AI in the Enterprise

OpenAI’s "AI in the Enterprise" guides companies on integrating artificial intelligence (AI) to enhance processes, automate repetitive tasks, and create personalized products. Through seven lessons, backed by real-world cases, it proposes an experimental, iterative approach.

The first lesson emphasizes evaluations (evals), rigorous tests to measure AI model performance. Morgan Stanley evaluated applications for financial advisors, achieving 98% daily AI use, reducing search times, and focusing on clients.

The second lesson suggests embedding AI in products. Indeed used GPT-4o mini for job recommendations, boosting applications by 20% by explaining their relevance, later optimizing with a model using 60% fewer tokens.

The third lesson encourages early investment. Klarna implemented an AI assistant handling two-thirds of support chats, saving $40 million while maintaining high satisfaction, thanks to continuous iterations and widespread adoption.

The fourth lesson focuses on customizing models. Lowe’s fine-tuned OpenAI models, improving product search by 20% and error detection by 60%, adapting to inconsistent data and search patterns.

The fifth lesson promotes empowering experts. BBVA enabled 125,000 employees to create 2,900 custom GPTs, streamlining banking, legal, and support processes, cutting times from weeks to hours.

The sixth lesson centers on unblocking developers. Mercado Libre built Verdi with GPT-4o, accelerating app development, enhancing inventory, and detecting fraud with 99% accuracy.

The seventh lesson advocates bold automation goals. OpenAI automated support responses, handling thousands of tasks monthly, freeing teams for key tasks.

It concludes by advocating for evaluations, security guardrails, and tools like Operator, which automates workflows without APIs, ensuring privacy and compliance.