For organizations wishing to move toward a product orientation for analytics and AI offerings, they should check that several key assumptions and prerequisites are in place and confirm that a number of capabilities are present within the organization. If they are not, the company will probably want to address the prerequisites and build the capabilities.
The following lessons are derived from product management principles in general; companies that have adopted product management approaches to analytics and AI; and consultants who work with firms on analytical product design and development. Brian O’Neill, such a consultant who is an IIA subject matter expert, provided a summary of his perspective on design issues.
Key Assumptions and Prerequisites
The value of a product focus in analytics and AI initiatives depends in part upon the initiatives themselves, and also upon the broader context of assumptions and prerequisites that facilitate them. Below are a set of attributes and prerequisites that should be in place if a product orientation is to succeed:
- The analytics or AI project being considered as a product involves significant technical capabilities and involves some form of software development or customization; it is not something acquired from a vendor and useful out of the box. If it is easily acquired and installed, it may not need an extensive product focus.
- The goal of a product is deployment of the system and analytical or AI capabilities, not just a pilot or proof of concept. As such, it’s critical that the system be widely adopted and employed by the intended users.
- The process or activity supported by a product should be somewhat structured (with a predictable set of inputs and outcomes) and repeated often. Otherwise, it won’t yield enough data to use in building analytical or machine learning models.
- Data products involve a substantial need for change in business activities as well as technical capabilities in the proposed system. Changes in business strategies or models, business processes or workflows, user skills, or customer behaviors would suggest that there are complex activities to be coordinated and that business stakeholders need to be influenced and communicated with. This is one of the primary benefits of a product approach to analytics and AI.
- There is a certain scale to the project. It affects large numbers of users or customers or has a substantial potential impact on the company’s fortunes. Otherwise, it may not be worth the trouble to take a product focus. One way to achieve scale is for the product to be used in multiple places within an organization. A dynamic pricing application, for example, that is created as a product should be intended for use by multiple business units and geographies within a company.
- The company implementing the system behind the data product employs some form of agile or iterative development process. That process is probably the easiest way to inject a product focus because it already incorporates frequent deliverables and communications with stakeholders. It also increases the likelihood that the project will be successfully developed in a reasonable period.
- A company should have a clear categorization of analytics and AI roles if it is going to establish a product management function. The product manager, of course, should be one of those roles. They should also include data scientists, data engineers, machine learning engineers, and perhaps machine learning operations specialists and analytics/AI translators. Analytics and AI translators may also play product management roles, or vice-versa.
- A well-designed process or pipeline for analytics and AI products is critical to the success of a product management function. It should begin with opportunity identification and encompass such steps as data collection, model development, system design, UX design, system integration, testing, and ongoing model management. If a company doesn’t understand the key steps in the process, it is unlikely to perform them well.
- A set of ethical guidelines and processes for analytics and AI projects will help to prevent issues with algorithmic bias, lack of transparency, data privacy issues, and inappropriate application of the analytics or AI system. A product manager can be primarily responsible for ethical compliance throughout the development, deployment, and maintenance process.
Key Capabilities
There are also a number of capabilities an organization needs to have in place in order to make analytical and AI product management successful. Some of these capabilities are generally desirable for analytics and AI groups within companies, while others are more specific to a product focus.
Design thinking is a critical capability of organizations pursuing data, analytics, and AI products. It should be the primary approach for identifying needed products and attributes from business strategies and needs. Design thinking can be used with a focus on external customers or on internal users. The output of a design thinking exercise can also provide high-level direction for user experience efforts.
If analytics and AI teams want their products to be used by customers or employees, they need user experience (UX) design capabilities. Just providing an answer or data-based insight is not enough. The information from analytics or AI models must be easily accessible, with clear explanations of their implications whenever possible. UX designers have not historically focused on analytics and AI systems, but many of the general principles for other types of systems are applicable. This self-assessment can help to identify the need for UX design.
Successful data products often require participative workflow and job design, because analytics and AI systems often lead to changes in business processes and the jobs of employees (or in some cases customers) who use the systems. These changes may involve only incremental change (as in Lean Six Sigma programs) or more dramatic change (as in business process reengineering programs).
Experimentation is another important capability for AI and analytics-driven change, even when the systems that contain the models are intended to be fully deployed into production. There is uncertainty around many aspects of AI and analytics development. Most sophisticated organizations have the ability to incorporate A/B and multivariate testing into their models.
A collaborative culture is essential for successful product management in analytics and AI. Product managers are “ministers without portfolio” and rely on influence, communication, and collaboration rather than authority. Organizations that don’t have collaborative cultures are likely to end up with frustrated product managers and unsuccessful data products.
Product-oriented organizations also require a self-service offering for less important analytics projects that can be done by their users or by business analysts. A product orientation implies that the products are important and of substantial value to their organizations. However, there is still a need for ad hoc reporting and small-scale analysis. Analytics and AI groups should offer easy-to-use self-service tools and provide training in their use.
Finally, there is an obvious need for organizations pursuing a product approach to have a general capability to deliver on analytics and AI projects and products. A set of structures and processes for creating products may be present, but if the involved employees don’t have the needed skills to develop and implement high-quality models, products won’t be successful.
Despite these recommended prerequisites and required capabilities for success, many organizations are moving in these directions anyway, and will not find it terribly difficult to adopt a product focus. The benefits of thinking about analytics and AI initiatives as products are many, and the downsides few—at least for large organizations that already have substantial AI capabilities.