In 2015, Kraft Heinz carried roughly $44 billion of goodwill and $49 billion of intangible brands on its balance sheet. The Oscar Mayer and Kraft names were treated as nearly indestructible assets, tested for impairment against management's own projections of pricing power that no longer existed. In February 2019, the company took a $15.4 billion writedown, cut its dividend, and disclosed an SEC subpoena. The stock fell 27% in a day. Nothing about the brands had changed overnight. What changed was the *judgement* — the assumptions management had been willing to defend, quarter after quarter, until reality forced a recalibration.
This is the uncomfortable truth of technical accounting: the largest numbers on your balance sheet and the most consequential charges on your income statement are not measured, they are *estimated*. Goodwill, provisions, expected credit losses, warranty reserves, decommissioning liabilities — each is a defensible opinion dressed as a figure. The CFO's job is not to compute these. It is to *govern the machinery* that produces them so that the outputs are consistent across periods, defensible under challenge, and free of the slow-motion bias that turned Kraft Heinz's brands into a cliff.
Not all estimates carry the same governance weight. The CFO's first discipline is triage: distinguishing the estimates that move the enterprise from the ones that merely tidy the ledger. Three families dominate the risk.
Under both IAS 36 and ASC 350, goodwill is not amortised — it sits until an event or an annual test proves it is worth less than its carrying amount. That single design choice pushes enormous judgement into two variables: the discount rate and the terminal growth assumption embedded in the recoverable amount (value-in-use) or fair-value-less-costs-of-disposal calculation.
The danger is asymmetry. A 50-basis-point shift in the discount rate, or a half-point in perpetual growth, can swing a cash-generating unit from comfortably above its carrying value to underwater. Management chooses those inputs. And because an impairment is an admission that a prior acquisition or investment disappointed, there is a structural incentive to keep the assumptions just optimistic enough to avoid the charge. This is not fraud. It is *anchoring* — the human tendency to defend last year's story with this year's slightly stretched inputs.
The 2020 wave of pandemic-era impairments — Boeing, Carnival, oil majors writing down tens of billions — was revealing precisely because so many charges landed at once. When everyone recognises simultaneously, it suggests the triggers were external and undeniable. The harder question for a CFO is the *quiet* period: the year when your CGU's headroom has been eroding for three consecutive tests and nobody wants to be the one to call it.
IAS 37 sets a deceptively clean bar: recognise a provision when there is a present obligation from a past event, an outflow is *probable*, and the amount can be *reliably estimated*. Every word in that sentence is a battleground. Is the obligation "present" or merely possible? Is litigation "probable" or "possible" — the difference between a booked liability and a footnote? US GAAP (ASC 450) uses a slightly different vocabulary ("probable" means a higher threshold than under IFRS), which is why cross-listed companies often disclose litigation differently in the two regimes for the identical fact pattern.
Restructuring provisions are the classic manipulation vector — the "big bath." A new CEO takes an oversized charge in year one, over-provisioning for redundancies and site closures, then releases the excess into earnings across the following two years to manufacture a recovery narrative. IAS 37 constrains this by requiring a detailed formal plan and a valid expectation created in those affected before a restructuring provision can be booked. The CFO's role is to hold that line even when the incoming CEO would prefer a generous bath.
IFRS 9's expected credit loss model and the CECL model under US GAAP pushed credit provisioning from *incurred* loss to *forward-looking* loss. That was a deliberate post-2008 reform — but it also imported macroeconomic forecasting into the accounts. A bank's loan loss provision now depends on management's weighting of economic scenarios. Warranty reserves depend on failure-rate curves. Pension obligations depend on discount rates and mortality tables. Each is an estimate with a model behind it, and each model has assumptions that someone chose.
Fundamentals told you *what* an impairment is. This is about the control architecture that makes yours defensible. Governance of estimates rests on four pillars.
The most common failure mode is that the person who wants the answer also controls the inputs. If a business unit head owns both the cash-flow forecast that *feeds* the impairment test and the accountability for the *result* of that test, you have built a conflict directly into the process. Best practice: FP&A owns the operating forecast, a central technical accounting or valuation team owns the model mechanics and the discount rate, and neither reports to the divisional leader whose CGU is under test. The assumptions become a negotiation between functions rather than a self-assessment.
Here is the single sharpest control a CFO can impose. The cash flows used in your value-in-use calculation must reconcile to the board-approved budget and long-range plan. If your CGU is projecting 6% growth in the impairment model while the same unit's operating plan — the one used for compensation and capital allocation — assumes 2%, you have a documented inconsistency that an auditor, a regulator, or a plaintiff's lawyer will find in discovery.
IAS 36 explicitly requires that cash-flow projections be based on the most recent approved budgets and forecasts, with growth beyond that period not exceeding the long-term average growth rate for the products, industries, or country — unless a higher rate is justified. In practice, this means the CFO must be able to answer one question in any audit committee meeting: *"Are the numbers in our impairment test the same numbers we're managing the business by?"* When they diverge, the divergence is either a signal of optimism bias in the test or a signal that the operating plan is sandbagged. Both are worth knowing.
A point estimate is not a governance artefact. A *range* is. For every material estimate, the CFO should require disclosed sensitivities: how much does headroom change if the discount rate rises 100bps, if terminal growth falls 50bps, if the litigation probability shifts from 40% to 60%? IAS 36 mandates sensitivity disclosure when a reasonably possible change would eliminate headroom. Treat that as a floor, not a ceiling.
Then back-test. Every year, compare last year's estimates to what actually happened. Was the warranty reserve adequate or was it consistently released? Did the restructuring cost what you provisioned? Did the CGU deliver the cash flows the impairment model relied on? A persistent one-directional error — reserves always too high, forecasts always too optimistic — is the fingerprint of bias, and it is the first thing a sophisticated audit committee should demand.
Judgement that cannot be reconstructed is indefensible. The estimate is only as strong as the documentation created *at the time* — not the memo assembled after a regulator calls. For every material judgement, the file should capture: the assumptions and their source, the alternatives considered and why they were rejected, the sensitivity analysis, the sign-offs, and the specific accounting literature relied upon. When the SEC subpoenaed Kraft Heinz, what mattered was not the impairment itself but whether the *prior* decisions not to impair were supportable at the time they were made.
Knowledge check
1. According to the lesson, what is the CFO's primary role with respect to accounting estimates like goodwill impairment and provisions?
2. The Kraft Heinz example illustrates which core conceptual point about large balance-sheet items and impairment charges?
3. Why does the non-amortisation of goodwill under IAS 36 and ASC 350 concentrate so much judgement risk?
4. Select ALL correct answers about the concept of 'triage' in governing accounting estimates.
Select all the correct answers.
5. Select ALL correct answers about why impairment and provision estimates are described as 'a verdict, not a fact.'
Select all the correct answers.
Governance frameworks are necessary but insufficient. On Monday morning, the CFO is making live calls under pressure. Three disciplines separate the CFOs who get this right from the ones who become the case study.
Impairment is not only an annual event. A triggering event — a lost major customer, a regulatory shift, a market cap that falls below book value, a sustained decline in a CGU's performance — obligates an interim test. The disciplined CFO maintains a *trigger watchlist* reviewed quarterly: for each material CGU and indefinite-lived intangible, what are the observable indicators that would compel a test, and where do we stand against each? This converts impairment from a reactive scramble into a monitored risk. A market capitalisation persistently below net asset value is the loudest trigger of all — if the market says the whole is worth less than the sum of your carried parts, an auditor will not accept "we're confident in our model."
A jurisdictional trap that catches even experienced finance leaders: under IFRS, impairments of most assets (other than goodwill) can be *reversed* if the recoverable amount recovers — but goodwill impairment can never be reversed. Under US GAAP, *no* impairment reversals are permitted for goodwill or long-lived assets held for use. This changes the strategic calculus. Under IFRS, an aggressive impairment of a plant taken in a downturn can be partially unwound in the recovery, which somewhat limits big-bath incentives on those assets. On goodwill, the charge is permanent — which is exactly why the pressure to *delay* is so intense. Know which lever is irreversible before you pull it.
Estimates are where the auditor's judgement meets yours, and increasingly where they diverge publicly. Impairment and provisioning now routinely appear as Critical Audit Matters (CAMs) in the US and Key Audit Matters (KAMs) under ISA 701 — the auditor is required to describe, in the audit report itself, the estimates that involved the most difficult and subjective judgement. That means your discount-rate assumption may be narrated to investors by your auditor. The CFO should never be surprised by a CAM. Walk into the audit committee with your *own* framing of the key judgements, your sensitivities, and your back-testing evidence *before* the auditor presents theirs. The CFO who lets the auditor define the judgement narrative has already lost control of it.
The through-line across all of this: consistency is credibility. The assumptions you use to justify *not* impairing in a good year are the same assumptions you will be held to in a bad one. A CFO who stretches terminal growth to preserve headroom this quarter has written the plaintiff's opening argument for next year.
1. Reconcile every impairment model to the board-approved plan. If the growth rate in your value-in-use calculation differs from the plan you manage and pay the business against, that gap is your single biggest audit and litigation exposure. Close it or explain it in writing.
2. Separate assumption-setting from result-accountability. Never let the leader who owns the CGU's performance also own the discount rate and cash flows that test it. Build the conflict out of the process structurally.
3. Back-test relentlessly and watch for one-directional error. Reserves that are always released and forecasts that are always missed are not bad luck — they are the statistical signature of bias, and they will be found.
4. Maintain a live trigger watchlist. Impairment is a monitored risk, not an annual surprise. A market cap below book value is a trigger you cannot argue away with a model.
5. Control the judgement narrative before the auditor does. With CAMs and KAMs, your subjective estimates are disclosed to investors in the audit report. Arrive at the audit committee with your own framing, sensitivities, and evidence file already built.