Beyond the Hype: How Real Companies Deploy AI for Scale and ROI
In Brief
Seven global enterprises — from Morgan Stanley to Mercado Libre — show that strategic, iterative, and expert-driven AI integration doesn’t just improve efficiency, it transforms how companies operate, build products, and deliver value at scale.
Despite the saturation of AI headlines, real-world enterprise adoption isn’t as simple as “plug in a model and go.” The reality is far more complex — and more strategic. OpenAI’s cross-industry deployments reveal that successful AI integration depends less on specific tools and more on transforming workflows, rethinking roles, and accelerating how organizations learn and evolve.
In a report grounded in seven enterprise case studies, OpenAI outlines what actually drives measurable outcomes in production settings — from document retrieval at Morgan Stanley to fraud detection at Mercado Libre. The lessons form a coherent architecture for AI adoption: start with rigorous evaluation, invest early, embed AI into products, fine-tune where it matters, empower experts directly, unblock developers, and set aggressive goals for automation.
Each lesson is reinforced with quantifiable outcomes — shortened support cycles, better job matching, improved product tagging, and significant profit improvements. Together, they show that enterprise AI isn’t a single system, but an evolving stack of capabilities, continuously iterated in context by people who understand both the technology and the business.
Rethinking AI Entry Points: Why Evaluation Comes First
Most enterprises are tempted to start AI adoption with pilots or small internal tools. Morgan Stanley, a global investment bank and wealth management firm headquartered in New York, took a different route: it began with evals — rigorous testing frameworks that assess how well models perform on real business tasks. Before anything went into production, AI outputs were benchmarked against human advisors across translation, summarization, and relevance.
This wasn’t a checkbox process. Evaluations gave Morgan Stanley the confidence to scale usage internally. Within months:
- 98% of financial advisors adopted OpenAI-powered tools in daily workflows;
- Document access rose from 20% to 80%;
- Time-to-response for clients dropped from days to hours.
Evals also created internal trust — a critical currency in regulated industries — by making performance, safety, and compliance measurable at every step. Evaluation wasn’t about proving a point. It functioned as a structured method for reducing risk and validating outcomes.
Embedding AI into Products
For AI to unlock business value at scale, it must leave internal back offices and become visible to end users. This is what Indeed achieved by embedding GPT-4o into its job recommendation engine. The real breakthrough came from the system’s ability to explain each match — not just make one.
Using GPT-powered systems, Indeed introduced “why” statements in job alerts. These contextual explanations — why this job, for this user — led to:
- 20% more job applications started;
- 13% increase in hires per application flow.
At Indeed’s scale — 350 million monthly visitors and over 20 million outbound messages — even modest gains compound. But there’s more: to maintain efficiency, the team fine-tuned a smaller GPT variant that used 60% fewer tokens without sacrificing accuracy.
Embedding AI went beyond personalization: it enabled product experiences to become more context-aware, relevant, and human-centric, with model performance treated as a strategic lever.
Early Investment, Compounding Returns
Klarna’s AI journey exemplifies the benefits of starting early. The fintech company introduced a generative assistant for customer service that now handles two-thirds of all support chats.
The results:
- Average resolution times dropped from 11 minutes to 2;
- Projected profit improvement: $40 million;
- Customer satisfaction remained consistent with human agents.
Just as important, 90% of Klarna’s employees now use AI in some form. Widespread adoption followed naturally as early integration led to feedback loops and incremental wins spread across departments.
The lesson is structural: AI investment is front-loaded. Delaying integration slows both impact and the development of organizational learning — a form of capital more difficult to replicate than code.
Sam Altman, CEO of OpenAI, emphasized the potential of AI to enhance human productivity on X:
Fine-Tuning for Relevance and Precision
Most general-purpose AI models don’t natively understand the nuances of a company’s data, taxonomy, or workflows. Lowe’s, a Fortune 50 retailer, tackled this by fine-tuning OpenAI’s models on their e-commerce product data — which was often inconsistent across suppliers.
The outcome:
- 20% boost in product tagging accuracy;
- 60% improvement in error detection.
The impact went beyond technical gains: search relevance improved, customer friction decreased, and internal QA workload dropped significantly. Fine-tuning gave Lowe’s more control over tone, structure, and domain specificity, making every model response aligned with the brand’s logic.
OpenAI likens fine-tuning to tailoring a suit: off-the-shelf models can work, but precision is in the fit.
Empowering Internal Experts as AI Designers
BBVA’s approach redefined AI adoption as a bottom-up, expert-led process. By giving 125,000 employees global access to ChatGPT Enterprise — with governance from Legal and Security — the bank enabled domain experts to build their own tools.
In five months, employees created over 2,900 custom GPTs. Examples include:
- Legal teams answering 40,000 policy questions annually;
- Credit Risk analysts accelerating creditworthiness evaluations;
- Marketing and operations streamlining internal workflows.
This distributed model eliminated bottlenecks in prototyping and unlocked AI’s potential within the constraints of real business logic — where experts know what matters and where models can fail.
The result: higher adoption, faster iteration, and a culture where AI functions as a direct extension of internal expertise.
Unlocking Developer Productivity with AI Platforms
Mercado Libre faced a common challenge: AI initiatives stalled as engineering teams hit capacity limits. To overcome this, the company built Verdi — an internal development platform powered by GPT-4o and GPT-4o mini.
By integrating LLMs with APIs, Python nodes, and guardrails, Verdi enabled 17,000 developers to build high-quality AI apps using natural language prompts — without writing boilerplate code.
This dramatically accelerated outcomes:
- Fraud detection accuracy rose to ~99%;
- Inventory capacity scaled via automated tagging with Vision models;
- Product descriptions adapted to regional dialects;
- Notification systems personalized at scale.
Verdi positioned AI as a core development layer, integrated directly into the organization’s operating model.
Automating Rote Work at Scale
Internally, OpenAI deployed its own automation layer on top of Gmail and support workflows. This system synthesized customer data, retrieved relevant knowledge, and generated contextual replies — turning multi-step manual tasks into automated flows.
The impact: hundreds of thousands of tasks processed monthly, freeing support teams to focus on high-context, high-impact interactions.
The system operated beyond standard dashboards or chatbots. It enabled process automation directly within existing workflows, using agentic capabilities such as browsing, data entry, and coordination across multiple tools — now applied to QA testing, system updates, and cross-platform operations.
The core principle: treat automation as infrastructure, not add-on tooling.
No More AI Pilots. Only Systems That Learn
Enterprise AI in 2025 is defined by resilience, adaptability, and scalable infrastructure. Companies focus on systems that evolve with use, supported by modular design, continuous testing, and clear operational governance.
Sam Altman, CEO of OpenAI, shared an example of their AI’s evolving capabilities on X:
Those who apply structured evaluation, integrate AI into core workflows, and decentralize development capabilities gain measurable leverage across their organizations. This approach reshapes how value is created — through speed, precision, and compounding intelligence in operations.
Disclaimer
In line with the Trust Project guidelines, please note that the information provided on this page is not intended to be and should not be interpreted as legal, tax, investment, financial, or any other form of advice. It is important to only invest what you can afford to lose and to seek independent financial advice if you have any doubts. For further information, we suggest referring to the terms and conditions as well as the help and support pages provided by the issuer or advertiser. MetaversePost is committed to accurate, unbiased reporting, but market conditions are subject to change without notice.
About The Author
Victoria is a writer on a variety of technology topics including Web3.0, AI and cryptocurrencies. Her extensive experience allows her to write insightful articles for the wider audience.
More articles
Victoria is a writer on a variety of technology topics including Web3.0, AI and cryptocurrencies. Her extensive experience allows her to write insightful articles for the wider audience.
