The use of AI is no longer experimental; it has become a strategic lever for companies worldwide. OpenAI gathered insights from seven leading organizations that have successfully integrated AI into their daily processes. This blog summarizes their approach and shows how your organization can learn from it. In this blog, you will learn when building an AI agent makes sense, how to approach it, and which design principles ensure reliable and safe deployment in your organization.
1. Start with evaluations
Example: Morgan Stanley
The bank did not start with technology, but with trust. By conducting systematic evaluations, such as comparing AI output with that of human experts, Morgan Stanley was able to deploy AI in a safe and controlled manner. Now, 98% of their advisors use AI on a daily basis, and the time required for customer follow-up has been drastically reduced.
Tip: Define what "good output" means for each use case (accuracy, relevance, security) and measure it systematically.
2. Integrate AI into your products
Example: Indeed
The job site used GPT-4 not only to make better recommendations to job seekers, but also to substantiate why a particular job suits someone. This led to a 20% increase in applications and a 13% increase in successful matches.
Tip: Let AI contribute to the customer experience, not just efficiency.
3. Start early, invest broadly
Example: Klarna
Klarna's AI assistant now handles two-thirds of all customer service calls, faster and cheaper, without compromising on quality. The key? Start early, roll out widely, and keep iterating. Now, 90% of employees use AI in their daily work.
Tip: Don't wait until you're "perfect" to start. Early experimentation yields exponential returns.
4. Create customized models
Example: Lowe’s
By fine-tuning a GPT model on its own product data, Lowe's increased tagging accuracy by 20% and error detection by 60%. This made the search process on their e-commerce platform much more relevant and consistent.
Tip: Fine-tuning offers better results than generic models, especially for specialized content or domains.
5. Put AI in the hands of experts
Example: BBVA
Instead of centralised control, BBVA gave 125,000 employees access to ChatGPT Enterprise. In five months, they built 2,900 custom GPTs. Teams used AI for risk assessment, legal support and sentiment analysis on customer feedback, among other things.
Tip: Your people know their work best. Give them access to AI tools and space to develop solutions themselves.
6. Unlock your developers
Example: Mercado Libre
The Latin American tech giant developed a platform ("Verdi") with OpenAI that enables 17,000 developers to build AI applications faster. The impact? Better fraud detection, faster product processing, and personalized customer notifications.
Tip: Take the pressure off development teams with platforms that accelerate and standardize AI development.
7. Dare to automate on a large scale
Example: OpenAI itself
Within OpenAI, AI tools are used to automate internal support processes, from emails to system updates. This automates thousands of tasks per month, freeing up time for work with greater impact.
Tip: Don't look for small optimizations, but dare to automate entire workflows.
AI success requires vision and experimentation
The companies in this guide prove that AI success is not accidental. It stems from:
- Start early and keep learning
- Practical testing with clear evaluation criteria
- Involving employees at all levels
- Investing in customization and infrastructure
- Strategic automation, not just optimization
Those who view AI as a new way of working rather than a one-off tool will reap the benefits in the long term.