← back to work case study · service design

Five tangled boards, read as one lifecycle.

A seven-person art studio had never mapped how it works as a whole. I sat with each department, drew its flow, and found five processes that barely connected. I rebuilt them into one student lifecycle over a single shared record, an MCP server every team reads and writes, so a student is entered once instead of re-typed and handed off by hand. A self-hosted AI sits on top and drafts the busywork; a person signs anything that leaves the building.

Client A small art studio, Vancouver- and China-facing, that helps students from around the world build the art skills and the portfolios they submit to world-class art and design schools. Identity withheld; operational details abstracted.
Sector Creative education, student services
My Role Service designer. Research, service blueprinting, and the systems architecture behind it, from the shared record to the AI on top.
Team Solo, working with the studio's director and its department leads.
Duration 2026. An as-is read, a to-be redesign, and a phased build plan.
Methods Stakeholder interviews and mapping, workflow mapping, service blueprinting, adversarial design critique, systems design for a shared record with AI on top
Outputs An as-is map, a to-be service blueprint, a shared-record architecture on an MCP server, a self-hosted AI layer on top, a dashboard per role, and a phased build roadmap.
Status An active engagement, mid-build. Studio anonymized pending its sign-off; sensitive operational specifics abstracted.

I drew how the studio works. Almost nothing connected.

Five accurate department boards that barely touched: a student's name re-typed at every stage, handoffs left to memory.

Before I could redesign anything, I had to see how the studio runs. So I sat with the director and each department lead and drew it out. Five large boards came out of it, one for each side of the business: how leads become students, how a student is enrolled and coached, how mentors deliver their teaching, how money moves, and how the studio finds the next student. Each board was detailed and accurate on its own. Side by side, they barely touched.

A student's name was typed fresh at every stage. A handoff between two teams was a message someone remembered to send. The same approval waited on one role in five different places. None of it was wrong. But a studio is really one path a student moves along, and running it as five separate machines left gaps for a name, a payment, or a handoff to fall through.

as-is · the real board

The original as-is enrollment board: eight colour-coded department lanes running left to right across five phases, dense with boxes and hand-drawn connectors, shown too small to read any single step.
Figure The board as I first mapped it, the real thing with only a client label blurred, shrunk until the words go too small to read. That is on purpose: its job is to show the tangle was real rather than to be read.

Once I followed a single student, the boards became one path.

One student's arc became the spine, and every department was rebuilt against it in a strict order.

I stopped treating the boards as five processes and traced the one thing that runs through all of them: a single student. They arrive as a lead, sit a consultation, get assessed, decide, enroll, get onboarded, get coached for months, apply to schools, and graduate. If they do well, their story becomes the marketing that brings in the next student, and the arc loops back to where it began.

So I fixed that arc as a spine and held it still, then rebuilt each department against it, one at a time. I kept to a strict order: cut the steps that should not exist, simplify what is left, automate, and only then connect the systems onto one shared record. Automating a broken step only makes a broken step faster.

I redrew the five boards as one.

The same departments and phases as the board I found, now organised into one flow over a single shared record.

The rebuild keeps every step but untangles the connections. Blue is the automated spine that carries one record the whole way. Gold marks the two things I added: the zones I tightened, and one shared record on an MCP server beneath every lane, carrying the student from one department to the next automatically.

to-be · the redrawn board

results loop back inquiry proposal contract & pay onboarding follow-up student sales operations education marketing finance director inquire decide enroll join platforms share results line of interaction triage proposal contract referral intake one lane · was three teams record created state machine handoffs fire assessment development plan notes, feedback content, publishing payment · reconciled exceptions queue routes by rule, not the inbox line of internal interaction MCP · one source of truth every lane reads and writes the same record, so a student is entered once records documents messages scheduling

→ scroll the board sideways

Figure The to-be, redrawn at the same scale as the board I found: the same lanes and phases, the sales teams merged into one, the director's inbox replaced by rules, and one shared record on an MCP server added beneath every lane. Blue is the automated spine that carries one record the whole way, plus the loop back into the next inquiry; gold is what I tightened and added.

I designed each department, then tried to break it.

Every redesign went back to the people who run the work, to be pulled apart before I trusted it.

A redesign that only its author has read is a guess. So every department went through the same loop. I drafted the new flow, then took it back to the director and the lead who runs that work and asked them to find everything wrong with it. Those critiques were uncomfortable and worth it.

One pattern kept surfacing. I would call a bottleneck solved when I had only moved it. Approvals for nearly everything funneled to one role, so I routed the routine ones away, then noticed I had rebuilt the same queue on a different screen. The fix was a rule that defines which decisions need that person, and a number on their own dashboard that shows when they are the thing holding work up. I kept the critiques in the work instead of hiding them.

Six departments, rebuilt one at a time.

Each of six lanes shows what it carried and the single move that changed it, before they recombine into one blueprint.

Holding the spine still, I rebuilt the departments one at a time, in the order the reframe set. The five boards became six lanes: I pulled the director's approvals, which had slowed all of them, into a lane of their own.

sales
eliminatesimplify
as I found itone approverthree teamsConversion had no home of its own. The work was spread across three teams, and every deal waited on one person's calendar and judgment.
what I changedexceptionsdirectorone lanepricedI gave it one accountable lane and replaced a by-feel price with a documented range, so only the exceptions still need the director's sign-off.
operations
simplifyintegrate
as I found itenrollcoachbillre-typed by handThe same student was typed in fresh at every step, and each handoff depended on someone remembering to send a message.
what I changedauto-handoffsone recorddeliveredsafety netOne record now carries the student end to end. Finishing a step triggers the next team's work on its own, with a safety net that catches anything that stalls.
education
automate
as I found itmentorhours by handMentors spent hours writing up session notes, progress reports, and essay feedback by hand, time taken straight from teaching.
what I changednotes, reports, feedbackAI draftsmentor signsportfolio critiquehumanThe model drafts those now; the mentor edits and signs them. The desk critique and the read on a portfolio stay entirely human.
marketing
integrate
as I found itresultunusedA student's strong result rarely turned into anything. The studio's best evidence of its own work mostly went unused.
what I changedresultcase studycontentnew inquiryI wired the growth loop, so a graduate's outcome becomes a consented case study, then content, then the next family's reason to inquire.
finance
simplifyautomate
as I found itpaymentmatcherrorsMoney moved by hand, and matching each payment back to a student was slow and easy to get wrong.
what I changedrecordreconcilerelease · humanI introduced a clean, auditable flow: amounts computed from the record, payments reconciled by an exact key, and nothing released without a person.
director
eliminate
as I found itdirectorEvery gate ran through the director, so the whole studio moved at the speed of one inbox.
what I changedmost decisionsone queueexceptionsby rulehandledI defined which decisions need them. The rest route by rule, and the few genuine exceptions land in one approval queue they can clear in a single pass.
Figure Every lane ran through the same order. The teaching and the relationships stayed where they were; only the administration beneath them moved.
The hard part

The hardest problems sat in the boundaries between departments rather than inside any one of them. Two teams would each assume they owned the same proposal, invoice, or case study, and the work stalled in the overlap. I settled every case with one rule: one team owns the artifact and the call to send it; another owns the mechanism beneath it, the templates and the shared record. Once that rule was written down, no two lanes claimed the same job twice.

The whole service on one shared record.

Four phases run across the top; roles and systems run down the side. Everything works from one record, and each backstage step is tagged human, auto, or AI-drafted for a person to approve.

The redesign is one blueprint. Across the top, the four phases a student moves through. Down the side, the people and systems that serve them, all reading and writing one shared record instead of copying it between each other. Every backstage step carries a tag, so the shape reads at a glance: what stays human, what runs on its own, and what an AI drafts for a person to approve. The dashed rules are the blueprint's own borders, between what a family sees and what they never do.

Unless tagged, a backstage step is AI-assisted: the model drafts, a person approves. human judgment, kept human auto runs on its own, reversible
attract & convertinquire → enroll
onboardset up
develop & delivercoach → apply
advocateloops back
student
inquire, decide, enroll
join the platforms
coaching, upload work, apply
share results, refer a friend
line of interaction
frontstage
consultation & assessment human
orientation human
coaching & portfolio critique human
results follow-up
line of visibility
sales
triage, proposal, contract
·
·
referral intake human
operations
CRM record & scheduling auto
onboarding pack
record state machine auto
·
education
assessment write-up
development plan
notes, reports, essay feedback
case-study draft
marketing
·
·
·
content & publishing
line of internal interaction
source of truth
recordsdocumentsmessagesscheduling One MCP server every lane reads and writes. The self-hosted AI reads it too, and drafts behind a human gate.

→ scroll the blueprint sideways on a narrow screen

Figure The to-be service at phase altitude. Every lane works from one shared record instead of re-typing it, so a student is entered once. The moments that earn the fee stay human; the busywork beneath them goes to an AI a person still signs off. Finance runs underneath every phase from enrollment on, and the director sits above it, handling the exceptions.

The AI reads the record and drafts the busywork.

One rule at the center: nothing irreversible happens without a person approving it. Student data stays on models the studio hosts itself.

The studio already used AI in the corners, informally. The redesign gives it one place to work, the shared record, and one rule: a person approves anything that sends, pays, or publishes. From that record the model drafts the proposal, the progress report, the invoice, the reminder. A person reads each one and commits it. Nothing irreversible fires on its own.

The work that touches a student's records runs on a model the studio hosts itself, on its own infrastructure. That keeps a young person's data where it belongs and where the law expects it to be. The teaching, the critique of a portfolio, the read on whether an idea has landed: I never put those on the table to automate. The studio sells exactly that.

A legend from the studio's own workflow board: a black lightning-bolt icon labelled 'don't know if automated or not but should be automated', and a green lightning-bolt icon labelled 'automated'.
Figure The studio's own key, off one of the boards. Black marks a step they thought should be automated; green marks one that already was. The boards are covered in black and almost no green. That gap is the work.
the shared record

One record per student on an MCP server: the source of truth every department and dashboard reads and writes. It replaces the copy-by-hand handoffs, so everyone works from the same state.

role dashboards

One screen per role, built on that record. The few numbers that drive an action, plus the one-click drafts the role needs.

the gate

Every send, payment, or publish waits here as a draft with a preview. A person approves it or sends it back, and a log records who did what.

connectors

Narrow, least-privilege links from the record to the systems the studio already runs. The model reads freely and writes only through them.

recordsdocumentsmessagesschedulingsocial stats
models

Self-hosted for anything that touches a student's data. A hosted model only for the rare hard problem on non-sensitive work.

Figure The system, top to bottom. Everything sits on the shared record, and the gate keeps a person on every risky action.

Every role gets one screen that says what to do next.

Each dashboard shows only the numbers that drive an action; every figure has to answer one question, what do I do about this?

Each team gets a dashboard scoped to the decisions that role makes, all reading the same record. Every figure has to answer one question: what do I do about this? The ones that cannot, I cut. A few carry a quiet signal the studio needed without having asked for it, like whether the work is creeping toward the limit of how many students the mentors can take.

Figure The sales view. Urgency reads through the accent and weight; there is no traffic-light palette. The starred signals tell the studio whether it is short of leads or short of capacity. Every other role gets the same kit, tuned to its own decisions.

The mistake I almost shipped.

The loudest signal was a worry heard at every intake: too many leads. I nearly built a faster funnel before checking whether the mentors were full.

The loudest signal in the studio was a single worry, raised again and again at intake: too many leads. I read it as the problem and started designing a faster funnel. The reviewer stopped me. A faster funnel only helps if the studio can teach the extra students it converts. If the mentors were already full, converting harder would fill a longer line and make the experience worse, which is the opposite of what this kind of studio sells.

So I held the redesign and asked the one question only the director could answer: are the mentors full? They were not, so the intake work went ahead. But I had come close to building hard for a constraint I never confirmed, on a hunch that happened to be right. Now I test the loud signal before I act on it.

What I handed over.

An as-is map, a to-be blueprint, one shared record on an MCP server, a self-hosted AI layer with a human gate, a dashboard per role, and a phased build plan.

The engagement is live and mid-build, so I will not report outcomes I have not measured. The design work itself, though, is done. An as-is map the studio recognized as itself. A to-be blueprint that holds together across every department. One shared record on an MCP server that every department reads and writes, with a self-hosted AI on top behind a human gate on every risky action, plus a dashboard for each role. A build plan sequenced so the cheapest, safest wins come first and earn the trust for the rest.

The one-lifecycle view held up against every department I rebuilt, and rebuilding against it changed what each one was for. The blueprint and the build plan are written to be handed off and run without me.

What I would do differently.

I would build the unglamorous prerequisites first, and design the visible parts second.

I would sequence the build the other way around. I designed the visible parts first and left the unglamorous prerequisites for last, when the build itself has to run in reverse: the studio needs a clean place to keep its money before any of the finance work is worth drawing.

On my first service-design project, I learned that most of the craft is restraint. Automating the studio was the easy part. The harder call was what to leave exactly as it is.