Lessons from seven cutting-edge companies
A new way to work
As an AI research and implementation company, OpenAI makes partnering with global companies a priority because our models perform best in complex, interconnected workflows and systems.
We see that AI is bringing meaningful, measurable improvements in three ways:
01 Increasing employee productivity
Helps people achieve better results in less time.
02 Automating routine operations
Frees employees from repetitive tasks so they can focus on creating value.
03 Product reinforcement
It provides a more relevant and responsive user experience.
But using AI is not the same as developing software or deploying cloud applications.
The most successful companies are those that see AI as a new working approach. This leads to experimental thinking and an iterative approach that delivers value faster and ensures greater user and stakeholder engagement.
Our approach: iterative development
OpenAI is organized around three teams:
- Research Team — develops the fundamental foundations of AI, developing new models and capabilities.
- Applied Team — turns these models into products like ChatGPT Enterprise and our API.
- Deployment Team — introduces these products in companies, solving the most pressing problems.
We use iterative deploymentto quickly learn from real-world cases and accelerate product improvement. This means frequent updates, collecting feedback, and improving performance and security at every stage.
Result: Users get access to the latest AI advances early and often — and their feedback influences future products and models.
Summary for executives
Seven lessons on how to implement AI in an enterprise environment:
01 Start with grades
Use a systematic approach to assessment to measure how models are performing for your tasks.
02 Embed AI into products
Create new customer experiences and more relevant interactions.
03 Get started now and invest in advance
The sooner you start, the more complex the effect you will get over time.
04 Customize and refine models
Tailoring models to your specific cases can significantly increase value.
05 Give experts access to AI
Those closest to the process have the best idea how to improve it with AI.
06 Unlock developers
Automating the development lifecycle multiplies the benefits of AI.
07 Set bold automation goals
Most processes involve routine work that is suitable for automation. Set ambitious goals.
Next, we'll take a closer look at each lesson using examples from customers.
Lesson 1
Start with grades
How Morgan Stanley ensured quality and safety through iterations
As a global leader in financial services, Morgan Stanley is a relationship-oriented company. It's no surprise that questions have been raised across the company about how AI can add value to such a personal and sensitive activity.
The answer was to conduct intensive assessments for every proposed AI application.
Assessment (eval) is a strictly structured process of measuring how an AI model performs a specific task using specified metrics. It's also a way to continually improve AI solutions with experts at every stage.
How it all started
Morgan Stanley's first assessment was aimed at improving the performance of financial advisors.
The idea was simple: if consultants could get information faster and spend less time on routine tasks, they would be able to give clients more and better advice.
They conducted three model evaluations:
01 Text translation
Assessment of the accuracy and quality of translations created by the model.
02 Summarization
Analysis of how a model reduces information, using agreed metrics of accuracy, relevance, and connectivity.
03 Comparison with experts
Comparing AI results with answers from professional consultants, assessing accuracy and relevance.
These and other assessments gave Morgan Stanley confidence that it was possible to incorporate AI examples into productive environments.
How are things now
Today, 98% of Morgan Stanley consultants use OpenAI on a daily basis; access to documents has increased from 20% to 80%, and the time spent searching for information has been significantly reduced. Consultants spend more time communicating with clients, thanks to automated tasks and faster insights.
Consultants' reviews are extremely positive.
They've become more engaged, and activities that used to take days now take place in hours.
Caitlin Elliott
Company-wide generative AI manager
📺 Morgan Stanley case video (YouTube)
What is “evaluation” (eval)?
Valuation is the process of checking and testing the results produced by your model.
Strict assessments lead to more stable and reliable applications that are resilient to change.
They are based on tasks that measure the quality of the model's output compared to the standard:
- Is it more accurate?
- Does it meet the requirements?
- Is it safe?
Key metrics depend on what's important in your particular case.
Lesson 2
Embed AI into your products
How Indeed makes job selection more humane
When AI automates and speeds up boring, repetitive work, employees can focus on things only humans can do.
Thanks to its ability to process massive amounts of data, AI can create a customer experience that is more felt personalized and humane.
Indeed, the world's #1 job search site, uses GPT-4o mini for new ways to compare job seekers and jobs.
The power of explaining “why”
Simply offering a suitable job is not enough — it's important to explain why this one the job is recommended.
Indeed uses GPT-4o mini's data analysis and text generation capabilities to formulate such explanations in emails and messages.
The popular “Invite to Call” feature now provides arguments about why a candidate is a good fit for the position based on past experience and skills.
The result after the introduction of AI into recruitment:
- Increasing the number of job applications started at 20%
- Increasing ultimate success (hiring) by 13%
Taking into account more 20 million messages per month et 350 million visitors on the site, that means significant business impact.
To increase efficiency, OpenAI and Indeed have collaborated on a smaller model that provides similar results, but with 60% fewer tokens.
The selection of suitable vacancies is a deeply human result.
The Indeed team is using AI to connect people to work quicker—a win for everyone.
Chris Hyams
Chief Executive Officer
Lesson 3
Get started now and invest in advance
How Klarna benefits from its AI expertise
AI is rarely a plug-and-play solution — real-life cases are becoming more complex and valuable in the process of iteration.
The sooner you start, the faster and more you'll benefit from cumulative improvements.
Klarna, a global payment network and trading platform, has introduced a new AI assistant to optimize customer service.
A few months later, the assistant was already processing two-thirds of all chats, doing the work of hundreds of operators and reducing the average response time with 11 minutes to 2.
This project is expected to bring $40 million in profit, while satisfaction rates remained at the level of human support.
And it's not all in one day. Klarna has achieved these results thanks to continuous testing and improvement assistant.
It is equally important that 90% of Klarna employees use AI every day.
Mass exposure to AI has made it possible to launch internal initiatives faster and continuously improve customer experience.
By investing early on and encouraging widespread adoption, Klarna is watching the effect of accelerating returns from AI across the business.
This AI breakthrough in customer interaction means
better service at a better price,
more interesting tasks for employees
and higher returns for investors.
Sebastian Semiatkowski
Co-Founder and CEO
Lesson 4
Customize and refine your models
How Lowe's is improving product searches
The most successful companies in using AI are those that invest in internal adaptation and training of models to their data.
OpenAI has invested heavily in the API to make customization easier, both in the form of self-use and with support from OpenAI.
We worked closely with Lowe's, a Fortune 50 household goods company to improve search in the online store.
Due to thousands of suppliers, Lowe's is often seen incomplete and inconsistent product data.
Improvement was key product descriptions and tags, as well as understanding customer search behavior, which differs by product category.
It is especially important to further train models here.
Fine-tuning results based on Lowe's data:
- Improving the accuracy of tagging products on 20%
- Increasing the efficiency of error detection on 60%
The team was thrilled to see the results
additional GPT-3.5 training on our product data.
We realized that this was our win!
Nishant Gupta
Senior Director for Data, Analytics and Computational Intelligence
📝 Product note:
OpenAI has launched Vision Fine-Tuningto further improve product search and solve problems in medical imaging and autopilot.
🔗 Learn more
What is fine-tuning?
If the GPT model is a suit from the store,
then further training is tailoring to order: adapting the model to your data and needs.
Why this is important:
- Higher accuracy — a model trained on your data (e.g. catalogs or FAQs) provides more relevant and brand-appropriate answers
- Industry expertise — the model has a better understanding of professional terms, style and context
- The same tone and style — whether it's legal links or brand descriptions — everything is designed the same
- Faster results — less manual editing, employees can focus on what matters
Lesson 5
Give experts access to AI
BBVA takes an expert approach to implementing AI
Employees know the company's internal processes and problems best — and they are often the ones who are able to find optimal AI solutions.
Putting AI in the hands of these experts can be much more effective than creating universal solutions from above.
BBVA, a global banking leader, has more than 125,000 employees, each of which faces unique challenges and opportunities.
The company decided to give employees around the world access to AI, working closely with legal, compliance, and IT security teams to ensure responsible use.
They deployed ChatGPT Enterprise across the company, after which they gave people the opportunity find ways to use it yourself.
Usually, even creating a prototype requires technical resources and time in our business.
With custom GPT, this has become easy — anyone can make an application for their own purpose.
Elena Alfaro
Global AI Implementation Manager at BBVA
5-month results:
- Staff have created more than 2900 custom GPTs
- Many of them reduced the duration of projects and processes from weeks to hours
Application examples:
- Credit Risk Team — determines creditworthiness faster and more accurately
- Legal department — processes 40,000+ requests per year on policy, compliance and other issues
- Customer service — automates mood analysis in NPS surveys
AI tools are also widely used in marketing, risk management, operations, and other departments, all because employees themselves found how to apply AI in their work.
We're considering investing in ChatGPT
as an investment in our people.
AI enhances our potential
helps to be more efficient and creative.
Elena Alfaro
Global AI Implementation Manager at BBVA
📝 Product note:
ChatGPT is capable of doing deep research.
You ask a query, and it synthesizes hundreds of sources, creating detailed, expert-level reviews in minutes.
Internal assessments have shown that such studies save on average 4 hours for every challenge.
📺 Video: BBVA brings AI to every team (YouTube)
Lesson 6
Unlock developers
How Mercado Libre makes AI programs faster and more stable
In many companies developers are the main bottleneck and growth constraint.
When engineering teams are overloaded, innovation is hampered and ideas pile up in the backlog.
Mercado Libre, Latin America's largest e-commerce and fintech company, partnered with OpenAI to create a GPT-4o-based development platform.
So he was born Verdi — a platform layer that helps 17,000 Mercado Libre developers speed up and unify the creation of AI applications.
Verdi combines language models, Python nodes, and APIs into a single scalable system where natural language is the main interface.
Developers can now create high-quality apps fasterwithout diving into the source code.
Security, routing logic, and security frameworks are already built in.
What we have achieved:
- Increased cataloging capacity: GPT-4o mini Vision helps you tag and describe products, allowing you to fill out the catalog 100 times faster
- Fraud detection: AI processes millions of product cards, achieving accuracy up to 99% for suspicious cases
- Localization of product descriptions: translation and adaptation to the regional language characteristics of Spanish and Portuguese
- Increasing orders: automatic summarization of reviews helps customers understand the point faster
- Personalized notifications: adapting push messages increases engagement and the quality of recommendations
We built our ideal AI platform based on GPT-4o mini
with a focus on reducing cognitive load
and an opportunity for the entire company to develop
and innovate.
Sebastian Barrios
Senior Vice President of Technology
Lesson 7
Set bold automation goals
How we automate our own work at OpenAI
At OpenAI, we work with AI every day and are constantly looking for new ways automate routine processes.
Example: Customer Service
Our support teams spent too much time accessing systems, analyzing the context, compiling responses, and taking necessary actions on behalf of the client.
That's why we we have created an internal automation platformthat works on top of our existing processes and systems.
It automates routine work and speeds up insights and actions.
The first case: automating the processing of emails in Gmail
The platform gets access to customer data and relevant articles, and then uses the results:
- to respond to an email
- to perform actions: updating an account, creating a ticket, etc.
Result:
- steel teams more efficient, faster and more customer-oriented
- the system is already performing hundreds of thousands of tasks per month, freeing people up for more value-added jobs
And all of this was possible because we set ambitious automation goals right from the start, rather than coming to terms with inefficient processes as “business costs”.
Conclusion
We learn from each other
As previous examples have shown, every company has opportunities to use AI to achieve better results.
Application scenarios may vary by industry and scale, but the principles remain universal.
The general idea:
The introduction of AI brings the greatest return when available open, experimental thinking, in combination with strict assessments and safety measures.
Successful companies aren't in a hurry to incorporate AI into every process — they start with simple but profitable cases, learn from them and then transfer their experience to new areas.
The results are clear and measurable:
- Faster and more accurate processes
- A more personalized customer experience
- Work smarter — employees do what people do best
We are now seeing how companies integrate AI into complex processes, often using tools and agents to achieve results.
We'll continue to share insights from the front line so you can use this knowledge in your strategy.
📝 Product note: Operator
Operator — an example of OpenAI's agent approach.
Its own virtual browser allows it to:
- navigate through sites
- click buttons
- fill out forms
- collect data like a human would
In addition, it is capable carry out processes in different systems and tools — without integrations and APIs.
Examples of use in companies:
- Automated software testing — Operator works like a real user, identifying bugs in the interface
- Updating accounting systems on behalf of users — without technical instructions and APIs
The result: full automation from start to finish, freeing teams from routine and increasing overall efficiency.
A reliable AI platform for business
Security and privacy
Security, privacy and control are the most important things for our corporate clients. We guarantee:
- Your data stays yours
We don't use your content to train models; your business retains full ownership. - Compliance with corporate standards
Data is encrypted during transmission and storage. Compliance with SOC 2 Type 2, CSA STAR Level 1 standards. - Flexible access control
It's up to you to determine who can see and manage data. This ensures compliance with management requirements. - Data storage settings
Flexibly configure logging and saving in accordance with your organization's policies.
🔗 More details: OpenAI Security | OpenAI Trust Portal
Additional resources
🧠 OpenAI is an AI research and implementation company.
Our mission is to make sure that artificial general intelligence benefits all of humanity.
