"Deployed Claude across the platform in under one month." / "We found that Claude is one of the highest performing models on our evaluations," said Niko Grupen, Head of Applied AI at Harvey.
"we are checking in 30 to 40% more [code], and we are not seeing any degradation. In fact, we are seeing the same quality as we measure." / "We route queries to the right model for the right task rather than running every interaction through the highest cost model."
"a lot of really hard work, we've been able to bring down the cost of video call. It is now under $0.01 per video call. And the reason for that is mainly a move towards open source models." / "For gross margin, we now expect to end the year closer to 70% as compared to the 69% we initially expected as we drive more AI content into our products offset by AI cost savings."
Vodafone (10,000+ field technicians, ~80,000km driven daily combined) built "Field Technician Assist" with system integrator Celfocus on Google Cloud, using GenAI/ML to turn unstructured incident data into recommendations. The tool presents engineers with the three best solutions for an incident at the troubleshooting stage of a 4-stage field-service process (travel, onsite troubleshooting, quality verification, ticket closing), engineers pick/rate the recommendation, feeding continuous improvement.
示唆
Results were presented by Vodafone exec Emilio Varas at TM Forum's DTW Ignite 2025 conference and reported by trade outlet TelcoTitans/VodafoneWatch. The 28% figure is an improvement over an earlier 18% reduction seen at proof-of-concept stage, showing gains compounding as the tool matured from PoC to MVP. Less-experienced engineers gaining access to the same AI-derived recommendations as veterans is framed as "harmonising quality" across a large, skill-variable field workforce — the value comes from standardizing a fleet's worst-case performance up toward its best, not from a single breakthrough fix.
成果・金額自己申告
28% reduction in repeated site visits achieved (up from 18% at proof-of-concept), targeting 35%; ~10 minutes average time saved per incident per engineer; deployed context covers 10,000+ field technicians and ~80,000km of daily combined travel.
"Vodafone has seen a 28% reduction in repeated site visits when using the app, and the amount of time spent by engineers on each incident has reduced by an average of ten minutes when using Field Technician Assist... the 28% reduction in repeated visits associated with the minimum viable product release representing a significant uptick on the 18% reduction initially seen with the proof of concept. Through continued feedback and enhancement, the operator is now targeting 35% reduction in repeated visits."
ブラジルの洗剤・石鹸メーカー Ypê(顧客22,000社超)が既存 SAP ERP を置き換えずに Rimini Street の「Agentic UX」AI レイヤーを追加導入。電子受注の価格/データ誤りによる例外処理を、SAPのブロック済み注文をServiceNowへ自動同期→営業担当にメールで判断オプション付き通知→AIが履歴データから是正案(価格修正等)を提案→承認済み変更をSAPへ自動書き戻し、という形で自動化。導入はキックオフから1ヶ月未満。
Publishers Weekly（独立系トレード誌、記者 Jim Milliot）が2026年6月16日、Wiley のFY2026(2026年4月30日期)決算発表・アナリスト向け電話会見の内容を報道。AIライセンス収益（IQVIA・OpenEvidenceなど19社の企業顧客+LLM開発企業4社と契約）と、法務・マーケ・コンテンツ運用での社内AI活用によるコスト削減の両方を、CEO/CFOの発言を引用しつつ伝えている。
FY2026 AIライセンス収益 $49 million。純利益は前年 $84.2 million → $221.6 million（+163%）。全社売上は $1.67 billion で前年比横ばい。研究部門売上は+5%で $1.13 billion。一方、学習部門は-7%（学術部門 $96.1 million で-4%、プロフェッショナル部門 $56.2 million で-9%）。AIによる恩恵で全社共通費が通期$23 million・第4四半期$9 million減。※候補にあった「adjusted EBITDA margin +220bps to 26.2%」は本文に記載なし（未確認のため不採用）。
"For the fiscal year ended April 30, 2026, Wiley said it generated $49 million in AI licensing deals, citing agreements with IQVIA, OpenEvidence, and a growing roster of corporate customers as key factors in its growth." / "Wiley's net income skyrocketing 163% to $221.6 million, from $84.2 million a year ago." / "Helped by AI, Albright said full year and fourth quarter corporate unallocated expenses were down $23 million and $9 million, respectively."
UAEのCommercial Bank of Dubai（CBD）が約300人にMicrosoft 365 Copilotライセンスを配布し、早期採用者を「Copilotチャンピオン」に任命。3回の「Prompt-a-thon」（部門別プロンプト改善→経営陣と従業員のペアリングによる部門横断協力）を実施し、有効なプロンプト集「Super Prompts」ライブラリを全社共有した。月次メール施策や「3C CBD Copilot Week」（Clarity/Control/Confidence）でAIリテラシー教育も継続した。
"84% of our respondents reported a 10%-20% productivity increase from using the product" / "saved over 2,300 person-hours since the start of the automation program" / "This process has reduced the audit report writing efforts by 30% so far" / "HR was able to engage with employees within a month of the survey"
オンライン取引を手がける英IG Groupが、複数のAIプロバイダーを比較評価したうえで Claude Enterprise を全社導入した。12月から4月にかけて段階的に研修を実施し、組織全体で利用率60%を目標にロールアウト。人事（評価テンプレート生成）、マーケティング（コンテンツ制作・多言語翻訳）、アナリティクス（SQLクエリ作成・データ処理の自動化）、開発（QAの高速化）に適用している。
"Super.com saves 1,500+ hours monthly and onboards employees 20% faster with Glean" / "By using Glean's AI-powered search, employees spent less time hunting for answers and more time on high-impact work. On average, employees found information 20 minutes faster per day, saving more than 1,500 hours monthly." / metric block: "17x" / "ROI"
弁護士は週8時間超(over eight hours weekly)の時間を節約と報告。新規案件対応では他事務所が1週間かかるところを48時間未満(less than 48 hours)で対応し受注。
"Attorneys report saving over eight hours weekly—time that is reinvested into high-value work like client counseling, men[torship]..." / "...key facts and craft a tailored, persuasive response, delivered in less than 48 hours, when other firms needed a week to respond. LPHS won the business – and a long-[term client]..."
"Vega accelerates threat investigations 44x while cutting legacy SIEM costs by 82% with Claude." / "Scaling frontier reasoning to every detection is the ultimate win for our customers." — Eli Rozen, Co-founder & CTO, Vega Security
"With Gemini 2.5 Flash on Vertex AI, CN2U.AI has reduced search and recommendation response times by more than 50%, improved response accuracy by over 10% and achieved a 15% increase in CTR, a key ecommerce performance metric."
In its best performance, the AI agents beat the 60-40 allocation by 0.7 percentage point a year, with lower volatility... every hypothetical backtest outperformed a 60-40 allocation.
Attorney General Dave Sunday stated: "Consumers deserve transparency and fairness throughout the insurance process, especially when losing coverage can leave someone unknowingly uninsured and vulnerable to penalties and financial risk."
DOJ Civil Rights Division の Protecting U.S. Workers Initiative の一環として、バージニア州のIT企業 Elegant Enterprise-Wide Solutions Inc. と和解。同社が使用したAI生成の求人広告に「H-1B・OPT・H-4ビザ保持者のみ考慮」等の市民権による制限が含まれ、Immigration and Nationality Act(INA)の市民権ステータス差別禁止規定に違反していたと認定。同社は米財務省へ民事制裁金9,460ドルを2回に分けて支払い、60日以内に採用ポリシーを見直すことに合意。
民事制裁金 $9,460(2回分割で米財務省へ支払い)。DOJ Protecting U.S. Workers Initiative 下の複数件目の和解(件数は情報源により8件目/13件目と表記揺れあり、要再確認)。
"It is unconscionable for companies to illegally exclude U.S. workers when recruiting and hiring. This Department of Justice will not tolerate discriminating against U.S. workers, no matter who – or what – drafts a job advertisement, or whether it is an employee, a recruiter, or an AI tool." — Assistant Attorney General Harmeet K. Dhillon
米国第6巡回控訴裁判所(6th U.S. Circuit Court of Appeals)が、2026年3月13日の意見書で、Whiting v. City of Athens, Tennessee(2022年花火大会を巡る紛争の統合控訴事件)の控訴趣意書に26件以上のAI生成による架空・誤引用の判例を記載したテネシー州の弁護士2名(Van R. Irion、Russ Egli)に対し、それぞれ$15,000(合計$30,000)の懲罰的制裁金の支払いを命令。加えて被控訴人側弁護士費用の連帯弁済、及びRule 38上「利用可能な最も厳しい罰」とされる二重訴訟費用の支払いも命じ、懲戒手続き検討のため首席判事への回付も指示した。元の候補にあった「控訴を棄却した」という点は、参照した複数の記事(Reuters/LawNext/National Law Review/Sixth Circuit Appellate Blog)いずれにも本案自体の処分(棄却/認容)についての明記がなく未確認。
"The company has deployed AI agents in its various operations as part of an efficiency programme that aims to boost productivity by 5%." / "Annualised savings of between 100 and 150 million francs are expected by 2027."
Google CloudがHSBCと共同でAnti-Money Laundering AI(AML AI)を試験導入。取引データ・KYC(顧客確認)情報・過去の疑わしい活動記録をもとにAIがリスクスコアを算出し、アナリストが検証する仕組み。既存のルールベース検知システムと並行稼働させ、段階的に導入することでリスクを抑えつつ信頼性を構築した。
Copilot will save 13,000+ hours per month of post-call admin time... the breakeven point is a month... the fastest I've ever seen with a piece of technology.
Starbucksが、シアトルのスタートアップNomadGo製のAIアプリを北米全店規模で導入し、店舗のミルクやシロップなど飲料材料の在庫をタブレット等でカウントさせていた。しかし本文によれば同アプリは「often miscounted or mislabeled items, failing to identify the presence of bottles on shelves（棚上の商品の存在を認識できず、頻繁に誤カウント・誤ラベル付けした）」状態で、過剰カウント時には本来補充すべき商品が発注されない不具合も発生。2025年9月の導入発表から9か月後の2026年5月、Starbucksは「静かに」このAI在庫管理システムを廃止し、単一の（人手による）カウント方式に戻す運用判断を下した。
示唆
在庫の自動カウントという一見定型的な業務でも、棚上の類似商品（ミルク・シロップ等)の識別精度が現場で維持できず、精度は導入後「悪化していった」("It started off not particularly accurate and got less accurate over time." — 現場責任者Carl Addison)。バリスタからの苦情が積み重なった末、企業側は大々的な発表なしに「静かに」ロールバックしており、AI導入の成功事例だけでなく、こうした無言の巻き戻しも実態として存在することを示す一次事例。
saving 3 million hours annually; 250 million vulnerabilities triaged in 20 hours (83% eliminated as benign); 2.5 million SCA findings from 5,000 images processed in under 1 hour (99.5% eliminated as false positives)
"Investigated and triaged 250 million vulnerabilities surfaced by enterprise scanners...in 20 hours...eliminating 83% as benign" / "Eliminated 99.5% of false positives" from "2.5 million Software Composition Analysis (SCA) findings from 5,000 images in under an hour" / "saving 3 million hours annually"
Walmart's "Self-Healing Inventory" system continuously monitors store-level stock across its network (example given: Mexico City, where shelf space is scarce). When it detects overstocks building up at one location, it automatically reroutes that supply to stores that actually need it, before the excess turns into waste.
示唆
The article frames this as one piece of Walmart's broader "supply chain playbook" going global (alongside quotes from CTO Vinod Bidarkoppa and CPO Tim Simmons). The $55M figure is stated as a standalone fact with no methodology, time period, or third-party verification given. The system is described as a general overstock-rerouting tool across Walmart's store network — the article does not specifically scope it to perishables/food waste as the hint suggested; that's a narrower framing than what's actually written.
成果・金額自己申告
$55M+ saved so far (per Walmart's own statement: "That one system alone has already saved Walmart more than $55 million")
"In Mexico City, shelf space is scarce, and timing is everything. That's why systems like Self-Healing Inventory keep watch around the clock. When overstocks appear, it automatically reroutes supply to the stores that need it most — before the excess becomes waste. That one system alone has already saved Walmart more than $55 million."
Customer outcomes from AI-powered solutions are compared against control groups, ensuring that the S$1 billion figure reflects tangible, measurable benefits rather than theoretical estimates.
Commonwealth Bank of Australiaの社内データサイエンス・エンジニアリングチームが3カ月でAIエージェントを開発。Snowflakeのデータ基盤上で1日8000万件以上の取引・詐欺報告シグナルを監視し、疑わしいパターンを検出して新しい検出ルールを自動生成する。生成されたルールは人間の監査(human-in-the-loop)を経て実装される。記事本文にAnthropicやClaudeへの言及はなく、使用モデルは明示されていない。
System monitors more than 80 million signals every day...fraud losses were down more than 20 per cent in the first half of FY26 compared to the first half of FY25
"Claude became the overwhelming choice...The platform processes 100 billion tokens monthly, and Claude is the preferred model" — Justin Watts, Distinguished Engineer
"37% improvement in median development time for 170-person engineering group" / "When the pilot group achieved a 42% improvement in development time, ComplyAdvantage then rolled out the solution more broadly in production."
.jpg)
