Rethinking Infrastructure Efficiency

Infrastructure efficiency is the process of identifying and recovering unused computing resources across Meta's infrastructure. The work saves hundreds of millions of dollars annually, but requires coordination across multiple teams, workflows, and decision makers.

Before designing anything, I mapped the entire efficiency lifecycle and the three user groups responsible for it. Understanding who did what, when, and why became the foundation for every design decision that followed.

The implication wasn't “build a single unified view.” It was “make each person's experience feel unified to them.” That's not a tool consolidation. That's a new product category — and it changed everything about what we were building.

A home screen, not a dashboard

The mental model that unlocked the team's alignment: an iPhone home screen. All the apps exist. You decide which ones go where. The system suggests sensible defaults based on your role — but you control your view.

Early conversations naturally converged on a unified dashboard. The problem was that dashboards are built for monitoring — they work well when users come to read information. But users weren't opening the efficiency tools to consume reports. They were opening them to complete work: find an opportunity, submit a win, clear a compliance item. A home screen shifts the frame from reporting to action.

This became the foundational product principle: modular, configurable, persona-aware. Rather than designing a single page, I designed a system of configurable content blocks — a to-dos surface, a metrics summary, an opportunities module, a wins tracker, a review queue, and more. Each widget is self-contained and can be surfaced, collapsed, or repositioned based on the user's role and preferences.

Persona-aware defaults, not a blank canvas

Different user types start with different configurations. Individual engineers see their assigned work and pending submissions. Infrastructure budget owners see compliance status and org-level metrics. Platform operators see fleet-level signals. Users customize from there — the defaults reduce “blank canvas” friction without removing flexibility.

Full customization was the obvious alternative — give users complete control and let them build whatever they need. But concept testing made the tradeoff clear: users didn't want to configure a tool before they could use it. They expected the system to already understand their role. Persona-aware defaults deliver immediate value while preserving the flexibility to adjust.

Contextual AI assistance: meet users where they are

Finding the right work was only half the problem. The same insight that drove the homepage — that relevance means meeting each user in the context of their actual work — applied just as much to workflow completion. Once users identified an opportunity, completing it still required navigating complex workflows: submitting a win, calculating dollar impact, tracking a review, understanding an unfamiliar metric. Each step meant switching context or recalling process details from memory.

A standalone AI destination was on the table — a dedicated interface where users could query the efficiency system. But research had consistently surfaced workflow fragmentation and context switching as the core friction. Sending users to another tool, even a smarter one, would have recreated exactly what we were trying to solve. The assistant needed to be embedded in the experience, not alongside it.

I introduced Efficiency Assistant — a persistent, context-aware conversational widget embedded throughout the experience. The first version targeted four JTBDs at the intersection of high frequency, high friction, and high business impact:

• Record a win — guided submission without opening the full form flow
• Review wins — surface evidence and an AI-generated verdict so reviewers can decide without switching tools
• Summarize wins this half — on-demand summary of submitted and approved work
• Answer page questions — explain metrics, surface context, clarify status in place

AI-assisted win submission

Engineers often already had the evidence — an IsItFaster experiment had run, results were in, contributors were known. But translating that into a review-ready submission still required manually calculating impact, attributing contributors, and completing structured metadata. I explored how the assistant could handle that translation: paste an experiment link, and it retrieves the metadata, calculates annual savings, identifies contributors, and drafts a submission ready for review.

Why this mattered

The challenge wasn't content generation — it was converting infrastructure evidence into standardized business records. The assistant reduced that workflow into a guided conversation, with every field traceable to a source and every decision remaining with the engineer.

This was the same principle as the homepage applied to task completion: don't send users somewhere else to do their work. Bring the right capability to where they already are.

The to-dos widget: the right first proof of concept

Rather than building everything at once, I scoped the initial development phase around a single high-impact widget: a configurable task surface showing each user what they need to do next.

Other starting points existed — the opportunities module, the review queue, the AI-assisted workflows. But the to-dos widget was the only one that cut across every persona. Engineers had pending submissions. Budget owners had compliance tasks. Platform operators had blockers to clear. No other module had that kind of universal relevance.

This was the right first build — high value, bounded scope, and a direct answer to the most consistent research finding: “I don't know what to do when I open this.” The to-dos module pulls pending work from across the efficiency ecosystem — compliance items, efficiency wins needing review, high-value opportunities worth acting on — into a single actionable list. No tool-switching required.

It also established a proof of concept for the widget architecture before we committed engineering resources to building the full system.

A single unified dashboard. The temptation was real — a big combined view would have looked impressive in a demo and satisfied the original brief. But it would have failed in production for the same reason the original framing was wrong: different users need fundamentally different things, and designing to the average serves no one.

The research gave me the evidence to hold that position. Without it, the organizational pull toward a simple, demo-ready deliverable would have won.

When I presented to senior leadership, they didn't ask to revisit the consolidation brief. They adopted the strategy. Engineering committed to building the full system — not a proof of concept, not a phased pilot — the whole thing. The team was genuinely excited. That kind of enthusiasm doesn't happen for a homepage redesign.

What shipped was Meta's first personalized experience in the capacity and efficiency ecosystem. Nothing like it had existed before. And because it established both the architecture and the design principles, it was positioned as the model for how future projects in the efficiency org should be built.

None of that happens without the conditions I created before a single pixel was designed. I wrote the strategy document that gave six stakeholders clarity on scope and ownership. I facilitated the cross-functional kickoff across two product organizations. I kept the project anchored to user needs when organizational momentum pushed toward a simpler deliverable. I co-authored the research plan, wrote the interview scripts, recruited participants, and ran the concept testing end-to-end.

The hardest part wasn't the design. It was making the case — early, consistently, with evidence — that the original framing was wrong, and that solving it right required more upfront investment than anyone had planned for.

My year-end manager review: “Set north star vision for Efficiency product ecosystem, securing H1 '26 commitment and presented to senior leadership to get buy-in.” The north star vision became the shipped product.