New on FinoLib · Volume IV: Finance Careers
I have just published a new article on FinoLib, my free finance education library: Finance Technology Stack: Where AI Belongs in Finance. It is part of Volume IV: Finance Careers. Below is a summary of the topic in plain, expert language, so you can follow the idea even before you open the full article.
G.1 The Finance Technology Stack
The idea in plain words
The finance technology stack runs from the ERP and its subledgers through the close, consolidation, planning, data and business intelligence, to filing and regulatory reporting, with treasury, tax and fund-accounting systems alongside. Errors happen mostly at the handoffs between systems, which is why reconciliations, master data and spreadsheet controls matter as much as the software itself.
Key points
■Each layer has an owner: the Controller owns ledger, close and consolidation; FP&A owns planning.
■Most breaks are interface and master-data breaks, not software bugs.
■The last mile is often a spreadsheet; high-risk ones need version control and independent recalculation.
A closer look
Every role in this volume works inside a chain of systems. A transaction is captured once in an operational system, summarized into the general ledger, closed and reconciled, consolidated across legal entities, compared with plan, and published to the board, investors and regulators. Each layer has an owner, and most finance failures happen at the handoffs between layers rather than inside any one of them.
The core is the enterprise resource planning (ERP) system: the general ledger plus the modules that feed it. Around it sit subledgers (detailed records such as receivables, payables, fixed assets or billing, which post summarized totals to the ledger), close and reconciliation tools, a consolidation engine, enterprise performance management (EPM) software for planning and forecasting, a data platform with business-intelligence (BI) dashboards, and filing tools. Regulated and investment firms add their own layers.
A worked example
Worked Example: One Sale, From Invoice to Board Pack
A software company signs a 12-month subscription for $120,000, billed upfront on January 1. The CRM holds the contract; the billing subledger raises the invoice; the revenue schedule recognizes it evenly. Step Calculation Result January revenue $120,000 ÷ 12 $10,000 Deferred revenue at January 31 $120,000 − $10,000 $110,000 Close check: AR subledger vs ledger $4,812,000 − $4,800,000 $12,000 break The $12,000 break is one invoice raised after the nightly interface ran. The reconciler traces it, the interface reruns, and the balances agree. Only then does the trial balance flow to consolidation, to the EPM tool for actual-versus-plan, and to the board pack and the quarterly filing.
Recap
The finance stack runs from the ERP and its subledgers through close, consolidation, planning, data and BI, to filing. Most breaks are interface and master-data breaks, not software bugs.
G.2 Where AI Is Used in Finance, and Where It Must Not Be
The idea in plain words
AI suits work that is high in volume and low in judgment, with an independent check behind it: matching, coding, extracting data, spotting anomalies and writing first drafts. It must not post journals without review, set values or reserves, approve models or sign certifications, because those carry personal accountability.
Key points
■Judge a use case by who notices an error, how fast, and at what cost.
■The four failure modes are hallucination, drift, leakage of confidential data and unexplainable output.
■SR 26-2 (April 2026) replaced SR 11-7 and leaves generative and agentic AI outside its scope.
A closer look
The useful test is not “can a model do this?” but “if the model is wrong, who notices, how fast, and at what cost?” AI in finance works best on high-volume, low-judgment work with an independent check close behind: matching, coding, extraction, first drafts. It must not make judgments that end in a signature, a reserve or a value on the balance sheet, because those carry personal accountability that no tool can hold.
Regulators have mostly applied existing rules to AI rather than writing new ones. The documents below are the ones a finance professional is most likely to be asked about.
Under the hood
Under the Hood: Why AI Fails in Finance
Hallucination. A language model produces fluent text that is not grounded in the source: a variance comment citing a cause no ledger entry supports, or a lease term that is not in the contract. Fluency hides the error, so the check must compare output with source, not read for sense. Drift. A model trained on last year’s data degrades when the data changes: a new product line, a pricing change, an acquired entity. A coding model that was, say, 97% right before a reorganization can quietly fall when the cost centers change. Accuracy must be measured every period, not once at go-live. Leakage of confidential data.
Recap
AI suits high-volume, low-judgment finance work with an independent check behind it. It must not post journals unreviewed or set values or reserves.
What the full article covers
In the full article I take you through these sections:
■G.1 The Finance Technology Stack
■G.2 Where AI Is Used in Finance, and Where It Must Not Be
It includes worked examples, a look under the hood at the mechanics, a recap, a reveal-on-click quiz and the official sources behind every figure.
Volume IV: Finance Careers
Read the full article on FinoLib
The complete article, Finance Technology Stack: Where AI Belongs in Finance, with every worked example, diagram, quiz and source, is free to read in Volume IV, Finance Careers on finolib.com.
Explore the Finance Education Series: Vol. II — Capital Markets · Vol. I — Global Economy · Vol. III — Finance Encyclopedia · Vol. IV — Finance Careers · Vol. V — Personal Finance