Chapter 13 Build or Buy

合集 - the business case for ai(10)

1.Chapter 6 Optimize decision making with AI - Simple versus Intelligent Data Analytics2024-12-252.Chapter 1: The Promise of AI2024-12-253.Chapter 7 ML development life cycle2024-12-254.Chapter 8, 9 B-CIDS: 5 pillars of AI preparation → Jumpstart Approach2024-12-255.Chapter 10-11-12. Find AI Opportunities - 4 Stages2024-12-256.Chapter 3-4. 5 tips for AI success & 5 myths of AI2024-12-257.Chapter 5. How AI can improve Business Process2024-12-25

8.Chapter 13 Build or Buy2024-12-26

9.Chapter 14. Measure Success2024-12-2610.Book notes: The Business Case for AI: A Leader's Guide to AI Strategies, Best Practices & Real-World Applications2024-12-26

收起

Many organizations today start by hiring a team of data scientists to implement AI. Yet, this is not the only way of bringing your AI vision to life, nor is it the most cost-effective, especially when you’re just getting started.

different strategies for integrating AI

Buy Strategy

AI applications that commonly come prepackaged include:

• Virtual AI assistants
• Facial recognition systems
• Sentiment analysis tools
• Language translation services
• Product recommendation engines
• Speech recognition systems

pros and cons of buy strategy

Pros of Buy Strategy:

• Convenience: Off-the-shelf solutions are easy to implement without needing a specialized team of data scientists.
• Cost-Effective: Reduces the need for extensive resources and expertise in machine learning algorithms.
• Low Maintenance: Typically, there is less need for ongoing maintenance and monitoring of models.
• Utilizes Existing Teams: Existing engineering and IT teams can evaluate, customize, and integrate these solutions into business systems.

Cons of Buy Strategy:

• Overgeneralization: Off-the-shelf solutions are designed to be broadly applicable, which can compromise quality and accuracy.
• Lack of Customization: The underlying assumptions may not align with specific organizational use cases and data needs.
• Inadequate for Specialized Needs: Solutions requiring high levels of accuracy in particular domains may be better served by custom-built options.

The Custom-Build Strategy

...starts with building out an MVP - geeting user feedback before developing final product. This feedback limits the risk of failure. MVPs are useful for iterative improvements. During testing, you can use your MVPs to evaluate your ROAIs, assess real-world model performance, get feedback from users, and then iterate as needed.
There are several ways to get your MVPs implemented:

“Hire” internal data science teams

.. pass.. 在内部找团队支持

Hire AI consultants

How to use consultants

• End-to-End Implementation: Consultants handle all aspects of the project, including data formatting, cleaning, model development, experimentation, testing, and delivering production-quality code. They may also assist with deployment if it involves a cloud-based service.
• Working Prototype Model: Consultants build a functional prototype, which your engineering team then develops into a minimum viable product (MVP) .
• Technical Advisory Model: Consultants act as coaches or architects, providing guidance while your engineering team collaborates with them to implement the AI solution.

... working prototype and technical advisory models are sufficient for our needs. They often independently test prototypes to ensure functionality before expanding to an MVP. This approach ensures that the data science components are developed correctly, allowing engineering teams to refine the solution for edge cases, integrate it into existing infrastructure, and enhance the source code to meet coding standards. Additionally, the MVP must undergo testing during the Post development testing (PDT) phase.

prototype is even earlier than MVP

What to look for in consultants

聚焦业务、有落地经验、项目管理能力强

• Problem-Focused, Not Techniques-Focused - 选择专注于解决问题并考虑符合要求的技术的提供商。一些较新的技术生产成本要高得多，并且可能会使您面临财务⻛险。
• Expertise and Past Projects - 为避免落入营销陷阱，请始终查看实施团队或个人的资格。他们有 AI 领域的经验和培训吗？他们完成了哪些项目？谁在领导团队？仅仅拥有研究生学位，尤其是在计算领域之外，并不意味着他们知道如何实施 AI 解决方案或处理数据。
• Cost-, Time-, and Scope-Sensitive - 项目管理的专业性。对于 AI 组件，咨询公司需要将它们进一步分解为多个阶段。这样，我们才能知道每个阶段会发生什么，我们可以分阶段快速测试解决方案，而不是等待整个计划完成。服务提供商是我的技术专家。他们应该知道如何将问题分解为具有时间线信息的合乎逻辑且范围明确的部分。

Hire new in-house data science personnel (the most expensive)

此外，行业经验法则是，对于您雇用的每1位数据科学家，要有2~3名数据工程师来支持这些数据科学家。换句话说，您需要雇用的不仅仅是数据科学家。
To be strategic, before bringing on data scientists, consider hiring a strong team of data engineers. They can be instrumental in many ways. They can start supporting your internal data needs, help implement parts of your AI and data strategy, and support AI initiatives whether you outsource, buy, or build internally.

Skills to look for in Data Scientists

A candidate’s expertise can be determined by their:

• Educational background or other formal training
• Past projects completed (in an industry setting)
• Ability to describe appropriate solutions for your company-specific problems
• Software engineering skills
• Ability to articulate and measure the success of AI initiatives
• Know-how to productionize models as a bonus

benefits of having an in-house data science team

Pros of Having an In-House Data Science Team:
考虑长期效应，但是也有风险，中国的离职率很高。

1. Deep Understanding of Company Infrastructure: In-house data scientists develop a strong understanding of existing data stores and company processes over time, enhancing their ability to support AI initiatives effectively throughout the development life cycle.
2. Implementation of AI Strategy: They can assist in executing the organization’s AI strategy, ensuring alignment with business goals and facilitating ongoing AI development.

Cons of Having an In-House Data Science Team:

1. Siloed Work Environment: Data scientists often work in isolation, which can hinder collaboration and limit their awareness of company inefficiencies, leading to suboptimal application of AI.
2. Misguided AI Applications: Without direct experience of organizational challenges, data scientists may guess where to apply AI based on data availability rather than focusing on high-impact problems.
3. Underutilization of Expertise: Companies may fail to realize the value of their data science teams unless they actively engage their expertise or integrate data scientists into specific departments.

Const considerations: In-house data science personnel versus Consultants

... if there are no demands, then I am dead ..

it makes the best sense to hire data scientists when you have a series of projects for them.
Otherwise, it can become a financial burden.
If you’re tackling one-off projects or just getting started with AI, it may be wiser to hire consultants until you’re able to plan out initiatives for the long term. Plus, don’t forget, you also need to hire data engineers to support your data strategy and your data scientists.

How the costs of implementing AI projects stack up after three years using a
consultant versus full-time data scientists. Scenario 1 assumes projects in the
first year and maintenance thereafter. Scenario 2 assumes new projects every
year.

购买与内部构建 AI 解决方案或使用顾问的比较