The 18-Day Ramp

Sometimes the job starts before the context exists.

I joined a finance team mid-transition and stepped into a live close before the full context existed. I used AI to rapidly map the systems, connect scattered knowledge across files and workflows, and turn ambiguity into a structured path toward an accurate close.

onboarding.tsTypeScript

// onboarding, declared as code, not learned by osmosis const ramp = new FinanceEngineer.Onboarding("NewCo", { day: 0, documentation: minimal, closeProcess: "live", strategy: deploy([ rapidSystemArchitect, // map the system knowledgePipeline, // capture what lives in heads fileIntelligenceLayer, // make files agent-ready ]), }); export const status = ramp.run(); // → "owning month-end close by day 18"

The three plays

Each one is a use case in this deck. Open any to see it run.

Use Case 01 · Rapid System Architect

At some point, you inherit books that no longer tie.

During periods of transition, context often lives across systems, spreadsheets, Slack threads, and individual workflows. The close still has to move forward, but the first challenge is making the operating picture visible enough to trust.

Sound familiar?

▦

Context is distributed

Important knowledge often lives in multiple places, not one clean source.

≠

History is hard to reconstruct

Balances may be correctable, but the path to get there is not always obvious.

≣

Methods evolve over time

Different people may have solved problems in different ways as the business changed.

△

Changes need care

Before adjusting prior work, Finance needs a clear trail from source activity to close impact.

This use case is the playbook.

Anchor to the last clean state, reconstruct what changed, and rebuild toward a close that is fast, accurate, and auditable.

A subsidiary had drifted out of sync, and the close still had to happen.

The AI workflow

This was not a one-shot prompt. It was a working reconciliation loop: a single context built from the raw materials, anchored to the last clean month, then refined discrepancy by discrepancy until the full trail tied back to the bank.

Building one working context 0% full

🗎ERP_GL_transaction_export.csv~8.2k tok

🗎ERP_IC_journal_export.csv~2.4k tok

🏦Trust_Account_Statements.pdf~11k tok

🗎Bank_Statement_export.csv~3.6k tok

🎙Nov2025_Process_Recordings~14k tok

▦Nov2025_Verified_Close.xlsx~5.1k tok

🗎Funding_Requests.pdf~1.2k tok

📋Payroll_Tax_Liability_Schedule.pdf~0.9k tok

🗎Pension_DirectDebit_Confirmations.pdf~0.7k tok

🗎Prior_JE_Detail_exports.csv~4.3k tok

📋Discovery_Notes.txt~1.5k tok

🗎ERP_GL_transaction_export.csv~8.2k tok

🗎ERP_IC_journal_export.csv~2.4k tok

🏦Trust_Account_Statements.pdf~11k tok

🗎Bank_Statement_export.csv~3.6k tok

🎙Nov2025_Process_Recordings~14k tok

▦Nov2025_Verified_Close.xlsx~5.1k tok

🗎Funding_Requests.pdf~1.2k tok

📋Payroll_Tax_Liability_Schedule.pdf~0.9k tok

🗎Pension_DirectDebit_Confirmations.pdf~0.7k tok

🗎Prior_JE_Detail_exports.csv~4.3k tok

📋Discovery_Notes.txt~1.5k tok

streaming files into context…

1

Built one big working context

Pulled every trust-account statement, the ERP ledger exports, and six recordings of the original close into a single Claude context, so it could reason over the whole entity at once, not file by file.

2

Anchored to the last clean month

Had Claude reconstruct how the books were supposed to work (account structure, entry patterns, the funding cycle) from the verified baseline and the walkthroughs.

3

Interrogated what looked wrong

A long back-and-forth against the source data: "this balance is negative, why?", "trace this clearing entry to the bank," "these liabilities don't split out." Each question I asked surfaced a concrete problem:

Intercompany clearing booked in USD, not GBP

Liabilities collapsed into one catch-all

Payments made but never journaled

Parent funding receipts never recorded

A reclass that unbalanced the one clean month

A recurring accrual silently stopped

4

Verified, corrected, documented

Pinned each finding to a specific bank line, structured one cumulative self-balancing correction, and reviewed every entry before posting, with an audit narrative explaining the why.

What this proves

The hard part wasn't correcting the numbers; it was reconstructing why they drifted. Tracing each account from the last clean baseline back to source activity turned a potential multi-period reopening into one cumulative correction, delivered as an audit-ready package explaining the why behind every entry.

2.5 days

not 2–3 weeks

0

periods reopened

100%

traceable to bank

Use Case 02 · Knowledge Pipeline Engineer

The workflows are already there. The documentation isn't.

Record-to-report: the workflow behind the numbers spans people, systems, approvals, reconciliations, and exceptions. The goal is to turn that context into a visible process map the team can improve, control, and eventually automate.

So I turn every process conversation into a map

◉

1 · Conversation

Sit with each process owner and record the walkthrough. Capture it once, properly.

▤

2 · Transcript

Speech-to-text turns the call into a clean, searchable record of how it really works.

⊞

3 · Visual map

Feed the transcript to AI and get the process back as a diagram anyone can read.

⚑

4 · Pain points

The map surfaces the gaps and dependencies that quietly impact the downstream close.

One method, every process.

Run the same loop across order-to-cash, procure-to-pay, and hire-to-pay, and you've mapped how the company actually operates: reusable knowledge for the next close, the next hire, the next automation.

The order-to-cash cycle, end to end.

The AI workflow

One recorded interview, fed to AI, comes back as a structured map of the whole cycle: every stage, system, and handoff made legible in a single read.

The interview9 areas · 40+ questions

Systems & rolesWhich system is the source of truth for revenue?

Deal → setupWhat gets validated before a deal becomes a customer record?

BillingHow do invoices flow from billing into the ERP, and on what cadence?

Revenue recWhich scenarios call for a manual adjustment outside the billing system?

Sales taxWhere is tax treatment non-standard (marketplace, international)?

CollectionsHow does a payment get applied once it lands in the bank?

Month-end closeWhat does the billing-to-ERP reconciliation actually involve?

Controls & accessWho can create or modify invoices, and who applies payments?

ImprovementsIf you could change one thing, what would it be?