What happened
Bloomberg reported on August 7, 2026 that with nearly 90% of S&P 500 companies having reported second-quarter 2026 earnings, an analysis by research firm 22V Research LLC found 25 index companies had explicitly quantified a margin impact from AI. The average lift those 25 companies cited was 180 basis points, or 1.8 percentage points of margin expansion. Excluding companies that bundled AI together with other efficiency programs, the typical gain was 150 basis points. The comparison to the prior quarter is the sharper part of the story. In the first quarter of 2026, only 17 companies in the index had quantified anything, and the average gain they cited was just 20 basis points. Both the number of companies willing to attach a specific figure, and the size of the figure itself, grew substantially in a single quarter. 22V Research president Dennis DeBusschere said that if this magnitude of improvement generalized across the full index, it would imply the S&P 500 is undervalued by at least 10%. He was careful to frame it as directional: "Direction matters more than precision in these early estimates." The named examples are what make the story concrete rather than abstract. Waste Management said its SmartTruck routing and dispatch system is running at more than $300 million in annual EBITDA impact, through better routing, customer service, and lower costs. Freight and logistics broker CH Robinson Worldwide cited a 60% productivity improvement tied to AI since 2022. Cybersecurity firm Fortinet reported its second-quarter operating margin rose 490 basis points. Insurance broker Willis Towers Watson said it expects $400 million in cost savings, "thanks mostly to process automation." Credit bureau Equifax's CEO told investors AI-related savings were beginning to show up in 2026. Johnson Controls projected 260 basis points of margin expansion and has seen its stock rise a cumulative 11% since its earnings release.
Why it matters for business owners
For most of the last two years, the dominant story about AI spending was uncertainty. A small number of technology companies were spending an enormous amount of capital on AI infrastructure, and nobody outside those companies had clear proof the spending was translating into a financial return anywhere, let alone in ordinary, non-technology businesses. This is the first earnings season with real, named, quantified evidence to the contrary. A waste hauling company, a freight broker, an insurance brokerage, and a credit bureau are now on record, in regulated earnings disclosures, describing specific dollar and percentage-point improvements tied to AI. That is a materially different, and more useful, kind of evidence than a vendor's sales pitch or a general industry survey. It gives an owner something concrete to compare their own AI efforts against.
What owners should not misunderstand
Do not read "AI margin gains are showing up in the S&P 500" as "AI adoption pays for itself." Only 25 of roughly 500 companies in the index, about 5%, could actually put a number on it this quarter. These are large, well-resourced public companies with finance teams built specifically to measure this kind of thing, reporting to investors who are actively asking the question. If 95% of them still cannot answer it, the difficulty is real, not a matter of trying harder. Do not miss what every company that could prove a gain has in common: a single, specific, named process, not a general AI rollout. Waste Management does not say "AI helped the company." It names SmartTruck, a defined routing and dispatch system, tied to a defined EBITDA figure. CH Robinson names one productivity metric it has tracked since 2022. That specificity is exactly what made the gain measurable, and it is exactly what is missing from most companies' broader, less defined AI efforts. Do not treat the reported figures as a net return. The reporting is explicit that these margin gains are not weighed against what each company spent on the AI tools, infrastructure, and vendor contracts that produced them. A 180-basis-point gain on a project that cost more than 180 basis points of margin to build is not a win, and the current disclosures do not tell you which is which. Do not assume durability yet. This is one quarter of self-reported disclosure, framed by the companies themselves on their own earnings calls, with no independent test yet for whether the gains hold up once the first wave of AI-driven changes has fully worked through the business.
The operational lesson
Look past the headline number and the pattern across every company that could prove a gain is the same: pick one bounded, high-volume process that was already being measured, apply AI to that process specifically, and track the same metric before and after. Waste Management already tracked route efficiency and service cost before AI touched the SmartTruck system. Adding AI to that specific, already-instrumented process is what let the company show a real before-and-after number. Companies that instead rolled AI out broadly, as a general capability employees could use however they wanted, are very likely the same companies that, one quarter later, still cannot name a figure. Broad, undirected AI adoption does not fail to produce value automatically. It fails to produce provable value, because nobody defined what to measure before the tool arrived. That is the discipline separating a business that can prove AI is working from one that only hopes it is: name the process, know the baseline number today, apply the tool to that process specifically, and check the same number again later. Skipping that sequence is a large part of why the best-resourced companies in the country, with finance teams built for exactly this kind of measurement, still could not answer the question for the overwhelming majority of their own AI spending this quarter.
What a serious business should do next
Before increasing AI spend or trying to build a company-wide AI strategy, pick one workflow your business already measures today, a process with a real number attached before AI touches it. Cost per order fulfilled, average handling time on a support queue, cost per lead qualified, invoices processed per hour, whatever your business already tracks. Apply AI narrowly to that one workflow first, rather than broadly across the team all at once. Keep the process and the metric the same on both sides of the comparison, so that a change in the number actually reflects the AI, not some other change happening at the same time. Set a review date 60 to 90 days out to check the same metric again. If you cannot point to a number that moved, you do not yet have evidence AI is working in that workflow, regardless of how much the team likes using the tool. Net the apparent gain against the real cost: subscription and API fees, implementation time, and any new vendor or infrastructure spend the project required. A margin gain that costs more than it saves is not a margin gain. Treat any AI vendor or consultant who cannot help you define this kind of before-and-after measurement, tied to a number your business already tracks, as a signal to look elsewhere. Every company that proved a gain this quarter started with a measured process, not a tool.
The Atlacis view
The S&P 500 data is the clearest evidence yet that AI can move the numbers that matter to a business, and it is just as clearly evidence that this does not happen automatically or by adopting AI broadly. The companies showing real gains did not roll AI out company-wide and hope. They pointed it at one process they already understood well enough to measure, and they can now show a board or an earnings call exactly what happened as a result. That is the same discipline Atlacis brings to a medium-size business. Before recommending a tool, a model, or a deployment, we help owners name the specific workflow in question, establish what it actually costs today, and set up a measurement that will tell them honestly, in 60 or 90 days, whether the spend was worth it. If the largest, best-resourced companies in the country are still struggling to prove that for 95% of their own AI spending, a business without a dedicated team measuring this in-house needs that discipline more, not less, before the next AI purchase.
The short version
- Bloomberg reported on August 7, 2026 that an analysis by 22V Research found 25 S&P 500 companies quantified a real AI-driven margin gain this earnings season, averaging 180 basis points, up sharply from 17 companies averaging 20 basis points in the prior quarter.
- Named examples span outside technology: Waste Management's SmartTruck routing system, CH Robinson's logistics productivity gains, Fortinet's operating margin increase, Willis Towers Watson's process-automation savings, and Johnson Controls' projected margin expansion.
- Only about 5% of S&P 500 companies, fewer than 1 in 20, could actually quantify an AI margin gain this quarter, despite having finance teams built for exactly this kind of measurement.
- Every company that proved a gain points to one specific, previously measured process, not a general AI rollout. That specificity is what made the gain provable.
- The reported figures are not netted against what each company spent to produce them, and there is no data yet on whether the gains hold up over time. Treat the numbers as directional evidence, not a finished return-on-investment case.
- The practical lesson for any business: pick one already-measured workflow, apply AI to it specifically, set a 60 to 90 day review, and net the gain against the real cost before assuming AI spend anywhere else in the business is paying off too.