Morgan Stanley is a well-known company in the financial industry, and their attitude towards AI is interesting. At the beginning, they were unsure whether AI could help, after all, the work of financial advisors involves customer relationships and financial information, which is both complex and sensitive.

But before they implemented AI, they first developed a rigorous evaluation process, similar to giving AI a "physical examination". Just like when we choose an important product in life, we have to compare and try it repeatedly. Morgan Stanley first conducted three experiments with AI:

• • Language translation  : Testing the accuracy and fluency of AI translation is critical because there are many documents in the financial industry and a wrong word in translation may lead to a huge difference in meaning.
• • Summary  : See how AI compresses long articles into the essence. This involves indicators such as accuracy, relevance and coherence. After all, no one wants to read a bunch of useless information.
• • Compare with human experts  : Compare the AI’s results with the answers of financial advisory experts, just like a teacher marking students’ homework, to see whether the AI’s answers are correct and whether they have reference value.

At the beginning, they just wanted to see if AI could enable financial advisors to get information faster, spend less time on repetitive tasks, and spend more time providing clients with in-depth insights. These evaluations ensured that AI was up to the task. As a result, 98% of Morgan Stanley's advisors now use OpenAI every day. Previously, they could only access 20% of the files, but now it has jumped to 80%, and the time to find files has been greatly reduced. Advisors no longer have to spend a lot of time on trivial matters and can concentrate on dealing with clients. And customer feedback is particularly good. Things that used to take several days to respond to are now done in a few hours.

This is like when we want to introduce a new tool, we must first check it carefully and test whether it works. Morgan Stanley's move is very wise and scientific. The OpenAI report also pointed out that unlike the deployment of traditional IT systems, the implementation of enterprise AI requires continuous iteration, deep customization, and close integration with existing business systems. Morgan Stanley's structured evaluation method has laid a solid foundation for the scale of its AI use, enabling it to safely expand AI applications in a highly sensitive and complex industry.

2. Integrating AI into products: Indeed’s personalized recommendations

When it comes to finding a job, you won’t be unfamiliar with Indeed. This is the world’s largest job search website, and now it has also landed AI. Indeed integrated GPT-4o mini into its job recommendation engine, allowing AI to generate contextual interpretations for why a job is matched with a candidate. This increased transparency has led to a 20% increase in job applications and a 13% increase in employer engagement. Later, a custom fine-tuned model reduced token usage by 60%, demonstrating that thoughtful integration and optimization can effectively expand influence.

It's like we open a store. In the past, we relied on manual product recommendations. Now, with the use of intelligent systems, not only can the recommendations be more accurate, but we can also tell customers why this product is suitable for them, and the value is more specific. If customers are satisfied, the business will naturally be good. Moreover, OpenAI's report points out that embedding AI into the core product experience is the key to achieving differentiation and customer value. Indeed's case perfectly illustrates this concept. By closely integrating AI with business goals, it has achieved significant business growth and improved customer satisfaction.

3. Start now and invest early — Klarna’s first-mover advantage

Klarna, a global payment network and shopping platform, was an early adopter of AI. Their early investment in AI resulted in measurable improvements. Their AI assistant now handles two-thirds of support interactions, reducing resolution time from 11 minutes to 2 minutes. With 90% of employees using AI regularly, the organization has accelerated internal innovation and achieved an expected $40 million improvement in profits.

What’s even more amazing is that 90% of Klarna’s employees cannot do without AI in their daily work. From ordinary employees to management, everyone is familiar with AI and can use it easily. It’s like in a company, everyone is proficient in new tools, and their work efficiency naturally takes off. And they don’t just throw AI into the process right away, but continue to test, optimize, and polish this assistant.

This early investment not only brings direct operational efficiency gains, but also accumulates valuable data and experience for the company, enabling it to gain an advantage over the competition. As OpenAI explains, capturing compounding benefits, Klarna’s case highlights the criticality of early investment to long-term success, with its benefits growing over time.

4. Customize and fine-tune the model — Lowe’s precision search

Lowe's is a home improvement company. They used to have pain points in e-commerce search. Product data was incomplete and inconsistent, customers had a hard time finding products, and the matchmaking efficiency was low. But their way of improvement was to try to apply AI to work with OpenAI and use fine-tuning models, just like tailoring AI. They fed the model with their own product data, shopper search habits and other unique information to make it specifically adapted to their own business. As a result, the accuracy of product labels increased by 20% and error detection increased by 60%. This is like customers used to find a needle in a haystack among a vast number of products. Now AI helps them locate accurately and find the products they want in an instant.

This shows that AI models cannot be "one size fits all". They must be customized according to the company's own situation, like carving a work of art, defining private problem domains, and customizing model applications for private vertical domains, so that they can exert their greatest power. OpenAI emphasizes that fine-tuning for specific use cases is crucial. Fine-tuning enables the model to reflect internal language, format, and industry nuances. For Lowe's, this means injecting precise product information and customer search intent into each product description to achieve an excellent shopping experience.

5. Let experts master AI - BBVA's employee empowerment

BBVA, a big player in the banking industry, has a vision for how to play with AI. Instead of centralizing AI development, they empowered employees to build custom GPT applications. In just five months, employees created more than 2,900 custom GPTs, covering areas from legal to compliance, customer service, and credit risk. This approach reduces the time to value and ensures that AI is applied where it is most needed.

BBVA believes that investing in ChatGPT is investing in employees. AI unleashes the potential of employees, making them more efficient and creative at work. It's like equipping employees with an intelligent assistant, which makes them more confident in their work. The expert-led AI development approach leverages the collective knowledge and creativity of employees to ensure that technical solutions are precisely aligned with business needs. This decentralized approach speeds up adoption and improves AI proficiency across the organization.

In fact, this sounds a bit like the initial practices of some domestic companies, which are to build a set of Dify, set up an application market, and run the prompt words first. Different professional positions will have a more specialized understanding of the business, and AI will be used in their own professional positions first. "Learn by doing, and do by learning."

6. Liberating developers - Mercado Libre's efficient platform

Mercado Libre is the largest e-commerce and financial technology company in Latin America. They have a problem: developer resources are tight and project progress is always delayed. So they teamed up with OpenAI to develop a platform called Verdi, which uses GPT-4o to help developers enhance their capabilities. This platform integrates language models, Python nodes, and APIs, just like creating a super toolbox for developers.

In the past, developers had to delve into the source code to develop applications. Now, with this platform, they can quickly build high-quality applications with simple operations using natural language. Moreover, security and rules are built-in, so developers don’t have to worry about them. With this platform, Mercado Libre’s developers are like wings, and their efficiency has been greatly improved. For example, they can optimize inventory faster, accurately detect fraud, customize product descriptions, and even personalize push notifications to allow users to interact with products more actively. This is like providing developers with a powerful integrated development environment (IDE), which not only improves the speed and quality of code writing, but also reduces errors through smart prompts and auto-completion functions.

The OpenAI report states that supporting developers is a key pillar of enterprise AI success. By providing powerful tools and platforms, organizations can unleash the creativity of developers, accelerate innovation cycles, and build scalable and secure AI solutions. As Mercado Libre's practice proves, the right tools can free developers from repetitive, low-value tasks, allowing them to focus on building strategic, innovative solutions that drive the business forward.

7. Setting bold automation goals: OpenAI’s self-innovation

OpenAI itself is the most thorough in applying AI. Their internal support team used to handle customer issues by spending time logging into the system, understanding the background, writing replies, and performing operations, which was a cumbersome process. So they developed an automation platform themselves, which was like installing an "intelligent engine" for their workflow. This platform runs on existing workflows such as Gmail, and can instantly retrieve customer data and related knowledge articles, and then integrate this information into reply emails or directly trigger specific actions, such as updating accounts and opening support tickets.

Now, this platform handles hundreds of thousands of tasks every month, freeing the team from repetitive work, so that everyone can devote their energy to more valuable work. OpenAI set a very high automation goal for itself, and then achieved self-innovation through AI. It's like we set a high goal for ourselves, push ourselves, and find that with the help of AI, we can really accomplish things that we thought were impossible before.

As OpenAI explains, setting automation goals can significantly reshape workflows and improve efficiency. By implementing automation solutions in its daily operations, OpenAI demonstrates how AI can be a catalyst for organizational transformation and growth.

Summary: AI helps enterprises reach new heights

After reading these seven cases, I feel that there is no grand narrative, just practical "down-to-earth" cases. We can clearly see that the application of AI in enterprises can effectively improve efficiency, optimize processes, and enhance competitiveness. The key to the success of these companies lies in: not rushing for success, but starting with evaluation, accurately locating the entry point where AI can really play a role; based on their own business characteristics, clarifying the problem domain, customizing and fine-tuning the AI ​​model to closely fit the needs of the enterprise; handing AI to employees, allowing those who understand the business best to fully tap the potential of AI; untying developers, creating an efficient development platform, and accelerating the implementation of AI applications; boldly setting automation goals and reshaping workflows with the help of AI.

OpenAI’s AI in the Enterprise report also makes a strong case for the need for structured, iterative AI integration based on real-world use cases. The report recommends starting small, investing early, fine-tuning for relevance, and expanding from high-impact use cases rather than rushing to full adoption. A common theme runs through the seven cases above: effective AI applications are built on rigorous experimentation, powerful tools, and empowerment to those closest to the problem. For technical and business leaders, OpenAI’s guide provides clear and actionable practical guidance for long-term AI success in the enterprise.