O'Reilly logo
live online training icon Live Online training

AI Project Selection and Management

AI for Business

Topic: Data
Roy Lowrance

The AI for Business series of online trainings helps managers determine what their organization needs to do about artificial intelligence and helps them launch their initial projects.

Artificial intelligence (AI) has moved from being a buzz word to providing capabilities that organizations need to consider, because those capabilities can augment products and services, reduce operational costs, and increase reliability. But these capabilities come with costs, including the costs of introducing new skills and knowledge into the management team, building and maintaining a cadre of AI experts, and increased computational and data costs. For AI to generate value, those costs must be managed effectively and balanced against their payoffs.

This course sequence is for you if you understand your business — what products and services it offers, how those are realized, what long-term advantages it is creating, and why customer and suppliers work with you instead of others. These courses will teach you what AI is capable of today by defining common use cases for AI. For each course, you will learn the skills and other capabilities that are needed for effective deployment. The course will also review upcoming AI technologies of which you should be aware.

AI Project Selection and Management is the second course in the series and covers additional material that those who manage AI projects need to know. Managers will probably want to take it also, so that they develop a better understanding of what the project managers will be doing. AI projects are typically iterative; therefore, we start with a useful framework that identifies and sequences the iteration stages and helps you achieve a common understanding of where you are in a project. AI project managers will need to work with data scientists directly, and it is important to have an understanding of what they are doing, so we’ll explain the steps that they follow. Much of your data is in databases which are routinely incorporated into models. However, increasingly you will want to use text and images to build predictions and we describe the major techniques for doing so. Finally, AI technologies are developing rapidly. We describe the most important technologies at present and those likely to emerge as important in the next few years.

What you'll learn-and how you can apply it

After completing the course, you will be able to:

  • Articulate the steps an AI projects goes through and anticipate what might go wrong at each step.
  • Help data scientists make technical decisions about model design so that the business value of the predictions is high.
  • Work with the data scientists on your team to incorporate text and image data into models.
  • Participate in discussions about technology choices for supporting AI-based projects.

This training course is for you because...

  • You are a manager, and you need to figure out what role AI should play in your strategy.
  • You are a manager or project manager, and you need to design and manage projects that use AI technologies.


  • A good understanding of your business or organization is required. That typically means at least several years of experience in business in general and a few years of experience in the specific area/field you are working in.
  • You need to know how to interpret formulas of several variable (for example, f(x,y) = 3x - 7y), but you don’t need to recall any calculus or statistics that you may have learned. Nor do you need to know how to program computers.

Course Set-up

The course is delivered through lectures and discussion. Your computer only needs access to the O’Reilly platform. You will receive handouts of the slides used, but you may want to take notes, perhaps on your computer, a tablet, or on paper.

Recommended Preparation

About your instructor

  • Roy began his career as a software engineer and has worked as a consultant at McKinsey and Company and the Boston Consulting Group (BCG) where he became the partner responsible for the information technology practice in the Americas. He was Chief Technology Officer at Capital One which was one of the first companies to leverage AI and at Reuters, which is a big data company.

    He also facilitated the data science initiative at New York University and became Managing Director of its Center for Data Science. He was the lead designer for the MS in Data Science and launched that degree. He is the co-founder of the AI-based startup 7 Chord which predicts corporate bond prices.

    He holds a BA in Mathematics cum laude from Vanderbilt University, an MBA with high distinction from the Harvard Graduate School of Business Administration, and a Doctorate in Computer Science from the Courant Institute of Mathematical Sciences at New York University.


The timeframes are only estimates and may vary according to how the class is progressing

Segment 1: AI Project are not like typical software development projects (50 minutes)

  • Because AI-based projects are exploratory, you need to use an iterative process.
  • CRISP-DM is an appropriate iterative process for which one can specify how a decision to iterate back to a previous step would be made.
  • CRISP-DM is compatible with agile project management processes like Kanban, which is used for software development when requirements emerge as the project progresses.

Break: 10 mins

Segment 2: working effectively with the model builders on your team (50 minutes)

  • The data scientists on your team will build the models.
  • They have specialized knowledge about statistics, mathematics, and software engineering but they typically lack a grounded understanding about how the business works.
  • You need to know enough about what they do so that you can work with them to set appropriate directions for the project.
  • In the first course, we taught how function-based models could be fit by searching for the parameters, but often, one must use gradients to do the search, and we’ll explain those intuitively in this segment.
  • Models are often combined into ensembles in order to increase their accuracy.
  • Two frequently used ensembles are random forests and gradient boosting.
  • When models are not accurate enough, you can use the guidelines we provide to decide whether to switch to different models or continue to use the same models but with additional training data.
  • The model you deploy into operations will be accurate on the training data, which means you need to evaluate the extent to which the training data conforms to your operational data and to your current business policies.

Break: 10 mins

Segment 3: beyond structural data (50 minutes)

  • So far, we have considered structured data, data that can be easily held in a perhaps really large spreadsheet.
  • In addition to structured data, you will often have text data and images.
  • Text in documents is often usable in predictive models by incorporating some preprocessing techniques.
  • Neural networks are needed to process image, video, and similar data.
  • Deep learning is a technique for composing neural networks in layers.

Break: 10 mins

Segment 4: relevant technologies (50 minutes)

  • AI projects usually create software; the most popular programming languages are R and Python, which target different use cases and come with extensive libraries.
  • Automated machine learning is emerging. It will help your data scientists get started and will let some business analysts build initial versions of some models.
  • When you deploy your model into operations, you will need to monitor its results. Software is emerging to help you do that.

Course wrap-up: 10 mins