This week: IFRS 18: the 2026 comparative needs work now, and 2 more articles
The Finance Decoder
This week's articles: IFRS 18 transition: restating the comparative year, step by step; IAS 23: when an off-plan development must capitalise its borrowing costs; and Service charges under IFRS 15: principal or agent? Below, three items on AI and planning in finance.
IFRS 18 transition: restating the comparative year, step by step
A December year-end group must restate its 2026 comparative under IFRS 18, the new standard on presenting profit or loss. The first place that comparative appears is the Q1 2027 interim, drafted in April 2027. So the restatement is 2026 work.
Appendix C of IFRS 18 applies the standard retrospectively. Paragraph C3 adds a reconciliation, line by line, between the restated comparative and the amounts first presented under IAS 1, the standard it replaces. Quarterly reporters carry more: paragraphs C4 and C5 require the new headings, subtotals and reconciliation in every interim of the first year, so each quarter of 2026 needs a restated comparative.
Start from the 2026 trial balance, not the published income statement. Published lines are aggregates, and finance income and other income are where the mapping breaks. In the article's illustration, rental income on investment property is investing, interest on lease liabilities is financing, and exchange differences on trade payables are operating. Profit for the year stays at 216 under both standards.
Management-defined performance measures, the figures a company defines and reports itself, get no relief in Appendix C. On my reading of paragraphs 31, 32 and 123 together, each one reported in 2027 needs a reconciled 2026 comparative.
For preparers, the order is fixed. Finish the business activity assessment in October and map the chart of accounts in November. A mixed account is cheap to split before year end, while in March it means analysing every transaction.
IAS 23: when an off-plan development must capitalise its borrowing costs
Two developers borrow the same amount at the same rate for the same tower, and one carries three years of interest in inventory and releases it through cost of sales at handover. The other expenses it as it accrues. Both can be right, because under IAS 23, the standard on borrowing costs, the answer turns on the revenue conclusion and not on the loan.
IAS 23.8 requires capitalisation of borrowing costs on a qualifying asset. There is no policy choice. The question is whether the developer holds such an asset at all. If control of each unit passes at handover, it holds inventory under construction and capitalises. If control passes over time under IFRS 15.35(c), the revenue standard's over-time test, it holds a receivable or a contract asset. The IFRS Interpretations Committee, which answers questions on applying the standards, decided in March 2019 that neither qualifies.
Measurement follows IAS 23.12 for a specific loan, net of income on idle funds, or IAS 23.14 for general borrowings, at a weighted average rate capped at the interest incurred. Capitalisation pauses while active development is suspended and stops once the asset is substantially complete.
In the article's example, one year of construction capitalises AED 29.0 million and expenses 42.0 million. Sold entirely over time, the same tower would put all 71.0 million through finance cost.
Auditors should ask for three papers: the IFRS 15 conclusion for each contract type, the funding map and the dates. Get the first wrong and every figure after it moves.
Service charges under IFRS 15: principal or agent?
A landlord that recharges cleaning, security and utilities to tenants can report the recovery as gross revenue only where it is the principal for the service; as an agent, its revenue is the fee it keeps, often nil. Profit is the same either way. The revenue line, the margins and any covenant tested on revenue are not.
Services in a lease are non-lease components, so IFRS 16.17 sends them to IFRS 15, the revenue standard. Its paragraphs B34 to B38 ask whether the landlord controls the service before the tenant receives it, and paragraph B37 gives three indicators: primary responsibility, inventory risk and pricing discretion. Vacancy acts like inventory risk. With two floors empty under a fixed cleaning contract, the landlord bears that cost.
Cleaning and security under contracts the landlord manages usually point to principal. Metered electricity billed at the provider's tariff often points to agent, and so does an investor recharging a building's service charge. Under Dubai Law No. 6 of 2019, the common parts are run by the developer or a management company selected by RERA, Dubai's Real Estate Regulatory Agency.
In the article's example, three contracts cost AED 3,000,000. At 80 per cent occupancy the landlord recovers AED 2,400,000 plus a fee. That fee is AED 240,000, so gross revenue is AED 2,640,000 and the 360,000 shortfall is a cost.
Preparers should keep a short memo for each recharged service. Paragraph 123 of IFRS 15 requires disclosure of significant revenue judgements, and on a large portfolio this one usually qualifies.
Where tabular AI could start in a finance team
The Rundown AI reports that Herzig suggests starting tabular foundation models (AI built for table data) on payment-delay prediction or cash-flow forecasting. The inputs are invoice dates, due dates, actual payment dates, customers and amounts. Per the interview, a few thousand rows can be enough to start, and cutting average collection time from 50 to 45 days at a large company can free tens of millions in cash. SAP's model adds explanations so a controller can check why a prediction was made.
Faster AI drafts move the bottleneck to review
Corporate Finance AI argues that AI makes FP&A (financial planning and analysis) drafts faster but moves the bottleneck to review. Its illustrative case: an AI draft blamed a €1.2 million material-cost variance on purchase prices. The controller found only €0.4 million was price, with €0.5 million from consumption per tonne, and the rest mix and unreconciled. Its advice is that review depth should follow decision risk, starting with reconciling the number.
Keep the budget fixed and let the forecast move
Corporate Finance AI says the approved budget stays the fixed performance reference, while the forecast joins closed actual months to a best estimate for the rest of the year. In its example, an EBITDA (earnings before interest, tax, depreciation and amortisation) budget of 15 million against a forecast of 12 million becomes a bridge by driver, each with an owner. Management actions stay out of the forecast until they have an owner, a timing and a delivery probability.
Tech in Translation
This week's two model releases arrive with outside rankings attached. The video workflows show a method but not yet a result.
Claude Code can turn a folder of assets into a finished episode
A YouTube video shows Claude Code turning a folder of assets into a full episode from one instruction. The folder holds a character sheet, locations, a voice profile and a script. Claude Code voices the script, splits it into shots, sorts character shots from background footage (B-roll), generates each clip through Higgsfield and returns a zipped set. Translation: The video ends before showing the finished output, so the quality is unproven. Read it as a method to try, not a result.
Old YouTube hits can show what viewers already want
A YouTube video suggests picking topics by searching YouTube for home-remedy and kitchen videos from 2014 to 2023 that drew three to ten million views. Those videos predate the wave of low-effort AI content, and the creator says their titles show what older viewers already want. The advice is to adapt those topics, not copy them. Translation: This is a demand check built on past views, which says nothing about whether a new video would find the same audience. Treat it as a way to shortlist ideas.
Kimi K3.1 Pro ranks fourth on an independent index
The Rundown AI reports that Moonshot has released Kimi K3.1 Pro, which Moonshot says matches the top proprietary models on key benchmarks. Artificial Analysis ranks it fourth on its Intelligence Index, behind two Anthropic 5.5 models and Meta's Muse Wonder, per The Rundown. Pricing is $2 input and $10 output per million tokens, with open weights promised by 4 November. Translation: The match claim is Moonshot's own, but the ranking is an outside tester's. Once the weights arrive, others can test it themselves.
Canva's own image model debuts fourth on DesignArena
The Rundown AI reports that Canva has built Craft, its first in-house AI image model, which debuted fourth on DesignArena's crowdsourced leaderboard behind three OpenAI models. It was trained on Canva's own designer templates, and its Magic Layers feature turns generated images into editable designs. It is live for all Canva users and through a public API (a connection other software can use). Translation: For anyone who makes marketing graphics, editable output matters more than leaderboard rank, because a flat image still needs rework by hand.
Deep Space Digest
A video model of a sun cooled by 30 percent shows how quickly plants and the food chain fail. Two more videos cover a possibly wider habitable zone and a glowing black hole star.
A sun cooled by 30 percent ends the food chain first
A Kurzgesagt video models a sun cooled by 30 percent, which leaves it about 75 percent dimmer. Its light shifts toward infrared, so plants lose the light they use, and vegetation and the food chain fail first. Surface temperatures fall about 10 degrees within a week and well below zero within a year, because spreading ice reflects more sunlight. Only geothermal areas such as Iceland stay habitable.
The habitable zone may be wider than assumed
A Cleo Abram video argues the habitable zone may be wider than assumed. Life's building blocks can work in ionic liquids, which are room temperature salts, as well as in liquid water. Those liquids resist evaporation, so they could stay liquid on planets with thin atmospheres or across a wider temperature range. For now, it remains an argument.
The red dot is a black hole wrapped in a glowing cloud
A Cleo Abram video describes the red dot as a black hole star: a supermassive black hole of about 100,000 solar masses. It is wrapped in a hydrogen cloud roughly the size of the solar system and glows like a star. The video says it releases about 100 billion times more energy than fusion alone could, unlike the flat accretion discs usually seen. That energy figure is the claim to watch.
The Knowledge Brief
This week's items share a habit: each asks what the evidence actually supports before repeating the popular version.
A test aircraft steers sonic booms away from the ground
A YouTube video explains that a sonic boom can be bent away from the ground. At a particular speed and altitude, the sound waves curve upward off the warmer air below, so the boom curves away. A test aircraft finds that speed and altitude automatically, with the aim of bringing back supersonic flight over land. The video gives no figures for how much quieter the boom becomes.
Low testosterone turns up in about 1% of people under 60
A YouTube video features guests who argue that low testosterone is diagnosed in roughly 1% of people under 60. They say hormone clinics market it through symptom lists that fit almost anyone. They cite a study of 51,000 men aged 65 and over who believed they had symptoms: about six in seven did not have low testosterone. Replacement into the normal range can help libido, erections, mood and bone density, they say, but it does not produce bodybuilder-level gains.
Creatine has data behind it, gray market peptides barely do
A StarTalk video on YouTube has its guest sort evidence from hype. Creatine has solid data for strength gains of about 10 to 20 percent, but weak evidence for cognition. NAD boosting has no clinical outcome data beyond one niche skin cancer trial, and gray market peptides such as BPC 157 have little human data and real risks. Testosterone replacement helps only truly low levels, which are uncommon under 60.
The pointed witch's hat arrived after the trials peaked
A WIRED video on YouTube traces the pointed witch's hat to later popular depictions, not to medieval origins. It suggests two sources: antisemitic stereotypes of Jewish headwear and Puritan style triangular hats. The hat only became standard in the 18th century, after the witch trials peaked.
