P / 01
Make the consequence visible
Every lever shows its downstream impact immediately. Nobody should approve a number they can’t feel.
Case study/001Enterprise · Workspace planning
Org Space
Manager
The story of how a 3-person team built and shipped Meta’s first workspace management tool in 6 months.
Overview
Org Space Manager is Meta’s first space planning and allocation tool — built to give leaders across organizations real-time insight into how employees use workspaces, and the confidence to make decisions that lift team productivity and workspace utilization.
Planning that is sprawled across spreadsheets, decks, and multiple internal tools now happens in one place: leaders see headcount, desk inventory, and policy impact update as they plan, and every decision moves through a single, visible workflow.
32
Campuses from Menlo Park to Singapore
16
Organizations
~45,000
Employees
1
Source of truth for the whole cycle
01 The research
Sitting with both sides
of the process
Before thinking about a solution, I partnered with our UX researcher and sat in ~15 study sessions with org leaders from the company’s four biggest orgs. We mapped out how planning really works today, and looped in the planners to understand every hand-off between them and the leaders.
Designer + UX ResearcherPaired through every session
4 organizationsFamily of Apps · Instagram · WhatsApp · Enterprise Engineering
~15 study sessionsAcross the four orgs
01Map the current processWalk through every step an org leader takes today to plan and allocate workspace for their employees.
02Listen to the problemsCapture the challenges and friction points leaders face, in their own words.
03Loop in the plannersTrace every step where leaders and planners interact — and the back-and-forth each hand-off creates.
02 The problem
A planner’s process,
an org leader’s burden
Org leaders had to navigate an overly complex process built around the planners — waiting weeks for answers while decisions bounced between tools, threads, and teams.
- 01
Built for planners, not leaders
Every allocation cycle forced org leaders through an overly complex, planner-defined process — lengthy wait times and excessive back-and-forth between the two parties.
- 02
Scattered workflow and data
The workflow and its data lived across multiple spreadsheets and internal systems, so every step meant manual reconciliation — slow and quietly error-prone.
- 03
Conversations everywhere
Communication was scattered across different channels, making discussions hard to follow and decisions hard to trace back to their context.
- 04
Misaligned decisions
Without one shared thread or source of truth, team members often reached conflicting conclusions — and plans moved forward misaligned.
03 The process
One prototype,
one source of truth
After plenty of trial and error, I landed on a new design framework that let the engineers and me co-develop the same product, minimizing manual design work and speeding up the delivery timeline.
Before · the classic framework
Research
Figma mocks
Testing
Stakeholder demo
Eng build
The design often got stuck in this loop for a long time before getting to engineers.
After · the AI-first framework
DesignerPM
Synthesize research insightsWith my PM, I ran every research insight through Claude: clustering them into themes, extracting the key problems, and mapping the user journey with its pain points and opportunities.
Feed the AI agentsThe agents took the pain points and key problems straight from research, plus the user stories, and turned them into accurate working UI with minimal manual Figma work.
Create PRD & user storiesTogether we turned those opportunities into concise user stories and a PRD, a backlog that traces straight back to observed pain.
Designer
Iterate front-end
Usability test
Stakeholder demo
One prototype · the single source of truthEvery discipline contributes to the same living prototype, one shared state from research to build.
Engineers
Back-end development
Review front-end
04 Principles & framework
How I made the
complexity decidable
Six principles kept a dense, policy-heavy domain honest while the product came together.
P / 02
Treat policy as an object
Policy is the thing you design — surfaced and editable — not a setting buried three menus deep.
P / 03
One spine, many contexts
A single plan model scales across every campus and org, instead of a bespoke sheet per quarter.
P / 04
Guardrails, not gates
Surface risk inline — a projected deficit, a broken threshold — and keep the human in the decision.
P / 05
AI as a co-pilot
The assistant answers across plans and recommends — it never silently automates the call.
P / 06
Design for the approval
Shared, legible state from draft to live, so authors and approvers read the same picture.
02
Frame
Defined the plan as a first-class object and a five-state lifecycle everyone could see.
03
Systematize
Built the policy + assessment model and a status system that scales to 9 × 4 contexts.
04
Validate
Modeled projected impact and pressure-tested it against real deficits and edge scenarios.
05 The solution
All plans,
one shared surface
Org Space Manager centralizes every plan into a single surface — org leaders see the current state at a glance, review impact as they plan, and track each plan through its entire lifecycle.
Plans dashboard
Designed to be scanned
Campus groups set the rhythm, paired utilization bars give each org a visual fingerprint, and colour-coded status chips — Live, Approved, Submitted — carry the whole lifecycle in a single glance.
Plan detail
A three-panel canvas
Policy controls on the left, the live assessment in the centre, projected impact on the right. The composition does the explaining — move any lever and numbers, warnings and guardrails respond across the canvas instantly.
Smart recommendations
One strict card anatomy
Every AI-drafted strategy card shares the same anatomy — title, approach, then outcomes as colour-coded pros and cons — so three options compare at a glance, and “Preview changes” keeps the final call with the human.
06 The deliverable
A shipped planning surface
Three connected surfaces — the dashboard, the plan, and the assistant — running on one model.
Weeks → days
Planning cycle, once the model did the reconciling instead of a person.
Caught pre‑approval
Desk deficits surface while the plan is editable — not after it’s signed off.
1 shared state
Authors and approvers read the same draft-to-live picture across every campus.
Outcome figures above are illustrative placeholders for this portfolio — replace Weeks → days and the rest with your measured results.