AIFinance AI Lab

AI for finance · zero to hero

Stop buying black boxes.
Build one first.

A code-first school for CFOs, Finance Leaders, FP&A teams, fractional CFOs and accounting partners. Open a blank notebook, write the tokenizer, the invoice parser, the Monte Carlo runway engine and the audit agent yourself — then evaluate Datarails, DataSnipper, MindBridge and Vic.ai knowing exactly what's under the hood.

Apply to the course →See the curriculum →
labs/m3_monte_carlo_runway.py
cash on hand             $2,400,000
monthly burn (p50)         $207,021
runway p10 / p50 / p90   10 / 13 / 17 mo
P(cash-out ≤ 12 mo)           46.4%

driver sensitivity (Δ runway, p50)
  ARPA -10%   ██████  -3 mo
  churn +1pt  ████    -2 mo
  CAC +25%    ████    -2 mo

✓ 10,000 paths in 0.5s · seed 42
88%

of finance & accounting pros call AI the most consequential technology for their discipline

8%

say their organization is fully prepared to deploy and govern it

56%

report a skills gap in generative AI and prompt design

Industry benchmark surveys. The gap is the course.

The ladder

Climb from primitives to production.

Level 0
Code primitives
Level 1
Custom scripts
Level 2
Agentic workflows
Level 3
Enterprise deployment

Curriculum

Seven modules. One blank notebook.

01free

Orientation & the AI Mental Model

Start with what a language model actually does: predict the next token. Separate the model, the chat interface and the tools before writing your first finance notebook.

Use case

Create a personalized finance tutor and learning plan.

What you'll build

A finance learning notebook and a sandbox with synthetic data.

What you'll learn

  • Explain how ChatGPT works and where it fits in the AI ecosystem.
  • Use a tutor to break a finance concept into examples and check your understanding.
  • Set up Python and notebooks without uploading confidential finance data.
Open module 01 →
Lesson syllabus
  1. Your Finance work is about to change
  2. AI at your work
  3. Your finance data sandbox
02free

Tokens, Embeddings & Finance Prompting

Turn text into tokens, then vectors. Build a small tokenizer and inspect similarity before asking a model for a structured finance answer.

Use case

Draft budget-owner support conversations and brainstorm finance, FP&A and accounting ideas.

What you'll build

A structured budget Q&A prompt and a chart-of-accounts classifier.

What you'll learn

  • Build a tokenizer and explain embeddings with a concrete account-mapping example.
  • Improve prompts using examples, explicit constraints and a structured output schema.
  • Draft clear finance communication and compare alternative planning ideas against supplied facts.
Open module 02 →
Lesson syllabus
  1. Writing a tokenizer from scratch
  2. Embeddings for chart-of-accounts mapping
  3. Prompting with structured outputs
  4. Lab: classify a GL extract
03

Invoice Parsing & Checked AP Pipelines

An invoice is a document; a journal entry is typed data. Build the conversion one field at a time, then add matching rules and checks before any posting step.

Use case

Summarize invoices and automate their handoff into an accounts-payable workflow.

What you'll build

An invoice extraction, three-way matching and journal-validation pipeline.

What you'll learn

  • Extract invoice fields into a schema and trace each value to its source.
  • Code PO, receipt and invoice matching with explicit exceptions.
  • Validate balanced entries and route uncertain outputs for human review.
Open module 03 →
Lesson syllabus
  1. Anatomy of an invoice parser
  2. Three-way matching in code
  3. Posting journal entries with checks
  4. Lab: an end-to-end AP pipeline
04

FP&A Drivers, Simulation & Reporting

Start with cash in minus cash out. Add drivers, sample uncertainty and run many possible futures; only then turn the calculated results into a report.

Use case

Generate high-quality FP&A reports and brainstorm budget scenarios.

What you'll build

A driver-based runway simulator with a source-grounded variance report.

What you'll learn

  • Build a driver-based model from transparent assumptions.
  • Simulate runway ranges and identify which inputs change the result most.
  • Generate management commentary from calculated outputs, checking every reported number.
Open module 04 →
Lesson syllabus
  1. Driver-based models
  2. Sampling burn and churn
  3. Monte Carlo runway engine
  4. Driver sensitivity
05

Document Retrieval & Evidence-Based Audit

A model cannot cite evidence it never received. Split documents, retrieve relevant passages and attach sources to every answer before testing a full transaction population.

Use case

Summarize documents, agreements and supporting invoices with citations.

What you'll build

A retrieval-and-answer pipeline for finance documents and a trial-balance audit lab.

What you'll learn

  • Build document chunking, retrieval and a cited answer step.
  • Separate a supported summary from a plausible but unsupported claim.
  • Test a full transaction population and inspect exceptions rather than assuming correctness.
Open module 05 →
Lesson syllabus
  1. From sampling to full populations
  2. Building a cited RAG agent
  3. Lab: audit agent on a trial balance
06

Context, Evals & Finance AI Controls

Treat an AI answer as a hypothesis, not an approval. Build a set of known cases, measure failures and make context, memory and permissions explicit.

Use case

Evaluate finance reports and support answers before they reach colleagues or clients.

What you'll build

A finance evaluation harness, red-team cases and a review checklist.

What you'll learn

  • Manage context and memory while keeping sensitive information out of unsuitable tools.
  • Test factual accuracy, numerical consistency and prompt-injection failure modes.
  • Define human approval, privacy and ethical-use controls with an audit trail.
Open module 06 →
Lesson syllabus
  1. Writing eval harnesses
  2. Controls and model risk docs
  3. Red-teaming finance agents
07

Tools, Workflows & Team Deployment

A useful demo is not a reliable service. Connect inputs, model calls and checked outputs; add retries, access boundaries and monitoring before handing it to a team.

Use case

Integrate AI into finance tools and workflows to automate repeatable work.

What you'll build

A reusable finance workflow and a capstone service with a human approval step.

What you'll learn

  • Connect AI steps to existing finance tools with validated inputs and outputs.
  • Design retries, logs, access controls and human approval points.
  • Measure time saved, running cost and business value before scaling a service.
Open module 07 →
Lesson syllabus
  1. Deploying for a team
  2. Pricing the value, not the hours
  3. Capstone: your AI service line

The stack

Where each module lands in your finance stack.

5Analytics & narrativeVariance commentary and board-ready stories.M2 · M4 · M7
4FP&A & planningDrivers, scenarios and probabilistic forecasts.M4
3Execution & automationAP pipelines, reconciliations, agents.M3 · M5 · M7
2System of recordThe ledger, controls and audit trail.M3 · M6
1Data foundationClean, typed, versioned finance data.M1 · M2

Pathways

Pick the route that matches your seat.

CFOs & Finance Leaders

M2M3M4M6

Close faster and forecast with ranges, not single numbers.

FP&A Teams

M2M4M6

Model drivers, compare budget scenarios and build checked management reports.

Fractional CFOs

M2M4M5M7

Serve more clients with reusable, productized models.

Accounting Partners & Audit Firms

M2M3M5M6M7

Test full populations and package it as a new service line.

Walkthrough preview

Every lab ships with a guided walkthrough.

M4 · Lesson 3
Monte Carlo runway engine
Use Case: 18-month cash planning
Presenter · Lena Levin
  1. 1. Load drivers
  2. 2. Sample burn
  3. 3. Simulate paths
  4. 4. Read percentiles
  5. 5. Sensitivity
Step 3
Simulate 10,000 paths

Each path draws churn, ARPA and CAC from their distributions, then rolls cash forward month by month.

WALKTHROUGH · 5 STEPSPress ▶ to play the walkthrough
◀▶▶▶1×🔇✕

Why build from scratch

You can't govern what you've never built.

Spot the failure modes

Once you've written a parser, you know exactly where vendor tools hallucinate, drift and break.

Own the handoff

Specify, review and sign off on AI work with the confidence of someone who's shipped it.

Price the value not the hours

Turn automation into products and retainers instead of giving the savings away.

Open a blank notebook. Start at zero.

Apply to the course →See the curriculum →Enter school