2024-05-26-麦肯锡-走过新一代人工智能的蜜月期_首席信息官从试点到规模的七个硬道理（英）_10页_545kb

Moving past gen AI’s honeymoon phase: Seven hard truths for CIOs to get from pilot to scale

McKinsey Technology and QuantumBlack (AI by McKinsey) have collaborated on this article to explore the difficulties of scaling gen AI, drawing from their research and client work in a May 2024 report. While gen AI pilots are easy to build, only 11% have achieved large-scale adoption, highlighting significant barriers to implementation.

There are three primary approaches to using gen AI:

• Taker: Use off-the-shelf tools like GitHub Copilot or Einstein.
• Shaper: Integrate bespoke capabilities using prompts, datasets, and connections to internal systems.
• Maker: Build custom LLMs from scratch.
  Most companies combine Taker and Shaper approaches, with Shaper delivering the highest-value initiatives.

The maturation of gen AI requires CIOs to focus on seven hard truths for scaling:

1. Eliminate the noise: CIOs must prioritize feasible use cases with clear business impact, reduce unnecessary experiments, and focus on solving important business problems.

2. Focus on integration: It’s crucial to understand how components work together securely, not just individual parts, and to automate workflows (using API gateways and MLOps platforms) for reliable scaling.

3. Manage costs: Despite only accounting for 15% of project costs, expenses can spiral due to rising model usage and workforce training. High performers allocate resources wisely and balance "build vs run" costs.

4. Tame tool proliferation: Avoid creating complex environments by selecting the right cloud providers and standard tools, simplifying infrastructure for ease of deployment.

5. Build teams for value: Assign cross-functional teams to ensure solutions are business-value-focused. Weak governance led lower ROI in unoptimized trials.

6. Use the right data: Clean and relevant data is a prerequisite for high-performance gen AI models.

7. Promote reusability: Driving development speeds through reusable components while ensuring they align with business goals.

The article emphasizes that successful scaling requires CIOs to overhaul how their firms operate, including strategic organizational changes (e.g., governance structures) and practical steps to manage costs and complexity.