This week: Intercompany differences: reconcile before consolidating
The Finance Decoder
This week's article is Intercompany eliminations: reconcile before you consolidate. The items that follow stay with finance and AI: a governance framework, adoption and barrier figures from a survey, and a dashboard built without a developer.
Intercompany eliminations: reconcile before you consolidate
An intercompany difference found at consolidation goes into a suspense line, while one found on day two of the month gets fixed by the people who booked it. So the reconciliation belongs in each entity's monthly close, and the matching can be automated.
The causes repeat. A month-end invoice reaches the buyer on the third, two entities translate one balance at rates from different sources, a management fee is disputed, or a posting carries no counterparty code. IFRS 10 paragraph B86(c), in the consolidation standard, requires full elimination of intragroup balances and flows with no tolerance, so a plug is a decision made without evidence.
Unrealised profit comes out too. One subsidiary sells stock costing AED 800,000 to another for AED 1 million. If 40 per cent is still on hand at year end, AED 80,000 of profit is eliminated.
The unit of work is a pair: one entity, one counterparty, one account type. Check every pair's balance each month, retranslating both sides at an agreed rate first, and match transactions only where the balance fails. In the article's matrix, four pairs show AED 1,075,000 of differences that need three different conversations.
For finance teams, the close becomes a calendar. Entities submit balances by counterparty on day one or two, each failing pair gets two owners and a deadline, and corrections go in the entity's own books. Automate extraction and matching, and keep disputed charges, cut-off and sign-off of unrealised profit with people.
Three foundations for governing AI in finance
Corporate Finance Hub argues AI in finance needs three foundations: trusted data, purpose-built logic and auditability. Trusted data means one reconciled chart of accounts and a link from every AI figure to a ledger transaction. Logic means deterministic systems for the arithmetic, with the large language model (LLM) kept to reasoning, search and narrative. Auditability means append-only logs plus human sign-off before any statement, board deck or tax submission leaves the department.
Most finance teams still have no AI policy
Corporate Finance AI's State of Finance Technology 2026 survey puts emails, procedures and summaries as the top finance use of AI, at 55%. Invoice processing is at 11%, audit and SOX (Sarbanes-Oxley) support at 4%, and 13% do not use AI at all. Data security is the biggest barrier at 57%, and only 12% of finance teams have a documented, enforced AI policy while 58% have none. Audit committees may want to check which side of that split their team sits on.
A finance dashboard built by describing it to Claude Code
In part one of a three-part series, Corporate Finance AI argues that a finance person with no coding background can now build a self-refreshing management dashboard by describing it to Claude Code. The stack is read-only accounting data, Claude Code as builder, GitHub for versioned code, and Cloudflare to host it behind a company-email login. The safety advice: a 12 to 24 month history export shown to Claude once, no payroll data, two-step login on every account, and an outside IT check at the end.
Tech in Translation
Two items this week hinge on a number that means little without context, and the third is an allegation with nothing yet proven. Treat all three as claims to weigh, not conclusions.
A video claims a Binance exploit drained $6.3B
A YouTube video reports that an exploit drained about $6.3B from Binance, though it is one creator's unverified account. It says the attacker's signed authorisation let one Ethereum wallet keep running attacker code, and pausing withdrawals did not revoke it. Removal took about four hours. Translation: Treat the details as unconfirmed, since the video covers one wallet only. The habit it points to is worth keeping: know what you approve, how long it lasts and how to cancel it.
Strata's viral speed figure measured reading, not writing
A YouTube video explains Strata, an open-licensed engine for running AI models locally. It says the viral 100-plus tokens a second figure (a token is a small chunk of text) measured prompt reading, not writing. Writing runs at about 4 to 6 tokens a second for DeepSeek V4.1 Flash on a 12 gigabyte card, and the video argues renting DeepSeek's service is far cheaper. Translation: These figures are the creator's and unverified, so ask which task a speed measures before costing a build. Prompt reading and answer writing are different jobs.
Report says a Megaspeed executive allegedly passed data to Beijing
The Rundown AI relays a New York Times report of 29 October that an executive at Singapore-based Megaspeed allegedly passed information to Beijing while close to Jensen Huang. Officials did not believe Huang knew of her alleged role, and Trump was reportedly briefed about a year earlier without action. Megaspeed, under a U.S. probe over steering Nvidia chips to China, bought at least $4.8B of Nvidia gear through a sister firm. Translation: Nothing is publicly proven, so read this as an allegation. Buyers of Nvidia hardware may face closer official scrutiny.
Deep Space Digest
Two explainer videos this week deal with scale. One covers what a black hole does to a falling body, the other how far away the Sun sits.
A black hole's size decides whether a faller survives
A YouTube video describes what one person falling into a black hole would experience. Near a stellar-mass hole, tidal forces pull the body apart (spaghettification), while a supermassive hole's horizon is gentle enough to cross intact. The video labels its Cauchy horizon idea, where physics stops predicting, hypothetical.
It takes 374 planet lineups to reach the Sun
A YouTube video says 374 lineups of all the planets, laid end to end, would reach the Sun. One lineup stops at the Moon, 95 reach Venus, and in a basketball-sized Sun model Earth is a pin-sized dot a court's length away. Distance dominates the picture.
The Knowledge Brief
Each item this week takes something people assume and looks at what the source actually found.
Witch accusations came from neighbours, a historian says
In a YouTube video, historian Mikki Brock answers Reddit questions about witch hunts. She says accusations usually came from neighbours who blamed a misfortune on someone they already distrusted, and that about 80 percent of those accused in Europe were women. Witch hunts were costly rather than profitable, and confessions taken under torture cannot be read at face value. She ends by arguing that dehumanising political language about marginalised groups follows the same pattern.
Brain chips read movement intent, not thoughts
A YouTube explainer compares four ways to read the brain: scalp electrodes, Synchron's implant delivered through a vein, Precision's thin film on the brain surface, and the Utah array, in humans since 2004. Neuralink's N1 uses flexible threads, which its R1 robot inserts at about 1.5 seconds each. Current systems read motor intent on its way to the hand or mouse, not thoughts.
Rest alone left her drained until she added a challenge
Miss Excel writes in her newsletter about a 2007 study by Sabine Sonnentag and Charlotte Fritz, which found that people recover from work in four ways: detachment, relaxation, mastery and control. Her own couch weekends were all relaxation and left her drained. Adding mastery, a challenge such as learning a new skill, gave her energy back.
A gas from gut bacteria gives the dead their green tint
A YouTube video explains why zombies are green by following a body after death. Without oxygen, brain cells die within minutes, and muscles lock within 2 to 6 hours before loosening after about 48 hours. Blood settles by gravity and purples the lower areas, then gut bacteria spread into the body and release gases. One gas reacts with a blood protein to produce the green tint and marbling.
