Two Similar Metrics Can Produce a Misleading Comparison
A margin comparison can look clean and still mix GAAP and non-GAAP figures. Learn the 9-step check for whether an AI-cited financial comparison is actually valid.
Who this is for
Financial analysts and researchers — Analysts, researchers, and finance professionals reviewing AI-generated summaries of company financial results
The problem
Two numbers can both be real, both correctly quoted from a filing, and still produce a false comparison the moment one is GAAP and the other isn't. "Margin expanded from 18% to 24%" sounds like a clean trend line — until you notice the 18% was a prior-year GAAP operating margin and the 24% was this quarter's non-GAAP adjusted EBITDA margin. Two different metrics, two different bases, presented as one continuous number.
AI models are prone to exactly this error because they're pattern-matching on metric names, not on reporting basis. "Margin," "EBITDA," and "EPS" all read as the same concept to a model regardless of which adjustments sit underneath them.
How ConvergePanel helps
ConvergePanel runs the source figures through five models with an explicit basis-check step: what reporting basis is each number on, what period does it cover, and what adjustments does the company apply to it. Where the AI's original comparison mixed bases, the panel flags exactly which two figures aren't actually comparable.
How they compare
| Cited Comparison | Metric A Basis | Metric B Basis | Period Match? | Compatible? | Correction |
|---|---|---|---|---|---|
| "Margin expanded from 18% to 24%" | Prior-year GAAP operating margin | Current-quarter non-GAAP adjusted EBITDA margin | No — full year vs. single quarter | No | Compare GAAP operating margin to GAAP operating margin across the same period lengths |
| "EPS grew 30% year over year" | Prior-year basic EPS | Current-year diluted EPS | Yes — same annual period | No | Recompute both years on a diluted basis before citing the growth rate |
| "Organic growth accelerated to 12%" | Company-defined organic growth (ex-FX, ex-acquisitions) | Reported revenue growth including a recent acquisition | Yes | No | State explicitly which growth figure — organic or reported — supports the claim being made |
How it works
- 1Extract every metric cited in the AI summary, with its exact wording
- 2Identify the reporting basis of each metric — GAAP or non-GAAP, and which specific adjustments apply
- 3Identify the period each figure covers — quarter, trailing twelve months, or full year
- 4Identify any adjustments the company itself applies (one-time items, currency, organic growth)
- 5Trace each figure back to the specific line in the source filing
- 6Compare like with like — the same basis, the same period, the same adjustment treatment
- 7Flag any comparison in the AI summary that mixed incompatible bases
- 8Rewrite the conclusion using only comparable figures, or note the basis difference explicitly
- 9Send material discrepancies for qualified financial review before the summary is used
Use cases
- Checking whether a reported 'margin expansion' compares the same metric across periods
- Verifying an AI-cited EBITDA figure is GAAP EBITDA or the company's adjusted definition
- Confirming EPS comparisons use the same diluted or basic share count
- Auditing an AI-generated earnings summary before it's cited in a note or memo
Ten checks worth running on any AI-cited metric
- GAAP versus non-GAAP — is this the reported figure or the company's adjusted version
- Adjusted EBITDA versus EBITDA — what specifically is added back
- Operating income versus adjusted operating income
- Diluted versus basic EPS
- Organic versus reported growth — does the figure exclude FX and acquisitions
- Constant currency versus reported currency
- Trailing versus forward figures — actuals versus guidance
- Quarter versus full-year period — are both sides of a comparison the same length
- Recurring versus one-time items — has a one-time gain or charge been excluded
- Company-defined adjustments — does this company's "adjusted" mean something different than the peer group's
Why the error is easy to miss and easy to make
Nothing about "margin expanded" signals that the two numbers behind it come from different reporting bases — the sentence reads as a clean trend regardless of what's underneath it. That's what makes this a documentation problem, not a math problem: the individual figures are usually accurate, and the error lives entirely in the comparison between them.
It's also an easy error to make honestly. A company's own investor materials sometimes present adjusted and GAAP figures side by side without heavy emphasis on which is which, and a model summarizing quickly can pick up whichever number appears more prominently for each period.
Frequently asked questions
Is a non-GAAP metric automatically misleading?
No. Non-GAAP metrics are common and can be genuinely useful — the problem isn't using them, it's comparing one to a GAAP figure (or to a differently-adjusted non-GAAP figure) as though they were the same measurement.
How do I know what a company's 'adjusted' figure excludes?
Check the reconciliation table companies are required to provide alongside non-GAAP figures — it lists exactly what's added back or excluded. An AI summary that cites the adjusted number without checking the reconciliation is the most common source of this error.
What if the AI summary doesn't specify which basis a figure uses?
Treat that as a flag on its own. If the basis isn't stated, trace the figure back to the source filing before using the number for anything — don't assume it matches whatever basis the comparison figure uses.
Does this only apply to earnings summaries?
It applies anywhere metrics get compared across time or across companies — investor decks, comp sets, forecast narratives. Earnings summaries are simply where it shows up most often because every quarter provides a new basis-mixing opportunity.
Can ConvergePanel confirm which basis is the 'correct' one to use?
No. It compares model interpretations and traces figures to their stated source, but which basis is appropriate for a given analysis is an accounting and financial-judgment question that requires a qualified financial professional.
Explore related pages
ConvergePanel provides AI-assisted verification for informational purposes only. Not forensic analysis. Not legal evidence.
More in How-To
How to Verify a Viral Claim with AI
How does AI claim verification actually work? Learn the mechanics: independent model queries, consensus scoring, and how to read disagreement as a research signal.
How to Review a Suspicious Video with AI
Use AI-assisted review to check suspicious videos for context, visual claims, manipulation risk, and source uncertainty.
How to Verify a Viral Claim Before You Share It
Viral claims travel six times faster than corrections. Check the source, date, and model disagreement in under two minutes before you share.