I turn financial operations into autonomous systems.

7 years running treasury at a €2.3B unicorn. Cash forecasts that moved 50% week to week. Liquidity crises at 2am. Covenant breaches that could let lenders convert debt into equity. I've lived the chaos that AI is about to fix. Now I design the agents that do it.

Schedule a conversation See my work
0%
Debt
Reduction
€0B
Fundraising
Supported
€0B
Receivables
Managed
0
European
Markets
About
Rafael Pardo

When I joined, the company was managing €2B in receivables across 10 countries through fragmented ERPs and 20+ Google Sheets. Cash forecasts moved 50% from one week to the next. I dealt with 2am liquidity crises, securitisation advance rates threatening to drop below covenant thresholds, and the very real possibility that our lenders would convert debt into equity if cash positions deteriorated further. I was fast-tracked from junior analyst to product manager, eventually owning treasury operations across all European markets. I reduced overdue debt by 85%, built the unit economics model (LTV-CAC, cohort analysis) that BlackRock, SoftBank, and the board used to make investment decisions, and helped build the financial infrastructure behind €1.3B in fundraising. Before that, I built analytics models for Coca-Cola, IKEA, and Mahou at MRM Worldwide, and cut my teeth on fintech at Fintonic.

That experience taught me where real value is created in financial operations, and where it's wasted. Treasury is repetitive, rule-based, and high-stakes: the perfect environment for AI. So I started building. I designed AI agents using Claude for cash application and invoice matching, achieved a 30% improvement in forecast accuracy, and built the product vision for autonomous treasury systems that reduce manual work by half. I know exactly where AI delivers ROI in finance, and where it's still hype.

I complement that operational depth with formal AI and ML training: Stanford's Machine Learning Specialisation, Anthropic's certification programme, and Reforge's AI Product Management curriculum. Not to collect credentials, but to build the technical intuition that separates a product manager who uses AI buzzwords from one who can actually ship autonomous systems.

I also invest. I've been on both sides of the table: building the financial models and data rooms behind €1.3B in fundraising rounds, and deploying my own capital as an angel investor in early-stage startups and as an LP in a venture fund. That dual perspective, operator and investor, shapes how I evaluate AI products: not just "does it work?" but "does it create defensible value at scale?"

I believe the best AI products will be built by people who've operated the systems they're replacing. Not by technologists guessing at business problems, and not by operators afraid of the technology. The intersection is where I live.

Education

IE Business School (BBA, top 5%) · Michigan Ross (Exchange) · LSE (Summer Programme)

AI Product Management Treasury Automation Autonomous Agents Cash Forecasting Collections Optimization Angel Investing
Thesis

Treasury will be the first back-office function to go fully autonomous. The people who build that future won't be technologists guessing at financial operations. They'll be operators who learned to build.

Every enterprise treasury platform automates the easy 70%: clean matches, straight-through processing, standard workflows. The remaining 30%, partial payments, cross-entity netting, country-specific collection norms, seasonal cash patterns, is where most of the value sits, and it requires AI agents that understand the domain deeply enough to handle exceptions without human intervention. I've spent seven years inside that 30%. Now I design the systems that resolve it autonomously.

Work
01

85% Debt Reduction Across 10 Markets

Treasury Operations / €2B Receivables Portfolio

Managed a €2B receivables portfolio across 10 European countries at a €2.3B HR-tech unicorn. Designed and executed a data-driven collections optimisation strategy that reduced overdue debt from €100M to €15M. Built the finance data infrastructure from scratch: started with a proof-of-concept data mart I prototyped using Google Scripts and SQL, then secured internal buy-in to scale it with a fully dedicated Data Science & Engineering team. The result was an enterprise-grade system: Redshift-based data mart, Looker dashboards for C-suite reporting, and an 80% faster reporting cycle serving treasury, finance, and executive stakeholders across all markets.

Collections Optimisation Data Infrastructure Redshift Looker Multi-Country Ops
02

€1.3B Fundraising Infrastructure

Capital Markets / €150M → €2.3B Valuation

Built the financial models, reporting infrastructure, and data rooms that supported €1.3B+ in fundraising across multiple rounds with BlackRock, SoftBank, and Atomico. Designed enterprise-grade Looker dashboards for C-level executives, including LTV-CAC modelling and cohort analysis that became core to the investor narrative. Created the company's first cross-market KPI framework, adopted across all 10 countries. Led analytics that reduced churn from 15% to 9%, supporting €30M+ year-on-year growth and a 15x valuation increase.

Financial Modelling LTV-CAC Analysis Cohort Analysis Investor Relations KPI Frameworks
03

AI-Powered Treasury Automation

AI Product Design / Claude Agents in Production

Led the end-to-end treasury transformation from manual operations to AI-powered systems. Designed AI agents using Claude (tool use, multi-step reasoning) for invoice matching and cash application. Built the business case, secured executive buy-in, and defined the product vision from scratch. Delivered: 30% improvement in forecast accuracy in pilot phase. Designed for: 50%+ automation rate and 5-10% DSO reduction at full deployment.

Claude / Anthropic AI Agents Cash Application Invoice Matching Forecast Accuracy
04

Autonomous Finance Platform: Product Strategy

Product Strategy / 22-Page Document / Anonymised

Authored the complete product strategy for an autonomous treasury platform at a European HR-tech unicorn (€2B+ annual receivables, 10 countries). Defined the three-layer architecture (data foundation, automation, AI differentiation), competitive moat through proprietary staffing data and multi-ERP orchestration, North Star metric with multiplicative decomposition, AI governance framework with human-in-the-loop requirements by decision type, phased 9-month rollout, and change management strategy including a resistance early warning system. The document represents 95% of the final version, completed before departure.

Product Strategy AI Governance Platform Architecture Change Management North Star Metrics
Download full document (PDF) → Read case study →
05

AI Treasury Analyst: From PoC to Production

Applied AI Research / Claude Skill Architecture

Designed and built a custom Claude skill that functions as an autonomous treasury analyst: ingesting real financial data, applying weighted moving averages with recency decay (λ=0.9), seasonal indices per collection bucket, and partial payment handling to produce CFO-ready cash flow forecasts. The proof of concept, iterated through 40+ versions, improved forecast accuracy by 30% vs. simple averages. Key design decisions: structured JSON outputs over free text, deterministic calculation rails (LLM orchestrates, code calculates), and country-specific configuration files encoding market-level payment norms. Production version currently under development, targeting a fully interactive demo with live data.

Claude Skills Tool Use Cash Forecasting WMA / Decay Factor System Design
Read case study →
Insights
Thesis / Treasury + AI 9 min read
Why Treasury Will Be the First Back-Office Function to Go Fully Autonomous

Every CFO talks about AI transformation. None of them start with treasury. That's a mistake. Treasury operations are repetitive, rule-based, and high-stakes: the perfect trifecta for autonomous agents. Here's why the first fully AI-driven financial function won't be accounting, compliance, or FP&A. It will be treasury.
Framework / Product Management 8 min read
The Operator's Edge: Why the Best AI PMs Have Run the Systems They're Replacing

The AI product management world is full of people who've never operated the processes they're automating. That's a problem. You can't design an intelligent collections agent if you've never called a debtor. You can't build a cash forecasting model if you've never managed a liquidity crisis at 2am. Operational depth isn't a nice-to-have: it's the moat.
Technical / Building with Claude 11 min read
How I Built a Treasury Analyst in Claude (And What It Taught Me About AI Product Design)

A practical walkthrough of designing, building, and iterating on a custom Claude skill for treasury analysis. What worked, what failed, and what I learned about the gap between prompt engineering and actual product design. Includes the full architecture and design decisions behind each component.
Architecture / AI Systems 9 min read
Why Most Enterprise Multi-Agent Implementations Will Fail

The bottleneck isn't the agents. It's the space between them. Most multi-agent architectures will fail because they're designed by people who've never operated the systems they're connecting. Five failure modes from shared state to error propagation, and what well-designed systems do differently.
Framework / AI Product Management 8 min read
The AI Product Manager Is Dead. Long Live the AI Orchestrator.

Feature roadmaps become behaviour roadmaps. User stories become agent stories. QA becomes evaluation. The PM role isn't evolving, it's mutating. The PMs who survive will be the ones who understand systems, not screens.

Let's build something worth building.

30-minute call. No pitch. Just a conversation about what you're building.

Schedule a conversation
rafael@rpardo.ai · LinkedIn · Download CV