Proof of Capability
We Built a Working Lease Subledger with AI. Here's What It Proves.
Bruce Conway
·
If you close the books under both US GAAP and IFRS, you know the month-end lease grind: one population of leases, two standards, and an auditor who expects everything to tie to the penny.
Over the past several months, our practice built a working lease subledger that does exactly that — dual-standard, ASC 842 and IFRS 16 side by side, proven to the cent — and we built it with agentic AI, Anthropic’s Claude, under CPA direction. No line of code was written by hand.
This article is the written companion to the two videos on our home page: the capability overview (about 14 minutes — the full landscape) and the live walkthrough (about 19 minutes — the engine running a deliberately difficult lease). Here we want to explain what the system actually is, how the numbers are proven, and why we built it in the first place — because the reason is not what most people assume.
What we built
It is not a spreadsheet model, and it is not a chatbot that talks about lease accounting. It is a calculation engine wrapped in a system of record — it carries a lease portfolio end to end: master data, Day-1 measurement, Day-2 lifecycle events, month-end close, ERP posting, and audit-ready disclosures.
The capability spans eleven domains. A few of the ones that matter most to a controller:
Both standards, in parallel. ASC 842 operating, ASC 842 finance, and IFRS 16 from one lease population, with a dual-reporter bridge that reconciles GAAP to IFRS per lease and across the portfolio, timing differences isolated to zero.
The contracts that eat an afternoon by hand. Multi-asset contracts where each asset balances independently and then consolidates. Assets commencing months apart. Real estate and equipment mixed in one contract, routing to different GL accounts. Multi-currency under IAS 21 and ASC 830. Every one of these composes — FX, times mid-month, times multi-asset, on one contract, across all three treatments.
The full Day-2 lifecycle. Payment remeasurements, partial terminations under both methods (because that is an ASC 842 policy election), term extensions and shortenings, renewal and termination option exercises, CPI resets — including the dual-standard divergence where IFRS remeasures the liability and US GAAP expenses the variance — and impairments under ASC 360 and IAS 36.
A real close. Closing a period posts its journal lines into an immutable register, like a real ERP. A change effective in an already-closed period never edits history: the posted months are flagged superseded, and a reverse-and-repost catch-up lands in the first open period, every line carrying the date it would have posted.
Five capture doors, one schema. A keyed form, an Excel workbook, an ERP integration, a versioned REST API, and a live AI agent all converge on one validated schema and run the identical engine. A hand-keyed lease and an agent booking tie to the same number, to the cent.
How the numbers are proven
Any system can display a number. The question an auditor asks — and the question we designed for — is how you know the number is right.
Tie-outs are independent, never by construction. Wherever a disclosure can be computed independently of the trial balance, it is, and a “difference must equal zero” line proves agreement. Every trial balance foots to a $0 check.
Every change is regression-gated. Before any change ships, the engine must reproduce roughly one hundred anchor workbooks cell for cell. Correctness is a gate, not a hope.
It replicates the profession’s own examples. The engine reproduces published worked examples from the FASB codification, the KPMG handbooks, all ten of Deloitte’s modification examples, and the BDO practice aid — down to the published dollar, with every rounding difference documented rather than absorbed.
That last point deserves a pause. When your numbers reproduce the illustrations the standard-setters and the major firms have published, the conversation with your auditor changes character.
Where the human stays in charge
“Built with AI” raises a fair governance question, so we built the answer into the system rather than into a policy document.
Five server-enforced roles carry the segregation of duties. The AI agent can read and capture — book leases, pull the trial balance, run the tie-outs — but period close is human-gated: a machine credential that attempts to close the period is refused with the same 403 a junior accountant would receive, and the denial is logged as audit evidence. Every action, human or machine, lands in a row-level change log — who, which field, the value before and after, and when.
We hold a firm view here: agentic AI belongs inside the control framework, not around it. The close is a human control point, and in our architecture the machine cannot even see the close button.
Why we built it — and what it means for your team
Here is the part we want to leave you with: this is not a product we sell.
We built it as proof — proof that a controller, directing agentic AI, can now stand up a dual-standard lease subledger that ties to the cent, capture to close, with the workpapers and governance to back it. That was not possible on this timeline, at this cost, even two years ago. It is possible now, and the discipline that makes it trustworthy is accounting discipline, not software discipline. That is precisely why it needs a CPA at the wheel.
What our practice offers is the capability, applied to your situation: your portfolio, your chart of accounts, your close calendar, your controls — whether that means building alongside your team, pressure-testing what your current platform produces, or automating the reconciliations around it.
Start with the fourteen-minute overview to see the full landscape, then watch the engine run live. And if the close we described sounds like your month, start a conversation.