An AI finance analyst for your team.

I set up an AI analyst that knows your books. It does the monthly assembly and the first drafts, and it can't send a number that hasn't been checked.

I've spent my career in finance: consulting, investment banking, and now strategic finance. I know what the board pack, the variance pack and the close cost a small team every month.

Let's talk
A storybook illustration of Jake riding the subway at night with a laptop open on his knees, its screen lighting him up.

What it does

three things, from day one
a sample

What a check looks like

This is a made-up company's board deck, run through the tie-out before it went out. Five things were wrong.

Sample Co: board deck tie-out

CheckResultScope
Source2 breaks16 figures traced to source
Unsourced1 breakevery figure has a source
Cross-slide1 breakfigures shown on two or more slides
Bridge1 break2 bridges sum
  • Slide 4, Rate: deck says 21,000, the variance model says 18,000.
  • Slide 9, Subscribers: deck says 41,200, the source says 40,600.
  • Slide 5, Net revenue retention: no source for this figure.
  • Subscribers: slide 2 says 40,600, slide 9 says 41,200.
  • Revenue bridge: the steps add to 1,793,000, the end shows 1,790,000.

Sample Co is not a real company. See the full report (PDF)

How it works

founding clients at pilot pricing

The workflows

what it can take on

Board reporting

Board prep that is review, not assembly. Inconsistencies get caught before the meeting, not in it.
The usual way
Most board packs are assembled by hand. Numbers get copied from the model, the forecast and the ledger into slides, bridges are rebuilt each quarter, and figures that appear on more than one slide are cross-checked by eye.
What I build
A backing workbook for each meeting with one tab per slide, charts linked into the deck, and the standard bridges (plan vs prior plan, forecast vs actual) built from the same sources every time.
How it's checked
A tie-out runs before the deck goes out. It traces every figure to its source, confirms the bridges sum, and flags any number that differs between slides.
How it fits together
  1. Your sourcesmodel, ledger, bank, KPI exports
  2. Backing workbookone tab per slide
  3. Tie-out
  4. Decklinked charts

FP&A variance reporting

The same pack every month, with explanations that tie to the books and no re-keying.
The usual way
Monthly actual vs plan commentary is pulled by hand from a large workbook, and the explanations are written from memory.
What I build
One pipeline that pulls actuals and plan line by line, flags what moved, and drafts the "why" from the ledger: which accounts and vendors drove each variance. Volume, mix and rate are separated where there is unit data.
How it's checked
Every account must be mapped, actuals must tie to the ledger total, and the pieces of each variance must sum before a pack is produced. Anything the ledger can't explain comes to you as a question, not a guess.

Getting budgets into the projection model

Budget-to-model tie-outs become a check, not a project.
The usual way
Department budgets arrive in spreadsheets with their own layouts. Mapping them into the model is manual and easy to get wrong.
What I build
A mapping layer from budget submissions to your model's account structure. It handles parent and leaf accounts so nothing is double counted, applies your plan-period rules, and supports labelled splits that tie back to the total.
How it's checked
Each submission's total must equal its mapped total, and every split shows its difference against the model, which should read zero.

Building financial projection models

One plan of record, faster scenarios, and reports that refresh from the model instead of being re-keyed.
The usual way
A large workbook with hard-coded links. Inputs drift, fixes break other tabs, and each scenario is another copy of the file.
What I build
It starts with an audit of the model you have. Then a rebuild in layers: actuals in clean tables, a driver-based forecast at the grain you plan on, scenarios as toggles, and a frozen snapshot of each approved plan.
How it's checked
The rebuilt model is tied to the old one line by line before it replaces it. Actualized periods must equal the ledger, and a test run follows every change.

The monthly close roll

A shorter gap between close and an updated forecast, with an audit trail for every change.
The usual way
Rolling the forecast forward after close means pasting actuals, matching cash by hand and reconciling in side spreadsheets.
What I build
A scripted, checklist-driven roll from the closed books into the forecast, with a changelog and a decisions register for every judgment call.
How it's checked
Cash must equal the bank, the model must equal the trial balance, and every gate must read zero before the roll is accepted.

Questions

the ones I'd ask
Do we need Claude already?
No. The pilot runs on my side, so you need nothing. For an install, I set the analyst up on a Claude account your team owns, so it and your data stay yours.
What happens to our data?
It gets read-only access. It can't post, pay or change anything in your books. Anything headed to a board, investors or a lender waits for a person, and the tools it uses are named in our agreement before any data moves.
What does it connect to?
QuickBooks Online and your spreadsheets to start. Where there isn't a live connection yet, it works from an export until there is one.