Buyers cannot calculate net financial returns from any of the four Claude customer accounts examined in this analysis—even where the headline benefit is expressed in money. The documents support narrower conclusions: a rebuild costing roughly half an estimated alternative, reported annualized procurement savings, and faster staff workflows. Those are different kinds of evidence, not interchangeable measures of return.
One question, four different measures
Superpower Daily compared four supplier-published documents covering LG CNS, Duvo, Cyera and Airtree. The question was deliberately narrow: which reported benefits support financial calculations after separating estimated rebuild costs, annualized savings, staff time and adoption? We classified each account by its benefit, baseline, measurement period, supported arithmetic, attributable costs and potential overlap. This was a document review, not a deployment test or an audit of customer accounts.
The nominal evidence cutoff was November 1, 2026, at 16:30:54 UTC. The supplied snapshots were retrieved less than a minute later, at 16:31:39 UTC, and their URLs checked live; exact historical versions at the cutoff were not established. The sample is limited to these four documents, whose outcomes have not been independently verified. Missing financial disclosure does not mean a deployment failed. It means its net return remains publicly indeterminate.
LG CNS: half of an estimated alternative
LG CNS supplies the clearest relative project-cost comparison. Its case study describes rebuilding a Korean construction company’s 20-year-old project-management system with Claude Code. The migration ran from November 2025 to June 2026 and reportedly cost roughly 50% of a conventional rebuild. That conventional route was an estimate: LG CNS judged it would require at least twice the available time and money, with costs of several billion Korean won.
The account contains concrete delivery evidence: 2,888 of 2,913 software interfaces converted, 1,340 screens migrated, and a June go-live. It also describes engineering work around the model, including task planning, code generation, testing, verification and quality management. The reported result belongs to that delivery system, not to an isolated model operating without engineering support.
The supported calculation is approximately 50% lower cost against an estimated alternative—not a reproducible amount of money saved. Neither exact actual spending nor the exact conventional baseline is disclosed, and cost inclusions are unspecified. The seven-month project period is also not a measured period of returns after deployment. LG CNS’s stronger qualitative argument is feasibility: it says the approach made a previously difficult-to-start project possible.
Duvo: a yearly savings rate, not three months of cash
Duvo’s account supplies monetary benefits, but in annualized terms. It reports €1.45 million in annualized savings at retailer Rohlik in the first week, from price monitoring and supplier outreach across more than 120 products and 15 suppliers. Across three months, the reported figure grew to more than €2.8 million annualized. These are successive descriptions of an expanding deployment; adding them would count overlapping benefits.
The mechanism differs from simply completing an existing task faster. Rohlik’s commodity-price monitoring previously happened quarterly at best, and sometimes not at all. Duvo says its agents now monitor continuously, prepare negotiation cases and initiate supplier outreach. That is evidence of a claimed benefit from previously neglected work. High-risk actions still require human approval, so the workflow includes an explicit human decision point.
An annualized figure expresses a yearly rate, not cash savings realized during the observation window. The document does not disclose the annualization workings, assumptions about persistence, purchasing baseline or customer costs. It also leaves unspecified whether expenses were deducted. Its separate estimate of 40% staff capacity released is a cross-customer average—not another Rohlik monetary benefit that can safely be added to the savings figure.
Cyera and Airtree: time arithmetic without a price
Cyera’s case study supports a straightforward subtraction. Daily triage of its internal Ask Claude channel reportedly fell from five–six hours to roughly 30 minutes. A scheduled task checks questions against knowledge bases and past answers, then drafts replies for the owner’s review. The implied time released is approximately 4.5–5.5 hours daily for that workflow, not a company-wide estimate.
The financial conversion remains open. Cyera supplies no observation-window length, labor valuation or demonstrated expenditure reduction for the triage saving. Its 88% weekly usage across roughly 1,500 employees measures adoption, not productivity or financial return. The surrounding work is substantial: permissions mapping, a governed data layer, training and office hours. Those activities are described but not priced.
Airtree’s interview offers another exact subtraction: board summaries reportedly fell from 30 minutes to two, implying 28 minutes released per summary. Marketing reporting separately saves two hours monthly. Research falling from two days to unspecified minutes cannot support an equally precise calculation. Nor can the board-summary saving be extrapolated into a monthly total without task volumes. Airtree requires human verification before outputs inform decisions or are published, but the reported durations do not specify whether that review is included.
The next disclosure needs a matched ledger
The accounts should not be ranked as though they measured the same outcome. LG CNS compares project costs against an estimated alternative. Duvo reports a monetary benefit rate. Cyera and Airtree report staff-time gains, while Cyera also quantifies adoption. Each answers a different operational question. None connects sufficiently matched benefits and full attributable costs to a reproducible net-return calculation.
Reader comments
Newest comments first. Replies stay oldest first.