13 yearsin product, analyst to Senior Director
$120M+recurring revenue in my portfolio
7 : 2engineering teams to PMs today
$5M+card revenue, product built from zero
95%+AI transcription accuracy, every Enterprise account
Case studies
Four pieces of work that show how I operate. Figures are rounded or generalized where the real numbers are internal. Happy to go deeper in person.
01 2026
The P&L we'd never had, and what it showed us
Emburse Enterprise · 2026 · Unit economics, pricing, abuse
The setup
We charge customers for warehouse API access and for pulling data out through our embedded analytics product. Both cost us money as usage grows. What nobody could tell you was what a given account earned us against what it cost us. Revenue and cost by account and product didn't exist in one place. The pieces sat in Excel files, Snowflake and Looker, and none of them lined up.
What I did
- Pulled revenue and cost for the embedded analytics product, by account, out of Excel, Snowflake and Looker, and wrote the Python to join it (AI tools did a lot of the typing). It publishes to an internal site.
- Every quarter I take the numbers from it into the leadership review. Leadership reads the deck. The site is my working tool.
- Used it to look for whitespace: which accounts and products had room to grow.
- It also showed a set of accounts whose warehouse API calls and analytics data extraction cost us more than they paid. Some of that was heavy, legitimate use. Some of it was abuse.
- Led the fix through pricing. Those capabilities moved to usage tiers, so heavy use gets paid for.
Outcome: leadership gets a revenue and cost view by account and product every quarter, and the heaviest users of our most expensive capabilities now pay by usage tier. I've left the dollar figures off this page.
What I'd tell you about it
Averages hide pricing problems. Once cost sat next to revenue for each account, our most valuable customers and our most expensive ones stopped looking like the same customers.
02
A card payments product from zero, then monetization a quarter early
Emburse · 2020 - 2025 · 0-to-1, payments, monetization
The setup
We had the expense workflow. We didn't have the spend. Cards put us in the transaction flow instead of after it, with controls before the money moves instead of cleanup afterward. The catch: an expense company has no native credibility in payments, so the product had to earn it.
What I did
- Built and launched Emburse Cards for the enterprise product: pre-spend controls for admins, real-time alerts, reporting. We went from our first customer to 15 in the first year.
- Came back to it at scale in 2024-25: UK card issuance, Amex card expansion, receipt-to-card matching on the last four digits to kill duplicate submissions, and one-click setup for real-time transaction alerts.
- When a late card partnership request came in mid-year, I rebalanced the roadmap to get it engineering-ready that quarter instead of letting it slip. That pulled first-year monetization of the partnership forward a full quarter.
Outcome: a card line past $5M in revenue, attached to the expense workflow, plus card-driven automation (real-time feeds, matching) that makes the core product stickier.
What I'd tell you about it
Distribution beat polish. We won card deals because the workflow was already in the building. The decision that mattered most wasn't a feature. It was re-sequencing the roadmap so revenue didn't wait a quarter.
03
One AI capture platform under three products, with no big-bang cutover
Emburse Enterprise · 2023 - 2025 · AI/ML, platform migration, build vs. buy
The setup
Expense automation lives or dies on transcription. Our OCR stack had grown into a patchwork: a third-party engine we paid real money for, with Microsoft and Google services stitched underneath. Accuracy had plateaued, and every renewal conversation landed on the same question: when does this get more automated?
What I did
- Owned the product side of moving capture to an in-house GenAI transcription platform, with our Data Science team. I took applied AI coursework so I could push back on model decisions with some credibility, and set up the PRD structure we now use for AI investments.
- Sequenced the rollout: internal pilot, early adopters, then 100% of Enterprise accounts on web and mobile. The same platform went under two more Emburse expense products in the same program.
- Kept a kill list next to the build list. We shut down the third-party OCR engine and two legacy services, and the savings paid for part of the program.
- Put our first category suggestion model into production. It replaced the legacy prediction engine, and the internal pilot cut manual edits by about 80%.
- Extended the platform into GenAI hotel and car folio itemization and AI auto-merge for card transactions. That's the unglamorous plumbing behind one-click expense.
Outcome: transcription accuracy up 20+ points to 95%+ on expenses, at or above competitor benchmarks, running for every Enterprise account. It's also the base for the agentic expense work I lead now.
What I'd tell you about it
AI products fail in the seams, not in the model. The model team can show you a great demo in a month. The two years go into folio edge cases, merge logic, rollout order, and deciding what to turn off.
04
Our angriest global account, back to a renewal conversation
Emburse Enterprise · 2024 - 2025 · Admin tooling, escalation, executive communication
The setup
One of our largest global accounts, a multinational manufacturer, had escalated to the executive level. Years of friction: payment workflows, how configuration changes got validated and rolled out, integration gaps. The company stood up a task force. I ran the product side.
What I did
- Turned a long grievance list into a committed, sequenced plan, then protected it.
- Shipped Changeset Management in two phases. It changed how admins validate and roll out configuration changes, which was their single biggest operational complaint.
- Fixed long-running payment workflow issues that were slowing their global rollout.
- Made trade-offs explicit. When two of their priorities collided on one engineering team, I laid out the conflict and a recommendation and let them choose. They chose, we delivered, and trust went up.
- Met their leadership alongside our CS team. Same playbook I've used on dozens of escalations across construction, legal, consulting and university accounts: show a year of delivered value first, then use the roadmap to defend the price.
Outcome: the task force closed out its commitments, the account de-escalated, and renewal conversations reopened while they kept expanding globally on our platform. The same playbook protected six-figure ARR at other at-risk accounts.
What I'd tell you about it
Escalations are discovery with the stakes turned up. Every angry workstream on that task force pointed at a real product gap, and half of what we built for them ended up on the core roadmap for everyone.