J

Notes from running finance: what works, what breaks, and where AI fits.

I write about the day-to-day problems in finance and operations, what I learn fixing them, and what I'm testing with AI.

Juan Cruz Pastorini
Approval flow
  1. Prepares
  2. Checks
  3. Approves
OnRamp 2026

01Experience

Where these notes come from.

2016 — Now

  1. Halo Media

    Head of Finance2022—25 · 2026—

    Built finance visibility across FP&A, business control, payroll, accounting coordination, reporting, systems, and executive decision support.

  2. Mercado Pago

    Business Control · Financial Operations2025—26

    Worked in a high-scale LATAM fintech environment, focused on planning, performance visibility, cost structure, and business control.

  3. Disney

    FP&A · Planning2020—22

    Worked inside a global media finance organization across reporting, planning, forecasting, and performance analysis.

  4. General Motors

    Financial Analyst2018—19

    Operational finance experience in a global automotive environment, working across financial analysis, controls, audits, and business support.

  5. KPMG
    KPMG

    Audit2016—18

    Audit foundation across accounting, controls, financial statements, and cross-industry exposure.

02Approach

Four problems I keep running into.

  1. 01

    Scattered numbers

    Leadership-ready reporting

    Inputs → Control → Decisions

  2. 02

    Margin uncertainty

    Clearer business control

    Revenue & cost → Margin → Capacity

  3. 03

    Manual workflows

    Operating rhythm

    Process → Ownership → Rhythm

  4. 04

    Financial noise

    Executive decisions

    Complexity → Synthesis → Decisions

03AI in my workflow

How I'm using AI in finance, one real problem at a time.

Most of it is unglamorous: cleaning up inputs, pressure-testing numbers, drafting the memo. This is the workflow I keep coming back to.

  • ChatGPT
  • Claude
  • Gemini
  • n8n
  1. 01 · Information

    Turn messy inputs into structured context.

    Parses exports, cleans columns, reconciles definitions across sources, flags inconsistencies.

    OutputA clean structured dataset or briefing, ready for analysis.

    ChatGPT · Claude · n8n

  2. 02 · Analysis

    Pressure-test the numbers, not just present them.

    Models scenarios, validates drivers, surfaces variance gaps, detects edge cases.

    OutputA short list of what changed, what's at risk, and what to dig into next.

    Claude · ChatGPT · Sheets

  3. 03 · Executive narrative

    Translate analysis into a memo leadership can act on.

    Drafts summaries, sequences arguments, frames options for non-finance readers.

    OutputA one-pager or memo with a clear recommendation and supporting logic.

    Claude · ChatGPT

  4. 04 · Decision support

    Frame the call, not just the data.

    Maps tradeoffs, surfaces blind spots, drafts counter-arguments, sequences next steps.

    OutputA decision frame with options, risks, and a recommended path.

    Claude · ChatGPT

04Contact

Working on something similar?

I'd like to hear how you're approaching it, whether it's finance, operations or AI.