Budget Dashboards & AI Assistant · Angela Lanza Vassallo

The Problem Space

Finance Business Partners (FBPs) and org leaders managed multi-billion-dollar spend across OpEx, headcount, and travel without dedicated tooling, manually stitching together Concur exports, Workday ledgers, and FP&A data in spreadsheets. Budget coverage sat at just 0.2%.

The work came in two parts: build a unified real-time tracking foundation, then layer AI on top of it to answer the routine financial questions that were consuming hours of manual effort.

The goal

Give org leaders and their delegates a budget view they can act on themselves, accurate enough to trust and simple enough to use without a finance background, and make it work across every category of spend at Meta.

Who We Designed For

Finance Manager

"I need to see the full financial picture across every pillar to catch risk early."

Org Leader

"I'm accountable for my budget, but I don't have time to become a finance expert."

Org Delegate

"I manage day-to-day budget details while my leader needs summaries."

The Approach

Budget accountability at Meta flowed hierarchically, moving from company-wide down to pillars, org leaders, and individual positions, trips, or purchase orders. I designed every surface to mirror that exact shape.

Whether monitoring the whole company or a single pillar, users operated at different altitudes within the same mental model. That consistency mattered beyond navigation: an AI assistant is only as trustworthy as the structured data underneath it.

How I Worked

Each tracker (Travel, OpEx, Headcount) operated as an independent pod with its own PM, Finance partner, 5 to 7 engineers, and shared UX research support. As the sole consistent design thread across all three, I drove cross-surface consistency in practice.

While another designer originated early concepts for Travel and Headcount, I independently designed OpEx from 0 to 1, establishing the component patterns and system foundation used to elevate Travel and Headcount in Phase 2.

What I Changed

Travel and Headcount had a Phase 1 before I took them on. It proved the demand and exposed four problems, and each one set the direction for what I built next.

Phase 1

What I changed

Phase 1

Leaders did not trust the numbers

Dense tables left executives questioning accuracy, with no story in the data.

What I changed

Digestible without losing detail

Visualizations for leaders who are not finance experts, and tables kept only where the detail was the point.

Phase 1

There was nowhere to go deeper

A rushed MVP used modal pop-ups, so there was no real detail behind any number.

What I changed

Dedicated detail pages

Real pages with consistent drill-down, so any figure traces back to its source.

Phase 1

There was no system to build on

One-off patterns meant every new spend category would need its own mental model.

What I changed

One paradigm, three trackers

Every data visual standardized, from variance framing through drill-down.

Phase 1

Answers took digging

Finding a specific number meant hunting through tables and filters.

What I changed

Ask the page

An assistant scoped to on-screen data, with every answer verifiable against the table beside it.

The Phase 1 travel budget tracker: a filter bar, a single summary bar, and two dense tables of budget against actuals, with Export to CSV as the only way to go further

Illustrative data

Phase 1, as it shipped before I took it on. Everything is a table, and the only way to go deeper is Export to CSV.

The Phase 1 modal pop-up showing a pillar’s travel detail, with five summary figures and a short list of trips above a Close button

Illustrative data

The one level of detail Phase 1 offered, in a modal you could not link to, filter, or drill past.

The Work

NDA note: Screens for this project are confidential under NDA. Everything shown here was rebuilt from memory to convey the design decisions. Every figure, name, org, and data point in these screens is fabricated and does not reflect real Meta budgets, headcount, financials, or organizational structure.

Travel Budget Tracker (Phase 2)

Delivered 8 design enhancements and 11 new capabilities focused on data visualization and resolving reporting discrepancies.

"This is looking really good... you guys should package it up and sell it to other companies."

VP of Finance, Meta

The full travel budget tracker page, from the summary cards through the monthly chart, top spend, and the spend by organization table

Illustrative data

The travel tracker end to end, at full page length.

Animation of the travel budget tracker drilling from the company-wide overview into the Reality Labs pillar page, then into that pillar’s trip ledger

Illustrative data

Drilling from company-wide travel spend into a single pillar, then down to its individual trips.

OpEx Budget Tracker (0–1)

Led the 0 to 1 design of a comprehensive budget owner dashboard, streamlining planning and increasing expense visibility by roughly 22 percent.

The full OpEx budget tracker page, showing what the total spend is made of, the monthly chart, and budget consumption by pillar

Illustrative data

The OpEx tracker end to end, at full page length.

Animation of the OpEx budget tracker drilling from the company-wide overview into the Infrastructure purchase order ledger

Illustrative data

Drilling from total operating spend down into a pillar’s purchase orders. The Budget AI flag that appears here was a proposed direction, not something that shipped while I was on the team.

Headcount Budget Tracker (Phase 2)

Delivered Phase 2 enhancements covering approximately 40.4 percent of Meta's budget. To close a major data gap, I introduced headcount debt, tracking positions still open after a plan committed to closing them as a first-class metric alongside Filled and Unfilled roles.

The full headcount budget tracker page, showing where projected cost sits today, projected cost versus budget by pillar, and open requisitions by age

Illustrative data

The headcount tracker end to end, at full page length.

Animation of the headcount budget tracker drilling from the company-wide overview into the Reality Labs headcount detail page

Illustrative data

Drilling from company-wide headcount down into one pillar’s org leaders and reporting lines.

Component Standardization

Partnered closely with design and engineering to unify the tool experience across all three surfaces into a single consistent shell and navigation pattern. The same interaction patterns carry across every tracker: one metric card that opens its own composition, one control for re-cutting a card by a different dimension, and one hierarchy that expands from the whole company down to a single manager.

The left navigation shown collapsed to an icon rail beside its expanded state, with Home, Travel, OpEx, Headcount, Insights, Settings, and Collapse

One navigation shell across Travel, OpEx, and Headcount, collapsible to an icon rail.

Status pill taxonomy showing budget status pills used on every dashboard, commitment lifecycle pills used in the travel and OpEx ledgers, and position state pills used in headcount

One status vocabulary, scoped by what it describes: budget health everywhere, commitment lifecycle in the ledgers, position state in headcount.

The metric card row shown for all three trackers: travel over budget, OpEx under budget with a positive delta on the projection card, and headcount in the dual-unit variant carrying dollars and positions together

The same five cards across all three trackers, over budget and under. Headcount carries both readings at once; the unit switch only decides which one leads.

Three summary cards from different trackers, each with its own breakdown and a footnote citing the system the numbers came from

Illustrative data

One card pattern across three trackers, each citing the system its numbers came from.

The committed versus spent monthly chart shown twice, once counting dollars and once counting trips with no budget line

The same monthly chart counts dollars in one tracker and trips in another.

Animation of a metric card being hovered to reveal its composition: the total broken into spent, committed, and still to book, with the source ledger and refresh time

Any metric card opens its own composition, down to the ledger the number came from.

Animation of the top travel spend card being re-cut from a breakdown by org leader to a breakdown by destination

Illustrative data

The same card re-cut by a different dimension, without leaving the page.

Animation of the spend by organization table expanding from company level down through pillar, org leader, and individual manager

Illustrative data

One table expands from the whole company down to an individual manager.

The filter bar showing a search field, three scoped dropdown filters, and a right-aligned count of how many records are being shown

One filter bar, scoped to whatever the page lists, always saying how much of the set is in view.

AI Budget Assistant

Co-led strategy and design for an in-workflow AI assistant, which shipped while I was on the team. Because financial decisions carry high consequences, the interaction model prioritized trust and transparency through several principles:

Show Your Work: Responses broke down reasoning by org with the exact underlying numbers, and cited the system each figure came from.
Actionable Outputs: Answers surfaced direct next actions, such as opening an affected org or modeling a fix.
Scoped to the Page: The assistant answered only against the data in view, so any number it returned could be checked against the table beside it.

Animation of the Budget AI panel opening beside the headcount tracker, taking the question which orgs are most at risk of missing their FY2025 headcount budget, working through its steps in view, then answering with a breakdown by pillar, the systems each figure came from, and next actions

Illustrative data

The steps it takes are visible while it works, and the answer arrives with the systems every figure was read from — checkable against the table beside it.

A Budget AI thread shared by two people: a Finance partner raises a projection gap, an Infrastructure leader mentions Budget AI to ask what is driving it, the assistant answers with sources, and the two of them agree on a fix

Illustrative data

Finance and the org lead working one thread together, pulling Budget AI in by name.

Animation of the Budget AI popover on the travel tracker answering where to host a three-day offsite, comparing cost by location against where the team already sits

Illustrative data

Not only variance questions: a forward-looking one, answered from the same page data.

Animation of the same Budget AI conversation moving between its three surfaces, from the side panel to a popover to full screen, using the display switcher

Illustrative data

One conversation, three surfaces: alongside the page, over it, or given the whole screen.

What Testing Changed

I concept-tested with all three personas. Three findings changed the design.

In OpEx, “committed” meant two different things. Some orgs counted committed spend as officially removed from remaining budget. Others treated it as money set aside but not yet deducted. Every card now states its own definition, and a third category that does not deduct from remaining budget was planned next.
In Travel, leaders track by trip, not by budget line. Org leaders wanted spend organized around the trips themselves, with a way to see who and what was driving each one. That became switchable rankings, a summary page that expands each trip, and a per-trip ledger down to every cost item.
In Headcount, debt payback needed more than a total. Org leaders could see the number owed but not who had paid back what, or when. I expanded the focus view to show payback by owner and by date.

Outcomes & Retrospective

−44%

Reduction in travel reporting effort

3,000

Hours saved in ad-hoc reporting

8.6K–33K

Projected annual hours saved by the AI assistant

40.4%

Of Meta’s budget with headcount data gaps closed

+22%

Expense visibility for OpEx budget owners

53→75%

Trust in travel data, Phase 1 to Phase 2

A blueprint, not a one-off. The tracker continues to be the model for budget tracking across other budgets at Meta.
Back into the design system. The components built here were in the process of being added to the enterprise design system used across all internal products.

Keep Doing

Building a component library for product-specific elements to hold consistency. Leveraging AI inside the solution so people can decide and act faster. Designing for trust in both the data and the AI responses.

Improve

Find ways to make this a more personalized experience per persona and per organization. Stress-test terminology with every organization earlier in the process. Incorporate AI workflows from day one rather than layering them on later.

Where I Left Off

Before leaving Meta, I was co-leading exploration into an AI-first budget transformation: proactive anomaly detection that would flag a risk like a bad accrual before anyone thought to ask, autonomous budget management, and personalized dashboards tailored to how individual leaders plan.

I designed this work in code rather than Figma, using Claude Code inside VS Code, and merged my own changes into the product. Working directly in the running system meant every exploration was real and testable the same day, with no gap between a mockup and something the team could actually put in front of a budget owner.

The Arc

This work sits at the intersection of making complex systems feel simple and operating at the strategic layer where product direction is set. Building these budget trackers from scratch required deep engineering collaboration and close partnership with Finance. Layering high-stakes AI trust models on top of that foundation represents some of the most forward-looking work of my career.