Bloomberg Tax's Planning Engine
Designed a system-generated tax planning engine, liberating Enterprise tax teams from managing calcs in fragile spreadsheets.
01 — The Overview
My Role: This project gave me my first blank-slate opportunity, designing a full product experience for tax professionals from the ground up. The original direction centered on a manual tool for managing asset calculations, but customer conversations revealed a bigger opportunity.
Collaborating with the PM, we shifted course and designed a tax planning experience that replaced fragile spreadsheets with a tool that automatically generates all permutations and helps teams finalize their tax strategy.
Product: Fixed Assets
Outcome: Replaced a spreadsheet-dependent tax modeling process with a tax planning generation scenario engine.
02— The Problem
Corporate tax teams manage thousands of assets had no reliable way to model depreciation scenarios at scale. Without a purpose-built tool, they were forced to manually build fragile spreadsheets in Excel, a time-consuming, error-prone process that limited them to evaluating only a handful of strategies at a time.
The stakes were high. Tax depreciation is one of the most consequential calculations a corporate tax team makes, involving thousands of assets, multiple election types, proposed regulatory changes, and year-end deadlines that leave no room for error. Getting it wrong isn't just inefficient. It's expensive.
And the person asking for the answers, the CFO, typically wanted multiple strategies, each of which could take weeks to prepare.
03 — The Research
What four enterprise customers told us
We conducted discovery interviews with tax teams at Amazon, Starbucks, Greenbrier, and Kwik Trip. Three findings shaped everything that followed.
EXCEL IS NOT A RELIABLE TOOL FOR PLANNING TAX SCENARIOS
Bloomberg Tax's Fixed Assets product didn't give customers a better option. When teams needed to compare bonus depreciation elections across asset classes or forecast depreciation totals against planned capital expenditures, they exported their data, built the model manually in a spreadsheet, and hoped they hadn't introduced an error somewhere in the process.
ASSET INVENTORY NEVER STOPS CHANGING
Enterprise customers acquire hundreds of thousands of assets within a single tax year. Every new acquisition meant recalculating depreciation scenarios from scratch.
THE CHIEF FINANCIAL OFFICER DRIVES THE STRATEGY, NOT THE TAX TEAM
When we asked why teams chose bonus versus non-bonus elections, the answer was consistent: the CFO requests multiple strategies and the tax team prepares them. That preparation, building each scenario manually, was taking weeks. Teams couldn't share more detail because the specifics were company sensitive, but the time cost was clear.
04 — The Pivot
The original product direction was a scenario builder. Set your parameters, toggle your assumptions, generate a result. Repeat as needed. It was a reasonable response to what customers asked for, the ability to adjust sub-calculations without starting from scratch each time.
The PM was convinced this was the right direction. I wasn't.
ChatGPT had just emerged. Microsoft CoPilot was on the horizon. I knew that a tool asking tax professionals to manually construct scenarios one at a time would have a shelf life of less than a year before a CoPilot integration made it obsolete. I brought this argument to the PM. It took multiple conversations.
ALIGNMENT MOMENT
I made a case that if we build a manual builder, we're solving yesterday's problem. If we build a system that generates every possible strategy automatically, we're building something CoPilot can't easily replace, because the value isn't the AI, it's the relief and confidence we can deliver to tax professionals.
She came around. We changed direction.
05 — The Insight
Tax depreciation planning has a finite combinatorial structure. There are five asset class lives. Each class life has two election types: bonus and non-bonus. The total number of possible depreciation strategies equals 2 to the power of 5, which is 32 permutations.
Every enterprise tax team was evaluating a subset of those 32 strategies manually. We could generate all of them instantly.
The design question stopped being "how do we help tax professionals build scenarios?" It became "how do we help them navigate 32 strategies and arrive at the right one?"
06 — Exploring Design Directions
The first iteration followed the original PM direction, a manual scenario builder where users set parameters and generated results one at a time. It solved the stated problem. It didn't solve the real one.
SCENARIO BUILDER
The planning engine replaced it entirely. Instead of asking tax professionals to construct scenarios, the system generates all 32 permutations automatically and surfaces them in a clean comparison interface. Teams can filter by election type or enter a target depreciation total set by their CFO and see which strategies get them there.
No toggling between assumptions. No exporting to Excel. No recalculating. Just a complete view of every strategy, ready to browse, compare, and decide.
TAX PLANNING ENGINE
OLD PROCESS vs NEW PROCESS
07 — Concept Testing
We ran four concept testing sessions with 2+ tax professionals per session, 8+ tax professionals in total. The response to the planning engine was consistent: the ability to view and explore all permutations at once, and enter a target depreciation total to filter toward a decision, was exactly what teams needed but hadn't known to ask for.
The manual builder asked them to know the answer before they started. The planning engine let them find it.
08 — The Outcome
After using the tax planning engine for Starbucks reported saving 30+ hours of Excel spreadsheet work after adopting the planning engine for the first month.
Their follow-up request: The ability to model the depreciation impact of future asset acquisitions before they happen.
Explore how I led three research studies, interviewed 40+ attorneys, and a self-directed design sprint shaped Bloomberg Law's AI workspace strategy from a blank slate.