from Agency roster, to Brand-eligible pool, to campaign fit — before presenting ranked matches.

Matches expose reasons and cautions, not just a score.

The user still chooses who to evaluate, shortlist or request.

Check content without taking over approval

When content is submitted, the prototype can check it against campaign requirements such as product mention, CTA, hashtags, disclosure and talking points.

Check content without taking over approval

When content is submitted, the prototype can check it against campaign requirements such as product mention, CTA, hashtags, disclosure and talking points.

Turn campaign results into questions worth investigating

Once a campaign is live, the interpretation problem changes.

Instead of asking “Who should we work with?”, the Agency and Brand are now asking:

What actually happened — and what should we learn from it?

Atelier Insights uses measured campaign performance to surface patterns that may be worth investigating: a creator driving a disproportionate share of attributed visits, a content format producing a stronger click rate, or conversions concentrating among only part of the campaign.

Turn campaign results into questions worth investigating

Once a campaign is live, the interpretation problem changes.

Instead of asking “Who should we work with?”, the Agency and Brand are now asking:

What actually happened — and what should we learn from it?

AI can reduce the work required to reach a decision without taking ownership of the decision itself.

I separated those insights from the underlying metrics deliberately.

Performance data describes what happened. AI suggests what may be worth noticing.

Each insight can expose the evidence behind it, and the system labels the interpretation as advisory.

What I deliberately didn't build

Scope was part of the product decision.

A product spanning agencies, brands and creators can expand almost indefinitely.

I kept several adjacent systems outside the core prototype.

No open marketplace

Discovery stays inside an agency-controlled roster rather than becoming global creator search.

No payment processing

Atelier tracks contracted, scheduled and paid states without becoming a financial product.

No standalone inbox

Campaign decisions stay attached to the work they affect. For communication outside a structured action, I'd explore lightweight messaging within the campaign context rather than creating a separate communication system.

No autonomous AI

AI can structure, rank, flag and surface patterns in performance data. It doesn't approve content, select creators, judge campaign success or commit campaign decisions on someone's behalf.

These weren't extra surfaces I needed in order to prove the coordination model.

A feature had to strengthen the model, not just make the product larger.

What I still don't know

A working prototype can test logic. It can't validate the operating model.

Building Atelier far enough to make the workflows interactive helped expose assumptions that polished static screens wouldn't have.

It didn't validate those assumptions.

There are five things I'd want to test next.

Does the attention hierarchy match how real agency teams triage work?

The current priorities are product hypotheses. I'd want to observe how agency teams actually decide what deserves intervention first.

Do people trust AI-assisted matching when they can see its reasoning?

Atelier exposes reasons and cautions rather than only a score.

But explainability isn't the same as trust.

What happens when a campaign stops rather than progresses?

The prototype models movement through the lifecycle much more deeply than cancellation, termination or partial completion. Those states could expose assumptions in payments, deliverables and campaign roll-ups.

Does role-scoping hold up as agency teams become more complex?

The current Agency role is intentionally broad. A production system would need to test how ownership and permissions divide across a real organization.

Can the performance layer support useful comparisons without overstating attribution?

The prototype distinguishes tracked campaign activity from observational signals, but a production system would need to validate which data sources teams actually have access to, how consistently attribution can be measured, and which comparisons are useful enough to influence future casting.

Those aren't screens I'd quietly fill in before testing.

They're the places where I'd want the next version to learn something.

The Takeaway

Designing the interface meant deciding where the complexity should live.

Atelier started as a creator-management concept.

Building it made the harder problem clearer: a simple interface doesn't make the underlying workflow simple.

The complexity still has to live somewhere.

I chose to put more of it into the product model:

one shared campaign system instead of three disconnected versions,
assignment-level state instead of flattened campaign status,
derived attention instead of another field to maintain,
visibility rules before discovery,
measured outcomes separated from interpretation,
AI inside decisions rather than above them.

That allowed the three interfaces to become less alike as the system underneath them became more coherent.

The Agency can operate the whole portfolio.

The Brand can make the decisions it owns and understand what those decisions produced.

The Creator can focus on what comes next.

All without requiring the campaign underneath them to become three different things.

The interface could stay simple because the product underneath it wasn't pretending the workflow was.