Best practices for building an enterprise artificial intelligence platform
The intelligence might be artificial, but the trend is real. AI has emerged as one of the most promising and in-demand capabilities for achieving any number of goals—from enabling new and different offerings to improving speed, quality and efficiency of existing products and services.
Go just a bit below the surface, though, and it quickly becomes apparent that AI represents a diverse set of tools and technologies. Most of these are niche products; many are increasingly “plug and play.” Yet these technologies are emerging and changing faster than budget cycles.
Meanwhile, as data-driven enterprises focus on building different types of AI, they face some common challenges.
In some cases, AI projects are taking too long. That can be due to a number of recurring obstacles, including not having a consistent platform, or due to a lack of talent as well as a lack of clean, accurate data for training the AI. Others are conducting proofs of concepts that demonstrate good value, but they remain hesitant to deploy them into production due to a lack of formal governance. And while some may be unable to identify any strong use cases, others are overwhelmed in trying to manage a set of highly dynamic use cases.
Chalk it up to business models that demand significant technological agility and fuel uncertainty about what the enterprise may need in the not-so-distant future.
Tackling this complexity may feel akin to building an airplane while in flight. How do you decide where to invest when you aren’t sure what you need, what solutions will meet those needs and what resources you’ll need to deliver them? The answer lies in developing an enterprise AI platform.
Enterprise AI Platform
An enterprise AI platform is a framework for accelerating the full life cycle of enterprise AI projects at scale. It gives organizations a structured yet flexible way to create AI-driven solutions today and over the long term. It also enables AI services to scale from proofs of concept to production-scale systems. It does so by incorporating specific guidelines drawn from the world of service-oriented and event-driven architectures.
When designed well, an enterprise AI platform facilitates faster, more efficient and more effective collaboration among AI scientists and engineers. It helps contain costs in a variety of ways—avoiding duplication of effort, automating low-value tasks and improving reproducibility and reusability of all work. It also eliminates some costly activities—namely, copying and extracting data and managing data quality.
What’s more, an enterprise AI platform can help in tackling skills gaps. It not only serves as a focal point for onboarding new talent, but also helps in developing and supporting best practices throughout a team of AI scientists and machine learning engineers. And, it can aid in ensuring that work is distributed more evenly and completed more quickly.
Within an enterprise AI platform, the elements are organized into five logical layers.
- The Data & Integration Layer gives access to the enterprise data. This data access is critical because in AI, developers are not hand-coding the rules. Rather, the machine is learning the rules based on the data it has access to. Data components also include data transformation and governance elements to help manage data repositories and sources. Data sources may be wrapped in services that make it possible to interact with the data at an abstract level, providing a single reference point to an existing platform data ontology. Most importantly, the data must be of good quality, and the AI scientists must be able to build the data pipelines they need without dependence on the IT team—ideally through easy self-service—so they can experiment as needed.
- The Experimentation Layer is where AI scientists develop, test and iterate their hypotheses. A good experimentation layer brings automation for feature engineering, feature selection, model selection, model optimization and model interpretability. Idea management and model management are both key to empowering AI scientists to collaborate and avoid repetition (see sidebar).
- The Operations & Deployment Layer is important for model governance and deployment. This is where the model risk assessment is conducted so that the model governance team or model risk office can validate and see proof-of-model interpretability, model bias and fairness, and model fail-safe mechanisms. The operations layer includes the results of experiments by AI DevOps engineers and system administrators. It offers tools and mechanisms to manage the “containerized” deployment of various models and other components across the platform. It also enables the monitoring of model performance accuracy.
- The Intelligence Layer powers AI during run time, with training-time activities handled in the experimentation layer. It is the product of technical solutions and product teams working in conjunction with cognitive experience experts. The intelligence layer can expose both re-useable components, such as low-level service APIs, to intelligent products that are composite orchestrations of many low-level APIs. Central to the orchestration and delivery of intelligent services, the intelligence layer is the primary resource for directing service delivery and can be implemented as simply as a fixed relay from requests to responses. Ideally, however, it is implemented using such concepts as dynamic service discovery and intent identification to provide a flexible response platform that enables cognitive interaction despite ambiguous directions.
- The Experience Layer engages with users through technologies such as conversational UI, augmented reality and gesture control. AI platforms are a growing area encompassing the components that enable the visual and conversational design work for a solution. It is generally owned by a cognitive experience team with traditional user experience workers, conversational experience workers, visual designers and other creative individuals who craft rich and meaningful experiences enabled by AI technology.
Below we examine the operations layer in greater depth to discuss the governance and deployment of AI.
In the Operations layer: Managing ideas, models and configuration
As an organization explores opportunities to use AI, it needs a formal approach for keeping track of those ideas: testing the possibilities, capturing what works and maintaining an “idea graveyard” for concepts that have been tested and determined to be untenable.
That might sound simple enough, but the potential quantity of ideas, and nuances among them, can quickly become overwhelming. To avoid that complexity, firms should design and implement an automated idea management process for tracking and managing the lifecycle of ideas and experimentation. Doing so helps in tracking idea performance and ensuring the quality of ideas. There are also efficiencies to be gained by providing team-wide visibility to successful ideas and managing duplicate works and potential conflicts.
A similar approach can be applied to managing models. Building real-world machine-learning algorithms is complex and highly iterative. An AI scientist may build tens or even hundreds of models before arriving at one that meets some acceptance criteria. Now, imagine being that AI scientist without a formal process or tool for managing those work products.
A formal process for model management will alleviate that pain for the individuals and the organization. It makes it possible for AI scientists to track their work in detail, giving them a record of their experiments. Such a process also enables them to capture important insights along the way—from how normalization affected results to how granular features appear to affect performance for a certain subset of data.
Across an organization, sound model management empowers data scientists to review, revise and build on each other’s work, helping accelerate progress and avoid wasted time. It also enables the organization to conduct meta-analyses across models to answer broader questions (e.g., “What hyperparameter settings work best for these features?”).
To succeed at an enterprise scale, an organization must be able to store, track, version and index models as well as data pipelines. Traditional model management should be expanded to include configuration management. Logging each model, its parameters and data pipelines enables models to be queried, reproduced, analyzed and shared.
Consider, for example, that model management will track hyperparameters that have been tested and record what was eventually used for deployment. However, model management will not simultaneously test what features were tested and discarded, what modifications were made to data pipelines or what compute resources were made available to support sufficient training, to name just a few key activities.
Together with model management data, tracking that kind of configuration information can accelerate the deployment of AI services while reducing duplicate work. An organization will never achieve that level of visibility and analysis when managing models via spreadsheets.
Make Knowledge a Service
An enterprise AI platform paves the way for an organization to deliver intelligent services and products, empowering not just AI scientists but all workers and customers to tap into the tool or combination of tools they need.
Untethered to any one AI type or solution, it enbales the rapid provisioning of a high-performance environment to support virtually any kind of AI. In short, it transforms AI from a series of finite point solutions to an enterprise capability that can be continuously improved as it is tailored and deployed to meet business goals over time.
As AI becomes a mainstay of every data-driven organization, it must be managed strategically—with a focus on agility and extensibility. The most successful organizations will be those that take the time to build their enterprise AI platform platform for the business. With this approach, they can deliver more value more quickly—not just today, but also as new opportunities take shape in the future.
With an enterprise AI platform—rather than a patchwork of standalone tools—an enterprise will be well positioned for advancements in AI use cases and other emerging and supporting technologies.