Revenue Recognition: ASC 606 vs IFRS 15

2026-08-18

The Five Steps Agree. The Judgment Calls Don't.

By the time the FASB and the IASB finally finished writing revenue recognition guidance together, in 2014, both boards had spent years staring at the same problem shaped in opposite directions. In the US, the guidance on when a company could book revenue had metastasized into more than a hundred narrow, industry-specific rules: one set of instructions if you were a software company, another if you built things, another if you sold real estate. Two companies doing economically identical things could report completely different numbers, not because their businesses differed, but because they'd been filed under different chapters of the rulebook. Everywhere else, under IFRS, the opposite disease was eating the same organ. There were only two thin, high-altitude standards, IAS 18 and IAS 11, and almost no worked-out guidance for anything complicated. So companies filled the gaps with judgment, and the judgment didn't agree with itself even within a single industry.

Different diseases, same symptom: you couldn't trust that two revenue numbers, sitting next to each other, meant the same thing. So the two boards did something they don't often manage: they wrote one standard together, gave it the same title on both sides (ASC 606 in the US, IFRS 15 everywhere else), and built it around a single five-step model that doesn't care what industry you're in. Identify the contract, identify the distinct promises inside it, work out the price, split that price across the promises, and recognize revenue as each promise gets kept.

What makes this comparison different from the usual "US GAAP vs IFRS" exercise is that there's no clash of philosophies to describe. Both boards wanted the same thing and wrote most of the same words to get it. So the real question isn't how these two systems differ. It's narrower, and I think more interesting: given that they started from an identical draft, where exactly did agreement break down, and what does the shape of that breakdown tell you about the two institutions that couldn't quite finish agreeing?

Testing the skeleton

Most of the five-step model turned out to be genuinely, boringly identical across both standards, which I only really believed once I'd tried to break it.

Take the fourth step: splitting a bundled price across two things sold together. A phone sold with a year of free service for ₹20,000, where the phone alone would go for ₹19,000 and the service alone for ₹2,000, a ₹1,000 discount buried in the bundle. My first instinct was to just put the discount somewhere reasonable-sounding: knock ₹1,500 off the phone, hand the service ₹1,500 more of the total. It felt harmless. It wasn't. The standard insists the discount be split proportionally, in the same ratio as the two standalone prices, and the reason only became obvious once I noticed which direction my "reasonable" instinct had leaned, toward the item that gets recognized immediately, away from the one that trickles in over a year. Give a company the freedom to allocate a discount however it likes, and the discount migrates toward whichever bucket books revenue soonest. That's not a hypothetical risk. It showed up, unprompted, in my own first guess.

It gets sharper when one item's standalone price isn't observable at all, when a company has never sold the add-on separately and has to estimate what it would have cost alone. Nudge that estimate up (claim the phone would "really" sell for ₹19,800 instead of ₹19,000, no market evidence required), and the residual assigned to the service shrinks from ₹1,000 to ₹200. Eight hundred rupees quietly move from the bucket that recognizes slowly into the bucket that recognizes on day one. Same mechanism, dressed up as an estimate instead of an allocation choice. The standard's answer is identical in both cases: take the discretion away. Estimate standalone prices only through specific, constrained methods, and keep the most manipulable one, the residual approach, on the tightest leash.

The second step, deciding whether a contract holds one promise or several, hinges on a test I first got half-right for the wrong reason. A phone bundled with an optional, separately-priced accidental-damage plan is obviously two promises, the plan is something a customer could buy or skip, from this company or a competitor. A phone with a proprietary chip embedded at the factory, without which the phone won't turn on, is obviously one promise, except I initially explained why by pointing at the wrong thing (that the chip "isn't sold separately"), when the real test is narrower: could the customer ever have wanted the chip on its own? No, it was never a thing-in-itself to the customer, only ever "a working phone."

The genuinely counterintuitive case sits between those two. A company sells the same accounting software license to hundreds of small clients as a clean, self-contained product, clearly capable of standing alone. Sell that identical license to one large client, bundled with a rewrite so extensive the software won't function for them without it, and the license stops being distinct, not because the product changed, but because this contract changed what the license means. Distinctness isn't a property of the product. It's a property of the specific deal. The same license can be one promise in one contract and half of a combined promise in the next, and the standard is entirely comfortable with that, because it isn't actually inconsistent. It's refusing to pretend a product has a fixed identity independent of what a customer is actually being promised.

The fifth step, when does each promise's revenue actually move, runs on three separate tests, and I initially collapsed two of them into one. A standard, mass-produced wardrobe, even painted a customer's chosen color, ships as revenue the day it's delivered: color is a preference, not a structural transformation, and if the deal fell through, the manufacturer could resell it with minimal loss. A wardrobe built to fit permanently into one customer's wall, with a contract clause guaranteeing payment for work done even on cancellation, recognizes gradually instead, because the manufacturer has no realistic escape route if the deal collapses, the asset is only valuable to the one person who ordered it. I first tried to explain the split by pointing at effort ("the wardrobe takes longer to build"), which is the wrong axis, a mass-produced item can take just as long. The right axis is risk: who eats the loss if the customer walks away.

A parallel pair of tests covers services rather than goods. A security guard's Monday shift is consumed the moment it happens, a replacement guard on Tuesday doesn't need to redo Monday, so revenue moves day by day. A three-month research report is the opposite: if the client fires the consulting firm two months in, the replacement firm typically can't just pick up where the last one stopped; they'll re-verify, sometimes substantially rebuild, because an unfinished analytical report isn't independently useful the way a finished day of guard duty is. I reversed the reasoning here too at first, describing the report as built from "independent pieces," when the opposite is true, it's exactly because the pieces are interdependent, each stage leaning on the one before it, that the report fails the over-time test. The third test is the simplest of the three: build on the customer's own land, and they already own the half-finished asset by definition, whatever state it's in.

None of this, none of it, differs between ASC 606 and IFRS 15. Every test above, every number, every trap, is identical in both standards. Which made it more interesting, not less, when the divergences finally showed up. The disagreement wasn't hiding in the mechanics everyone assumes matter most. It was hiding somewhere quieter.

The first real fork

Step one asks whether a "contract" exists at all for accounting purposes, and one of its five tests is whether collecting payment is probable. Both standards use that exact word. They don't mean the same thing by it.

Under IFRS 15, probable means more likely than not, over fifty percent. Under ASC 606, the same word carries a much higher bar in practice, closer to seventy-five or eighty percent and up. Hand both standards an identical customer with a sixty percent chance of paying, and IFRS 15 says yes, book the revenue; ASC 606 says no, this isn't a contract yet.

A company reporting the same transaction under both regimes would show healthy revenue in one jurisdiction's books and nothing at all in the other's.

The reason isn't arbitrary, and it's worth tracing mechanically rather than just labeling it "US conservatism," because the label alone explains nothing. Say a company books that revenue early, its stock ticks up on the strength of it, and months later the customer defaults, forcing a reversal in a later filing. Anyone who bought stock at the inflated price has now lost money on information that turned out to be wrong, and in the US specifically, that's the exact fact pattern that produces shareholder class-action suits: you showed us a rosier picture than the facts supported, and we relied on it. The US carries a heavier litigation and SEC-enforcement burden around premature revenue recognition than most IFRS jurisdictions do collectively. A higher collectibility bar isn't decoration. It's the FASB refusing to let a company get far enough into the early-recognition-then-reversal cycle for that lawsuit to become possible at all.

Whether that's the right trade is a genuinely open question, and I don't think it resolves cleanly. One case says the divergence is a feature: different jurisdictions carry different litigation and enforcement environments, and a single global threshold would be wrong for at least one of them by construction. The other case says it's a failure: if the entire point of writing one joint standard was so investors could compare a company's numbers across borders, a fork this basic, same word, opposite meaning, opposite bottom line, undermines that promise exactly where it matters most. I didn't land on a side here, and I don't think the material forces one. It's worth sitting with as unresolved, because the same tension resurfaces, sharper, a little further into the standard.

A fork that wasn't

I expected the third step, pricing when part of the consideration is variable, like a volume rebate, to produce a similar split. It didn't. Both standards apply essentially the same machinery: estimate the variable amount using whichever method actually produces a realistic number (a single most-likely outcome when there are only one or two clean possibilities, like "the rebate triggers or it doesn't"; a probability-weighted average only when there are enough spread-out possible outcomes for an average to mean something), then apply a constraint blocking the company from booking an optimistic figure unless it's highly probable that figure won't later need a significant reversal. The logic runs in both directions: a manufacturer confident a rebate will trigger has to book the lower, discounted price now, so it doesn't overstate revenue; a manufacturer with only fifty-five percent confidence isn't allowed to book that same discounted price, so it doesn't understate confidence into a number likely to reverse. Both boards are protecting against the identical failure mode, premature optimism that has to be walked back later, and neither wrote the fix any differently than the other. It's the cleanest evidence in the whole comparison that convergence, where the two boards actually agreed on the underlying risk, produced genuinely identical text, not just similar-sounding text.

The deeper fork

Licensing intellectual property runs through the same access-versus-use question as step five generally, does the customer get something frozen at the moment of transfer, or something that keeps changing because the licensor keeps working on it, but it's here that the two standards stop asking the question the same way.

ASC 606 answers it with a category. Classify the IP as functional (it has standalone usefulness on its own, software, a finished film) or symbolic (its value comes entirely from association with the licensor, a brand, a logo, a team name), and the category all but decides the outcome: functional usually means point-in-time, symbolic usually means over-time. IFRS 15 refuses the shortcut. It asks, case by case, every time: will the licensor actually do anything going forward that meaningfully changes what the customer gets out of this IP? If the honest answer is no, IFRS 15 says point-in-time, regardless of what kind of IP it technically is.

That difference in method, not just outcome, is easy to state and easy to underrate until you run a case where the two methods collide. Take a cricket league that folded a decade ago, no matches, no roster, nothing new added to the brand since, whose old logo still carries enough nostalgia that a clothing company licenses it for a one-off run of retro jerseys, with zero ongoing involvement from the league. Under ASC 606, that logo is symbolic IP by definition, no standalone functionality, value purely from association, and symbolic IP recognizes over the license period, full stop. The categorical test doesn't ask whether the league is actually doing anything; it was never designed to ask that, because once you know the type of IP, the answer is meant to follow automatically. So revenue spreads out over the license term even though, materially, nothing is happening during that term to justify the spread, the customer got everything of value on day one, and the "ongoing" portion represents activity that doesn't exist.

Under IFRS 15's version of the same question, the answer flips. Nothing ongoing is happening, the league is dead, no future activity will change what the licensee received, so the case-by-case test says point-in-time.

That's the same argument I'd parked over collectibility, wearing a sharper, more legible costume, and here, unlike there, I found myself willing to actually take a side. The categorical rule only misfires at the edge; in the much larger set of ordinary, active-brand licenses, it agrees with the activity-based test anyway, because an active brand is generating the ongoing effects the activity-based test is looking for. A trade that buys predictability across the common case at the cost of occasional strangeness in a rare one feels, on balance, like the better bargain, an auditor doesn't have to adjudicate a forward-looking guess about a licensor's future conduct, which is exactly the kind of soft prediction that turns into a dispute after the fact. I'd hold that lean loosely, though, because "the edge case is rare" is precisely the kind of claim that ages badly the one time it isn't. Collectibility never resolved the same way for me. The litigation-risk justification explains the choice; it doesn't settle whether the choice was right, and I don't think it fully will.

The one place nobody even tried

Here's where the story loops back on itself in a way I didn't expect going in.

When a contract turns bad enough that the cost of fulfilling it exceeds what the company will ever get paid, a loss contract, or in the more formal language, an onerous contract, IFRS routes the problem to IAS 37, a single, general-purpose standard covering the situation the same way regardless of industry: if the unavoidable cost of finishing exceeds the economic benefit still coming in, recognize the loss now, before it's actually incurred. One rule, one place, applied uniformly.

US GAAP has no equivalent, not a stricter version, not a looser one, no general rule at all. Coverage survives in exactly the shape it had before the 2014 reform ever happened: one piece of guidance for construction- and production-type contracts, another for extended warranty and maintenance contracts, another for certain software arrangements, another for certain insurance contracts, another for certain federal contracts. If a company's loss-making contract doesn't fall inside one of those named buckets, there is, as of today, no general requirement under US GAAP to recognize the loss in advance at all.

Both boards knew this while writing the joint standard. The FASB's own deliberations describe an explicit decision not to write general onerous-contract guidance into ASC 606, choosing instead to preserve the old industry-specific patches exactly as they stood. Which means the one disease the entire revenue-recognition reform was built to cure, a hundred narrow, industry-specific rules standing in for one coherent principle, was allowed to survive, fully intact, in precisely the single corner of the new standard where the boards looked directly at it and chose not to operate.

I don't think that was laziness. Rewriting general loss-recognition guidance touches balance-sheet questions well beyond revenue, inventory write-downs, decades of case law built around the existing fragmented rules, and unwinding all of it inside a revenue project probably wasn't a fight either board wanted to pick in 2014. But whatever the reason, the result is a small, precise irony sitting in the middle of a standard that otherwise spent over a decade trying to stamp fragmentation out: the fragmentation didn't lose. It just got fenced off.

The smaller cracks

A handful of narrower differences don't reshape the architecture the way collectibility or licensing do, but together they sketch the same personality difference in miniature.

US GAAP offers an explicit accounting policy election, a company can simply choose, once, to treat shipping and handling performed after a customer already has control of the goods as a fulfillment cost rather than a separate promise requiring its own revenue allocation. IFRS 15 offers no equivalent election; every company reasons through, case by case, whether its own shipping arrangement counts as a distinct service. A near-identical split exists for sales tax: ASC 606 lets a company elect, once, to strip similar taxes out of the transaction price entirely; IFRS 15 offers no comparable shortcut. In both cases the pattern repeats: a fixed election on the US side, ongoing judgment on the IFRS side.

There's a genuine measurement asymmetry, too, in how the two standards treat impairment on costs a company capitalizes to obtain or fulfill a contract. If those capitalized costs are later written down and the situation that caused the write-down improves, IFRS 15 requires the company to reverse the impairment. ASC 606 doesn't allow that reversal under any circumstances, consistent with US GAAP's broader, long-standing discomfort with walking back an impairment once it's booked. Small provision, same underlying instinct as the collectibility threshold: once a loss is on the books, US GAAP doesn't want it undone easily, even when the facts genuinely improved.

Nonpublic companies get materially different treatment, too. ASC 606 carves out real, specific disclosure relief for private US entities, less required detail, some transition simplifications. IFRS 15 offers nothing built into its own text for smaller entities; a separate standard exists for some smaller companies, but it isn't a scaled-down version of IFRS 15 the way the ASC 606 relief is a scaled-down version of the same standard.

None of these four differences is individually large. But they all lean the same direction: the US side keeps handing companies a fixed, elected, or bright-line answer, while the IFRS side keeps sending the same question back to judgment, applied fresh each time. That's not four unrelated footnotes. It's the same fork from collectibility and licensing, showing up as a reflex, in places too small to have earned their own debate.

Why revenue converged and almost nothing since has

It's worth asking why this particular project actually finished, when the two boards' attempts at leases and financial instruments, both launched with similar ambitions around the same period, ended in standards that look meaningfully different from each other. Leases split into IFRS 16, which puts almost every lease on the balance sheet under one model, against ASC 842, which kept the old operating-versus-finance distinction largely intact for lessees. Credit losses on financial instruments split into IFRS 9's staged, deterioration-triggered model against the FASB's CECL, which asks for a lifetime expected-loss estimate from day one. Two more joint projects, two more final products that don't match.

The likeliest explanation isn't that the two boards ran out of goodwill. It's that revenue recognition, unusually among major accounting topics, doesn't move the two numbers everyone outside the accounting department actually watches: leverage and regulatory capital. Moving leases onto the balance sheet changes a retailer's or an airline's reported debt ratios and can trip loan covenants written years before anyone imagined the rule would change. Changing how banks provision for credit losses changes regulatory capital requirements in real time. Both of those projects walked directly into rooms full of stakeholders, banks, lessees, prudential regulators, with a direct financial stake in exactly how the new rule landed, in a way revenue timing, for most companies, simply doesn't provoke. Revenue recognition reform could proceed as a comparatively contained, technical conversation between two standard-setters and their accounting constituencies. Leases and credit losses couldn't.

There's a second layer under that first one. The IASB answers to a genuinely heterogeneous, multi-jurisdiction constituency, which tends to push it toward compact, principle-level standards flexible enough to work across very different legal and economic environments. The FASB answers to a single jurisdiction with an unusually litigious enforcement culture sitting directly downstream of the SEC, which tends to push it toward specific, defensible, elected answers that leave less room for a plaintiff's lawyer to later argue the company exercised bad judgment. Revenue recognition let both instincts coexist inside one document, because the topic was narrow enough that the two philosophies rarely had to fight over the same sentence. Leases and credit losses weren't narrow the same way, and when the two instincts finally collided directly, on a topic with real balance-sheet teeth, they didn't converge. They produced two different standards and moved on.

Seen this way, ASC 606 and IFRS 15 aren't really the exception to a pattern of failed convergence. They're closer to the one case where the conditions for convergence, comparatively low political stakes, a topic narrow enough to isolate, actually held all the way through. Which makes the divergences that did survive inside this one success story more informative, not less: they're what's left over even when almost everything about the environment favored full agreement.

What it looked like once it landed

The clearest place to watch this standard actually bite was software and subscription businesses. The shift away from older US guidance, which frequently required all the revenue on a bundled software-plus-support deal to wait until every element had reliable pricing evidence, toward the new distinct-obligation model let many SaaS companies recognize revenue earlier and more granularly than before, spread across a subscription term rather than held hostage to a single pricing technicality. Telecom bundles moved in a similar direction, unwinding older guidance that had forced handset-and-service bundles into awkward, non-economic allocations. Media and licensing companies had the roughest transition of anyone, for the obvious reason: they were the industry where the functional-versus-symbolic line, and the access-versus-use question generally, actually determined a large share of reported revenue, not an edge case buried in a footnote, but the main event.

Where that leaves the comparison

None of the disagreements that survived this reform are accidents of drafting. They cluster, every time, around the same axis: US GAAP keeps reaching for a fixed, elected, or categorical answer, a specific collectibility threshold, a functional/symbolic label, a shipping policy election, an impairment rule that never reverses, while IFRS 15 keeps sending the same questions back to judgment, applied fresh to each contract.

The sharpest version of that disagreement isn't even in one of the named divergences. It's in the one place the two boards looked at squarely and chose not to touch. Onerous contracts are still governed, on the US side, by exactly the kind of scattered, industry-by-industry patchwork that revenue recognition reform spent over a decade trying to eliminate everywhere else. The boards didn't fail to notice the fragmentation there. They noticed it, and left it standing, which is, in its own way, the most honest thing about the whole convergence project. Even a standard built specifically to replace a hundred narrow rules with one coherent principle couldn't quite bring itself to finish the job in the one corner where finishing it would have meant writing something genuinely new, rather than agreeing on something both boards had already half-written before they ever sat down together.

A note on sourcing: the five-step model and the collectibility, licensing, and variable-consideration mechanics above reflect the actual text of ASC 606 and IFRS 15 as issued in 2014 and subsequently amended. The more granular comparative points, the functional/symbolic IP split, the onerous-contract gap, and the practical-expedient and disclosure differences, were checked against current professional guidance from KPMG, PwC, Deloitte's Roadmap series, and RevenueHub before being written up here, since these are exactly the kind of narrow, periodically-amended technical details worth verifying rather than recalling from memory. The reflection on why leases and financial instruments didn't converge the way revenue did is interpretation, not standard-setter testimony, a reasoned account of institutional incentives, not a claim about what either board has stated as its own reasoning.