← Multimodal Sources 🕐 6 min read
Multimodal Sources

AWS re:Invent 2024–2025: What F500 Companies Said On-Record About Enterprise AI

Brandon Sneider · April 2026

Seven sessions across re:Invent 2024 and 2025 yielded on-record metrics from named companies.

Executive Summary

  • Named F500 companies at AWS re:Invent disclosed production AI metrics that map onto patterns already visible in the research corpus: data readiness is the bottleneck, not model capability; workflow redesign drives value; and HITL governance scales adoption in regulated industries.
  • United Airlines compressed delay messaging from 5–15 minutes to 28 seconds and expanded a mainframe access tool from 3,000 to 15,000 frontline employees — both built on Amazon Bedrock. Capital One deployed GenAI workflows to 10,000+ customer service agents with explicit observability guardrails.
  • In financial services, MUFG achieved 10x deal generation at maintained conversion rates; Verafin cut fraud investigation time by 70%. Cleveland Clinic reached 90% adoption of AI-drafted discharge summaries.
  • Infosys surveyed 1,500 CXOs and found less than 50% are satisfied with their AI results, less than 20% of PoCs reach production, and only 12% have an AI strategy in place. These numbers corroborate the BCG 5% / McKinsey 6% high-performer findings from independent sources.

The Production Metrics

Seven sessions across re:Invent 2024 and 2025 yielded on-record metrics from named companies. The pattern: companies that redesigned workflows around AI captured measurable value; companies that bolted AI onto existing processes saw limited returns.

Operations and Customer Experience

Company Metric Context
United Airlines 5–15 min → 28 sec delay messaging; 100K+ messages sent Amazon Bedrock + Lambda + ElastiCache
United Airlines 30 min research → seconds; 3K → 15K employees enabled Mainframe access via natural language
Capital One 10,000+ agents using GenAI workflows in production Explicit observability monitoring; gradual risk slope-up
Cisco (via MongoDB) >$10M/year support cost savings AI virtual assistant, 3 years in production
Amazon Connect 6B → 12B minutes processed; 2x YoY Contact center AI platform
Priceline QA coverage: 3% sample → path to 100% Real-time coaching in 2–5 seconds post-call
Spectrum/Charter 35% cost reduction on 600TB data warehouse 32,000 jobs/day, 25,000 reporting users

Financial Services

Company Metric Context
MUFG 10x deal generation; 30% conversion maintained 2,000 employees, 1M corporate clients
MUFG Sales prep: hours/days → 1–2 minutes Corporate pitch materials
Verafin (NASDAQ) 70% time savings in fraud investigation Compliance document summarization
Genpack + NASDAQ 33–60% time savings in compliance automation Regulatory filing workflows
Crypto.com 90%+ accuracy; 6-week POC-to-production 100M+ users, 90 countries, 25 languages

Healthcare and Life Sciences

Company Metric Context
Cleveland Clinic 90% adoption of AI-drafted discharge summaries Pieces Technology platform
Pieces Technology 60–90 min saved per physician per shift Discharge summary drafting
Genentech 43,000 hours/year automated Target identification: weeks → minutes
Merck 90% accuracy on SDTM clinical data standardization Bedrock/Anthropic models for clinical trial data
Epic on AWS 100% chart availability, 25% performance gain, 30% cost reduction vs. on-premises EMR

AI Infrastructure Cost

Company Metric Context
Ricoh 50% lower training cost, 25% faster Trainium vs GPU cluster (Llama 3 Japanese)
ByteDance 20%+ throughput, 13% lower cost vs NVIDIA Inferentia 2 for multi-modal inference
Arcee AI 32% cost improvement at 8B parameter scale Inference cost vs P4d/P5 GPU instances

The Strategy Gap

The most striking data point came not from a customer but from a vendor survey. Infosys polled 1,500 CXOs and found:

  • Less than 50% of executives are satisfied with AI results
  • Less than 20% of AI PoCs reach production
  • Only 12% have an AI strategy in place
  • Only 11% have formal AI governance

These numbers align with the corpus’s existing triangulation: BCG finds 5% of organizations capturing substantial financial gains from AI (n=10,635, 2025). McKinsey finds 6% qualifying as high performers on AI (n=not disclosed, Nov 2025). The Infosys data adds a third anchor from the systems integrator perspective, confirming the strategy and governance deficit is structural, not anecdotal.

Key Data Points

Metric Source Date Credibility
<20% of AI PoCs reach production Infosys survey, n=1,500 CXOs Dec 2024 MEDIUM — vendor survey, methodology not published
10,000+ agents on GenAI workflows Capital One (Prem Natarajan, on-stage) Dec 2024 HIGH — named exec, production deployment
5–15 min → 28 sec messaging United Airlines (Krishna Srinivasan, on-stage) Dec 2024 HIGH — named exec, production metric
10x deal generation at 30% conversion MUFG (Tetsuo Horigane, on-stage) Dec 2024 HIGH — named exec, production metric
70% time savings in fraud investigation Verafin/NASDAQ (via AWS) Dec 2024 MEDIUM — cited by AWS, not direct exec
90% adoption of AI discharge summaries Cleveland Clinic (Dr. Amy Merlino, on-stage) Dec 2024 HIGH — named CMIO, production metric
43,000 hours/year automated Genentech (via AWS) Dec 2024 MEDIUM — cited by AWS, not direct exec
>$10M/year support savings Cisco (via MongoDB partner) Dec 2024 MEDIUM — secondhand, no direct Cisco exec
90% of GenAI pilots are RAG use cases IBM (Armand Ruiz, on-stage) Dec 2024 HIGH — named VP, 1,000+ pilots

Source Credibility

Overall: MEDIUM. AWS re:Invent is a vendor conference. Every customer on stage was selected by AWS to tell a success story on AWS infrastructure. No failures were presented. No control groups exist. No independent verification of claimed metrics.

The value of these case studies is directional, not definitive. They show what is possible when deployment goes well, on AWS infrastructure, with AWS support. They do not represent median outcomes.

These case studies are vendor-published and represent selected wins with no control group and no independent verification. Cross-reference against: METR RCT (experienced developers 19% slower), CMU study (40.7% code complexity increase), Atlan 200-deployment analysis (median +159.8% ROI requires workflow redesign first).

That said, named executives from Capital One, United Airlines, MUFG, Cleveland Clinic, and Crypto.com stated specific metrics on-record at a public conference. These carry more weight than anonymous survey data or vendor marketing claims.

What This Means for Your Organization

Three patterns from these sessions apply directly to mid-market AI planning:

Data readiness remains the gating factor. Confluent’s Andrew Sellers stated that “90% of any new application is the data modeling.” Capital One’s Prem Natarajan framed it as “your data advantage is your AI advantage.” IBM’s Armand Ruiz reported that 90% of their 1,000+ GenAI pilots were RAG use cases — retrieval over existing enterprise data, not novel model training. The message is consistent: investing in data quality and access before model selection produces better returns than chasing the newest model.

Workflow redesign, not tool deployment, creates the value. United Airlines did not give existing workers an AI chatbot. They rebuilt the delay messaging workflow from scratch — and compressed it from minutes to seconds. MUFG did not add AI to existing sales prep — they rebuilt the pitch generation process entirely. The 10x and 100x gains in these case studies came from rethinking the process, not from adding AI to the old one.

HITL governance enables adoption in regulated industries. Cleveland Clinic reached 90% adoption of AI-drafted discharge summaries — in a clinical setting where errors have life-safety consequences. The mechanism: Pieces Technology runs a “double human in the loop” where adversarial AI checks AI output before clinicians review. Capital One explicitly mentioned “a lot of observability” and gradual risk slope-up across 10,000 agents. In financial services, every deployment discussed assumed compliance governance as foundational infrastructure.

If these patterns raise questions about where your organization’s AI investment is actually creating value — or where it might be stalling — that conversation is worth having. brandon@brandonsneider.com

Sources

Brandon Sneider | brandon@brandonsneider.com April 2026