This blog post was first published on the Element AI blog and is the third post in a four-part series on intelligent AI adoption.
In the first two posts of this series, we discussed the why and when of intelligent AI adoption. To unlock the transformative benefits of artificial intelligence, you must adapt AI to your business and adapt your business to AI. And because of the distinct challenges of starting, executing, and scaling your AI efforts, you need to begin this work as soon as possible.
This post is about the how, unlocking the benefits using the four pillars of intelligent AI adoption: Strategy, Technology and Data, People and Organization, and Governance.
These four pillars address the biggest questions that face every business, at every step of the AI adoption journey. The straightest path through is around: starting with Strategy and moving through each subsequent pillar, then repeating over time as your business, industry and AI itself continuously change.
In Strategy, you consider the paramount question of what value you will create with AI and how. Then, you prepare Technology and Data to power and differentiate AI capabilities; prepare People and Organization to be ready, willing, and able to work smarter with AI; and prepare Governance to secure safe, trustworthy, responsible AI into the future. This simple framework is the cornerstone for incorporating AI into your business.
Strategy comes first, because building a coherent AI strategy is the only way to achieve your short- and long-term goals. AI is reshaping the competitive landscape, and there are many potential areas where AI could be applied in any business. It’s important to be deliberate about which AI opportunities you pursue, and in what order.
Our observation is that companies that skip the strategy step end up making false starts: AI experiments that don’t lead to strategic benefits or lasting success. At a minimum, we think businesses need to spend time on three activities:
1. Studying possible future scenarios,
2. Aligning on a shared vision of which future to pursue, and
3. Building a strategic roadmap to achieve that future.
First, we advise pausing to assess trends that impact the future of your business, your industry, and AI itself. Exploring possibilities at the outset enables you to take next steps using an entrepreneurial frame of mind instead of a reactionary one.
We rely on foresight for this activity, though you can use any discipline that helps you increase the field of strategic options and stockpile thinking in the face of multiple possible futures. For example, AI systems may help your business attract talent in a tightening labor market, redefine your customer experience, or solve core industry challenges in new ways. To discover opportunities like these, you need to look beyond the boundaries of today’s operational environment, and beyond quarterly and yearly reporting cycles.
Second, align your leadership team on which possibilities take priority by setting a shared vision. Aligning on a shared vision will clear the path for all future work that comes next: unlocking budgets, guiding the decision-making of managers and employees, and designing the right checkpoints to stay on track.
Finally, once a vision is established, you can define a roadmap of AI use cases and applications. An AI use case combines one or more AI capabilities, such as machine vision or natural language processing, with a business task, such as reading and understanding loan documentation at a bank. Use cases create enough definition around opportunities to validate which ones are technically feasible and will have a positive impact on users and the business.
Use cases create enough definition around opportunities to validate which ones are technically feasible and will have a positive impact on users and the business.
Technology & Data
After establishing your AI roadmap, it’s time to bring Technology and Data up to speed for AI. The roadmap should start from feasible use cases, but businesses almost always need to complete significant prep work before their technology and data are ready for AI. We’ve observed that businesses often need to augment data, upgrade infrastructure, or both.
In data, there are 5 dimensions we typically validate:
- Volume: Is there enough data to train AI models?
- Representativeness: Is the data varied enough to capture the range of cases found in production?
- Quality: Is data sufficiently well structured and free of gaps and errors to be used using existing methods?
- Labeling: If using supervised learning techniques, is data labeled properly to enable AI models to understand examples?
- Accessibility: Is data accessible for development as well as production environments?
One of the typical gaps in data is having it labeled properly for training AI models. There are many ways to accelerate data labeling, but not all of them are useful beyond the simplest cases. If data is too sensitive or requires subject matter expertise to interpret, you’ll need to decide whether to dedicate expert talent to labeling data all at once, or incrementally by adding a labeling step to existing work processes. The first approach takes less time and concentrates cost, while the second approach is cheaper and takes much longer.
Once data is in place, you still need technology to onboard it, build models with it, and scale data-driven solutions in production. The biggest gaps in technology are usually around supporting new workflows with heavy yet rapidly evolving requirements. Tools need to be flexible enough to connect to different kinds of data, handle different kind of models, and scale to different production scenarios.
Flexible deployment models enable faster and easier adoption of AI within the enterprise — whether public/private cloud, on-premises, or leveraging existing Intel® Architecture infrastructure through optimized frameworks.
In practice, many different tools and frameworks exist and each has different strengths. Rather than relying on a single solution, we recommend setting up a flexible toolchain that helps you with the best tool or framework for each task along the way.
Finally, we advise building a deliberate technology and data strategy for the long term. Technology and data are key for differentiating your use of AI over time, and you need to be intentional about supporting new and creative ways of acquiring data sets, storing and serving data in privacy-sensitive contexts, and scaling capabilities across an ever-evolving regulatory landscape. New frameworks and modeling platforms, policies and processes, partnerships and M&A may all be required to succeed.
People & Organization
After developing a roadmap and preparing your technology and data, you need to work on training people and designing the organizational support they need to confidently invest, manage, and work alongside AI applications. Even the most sophisticated AI systems will not be used if people aren’t ready and willing to use them.
People across your organization will need to build AI literacy and skills, and that means more than just technical training. As people work more closely with AI at all levels of an organization, the nature of their work changes. People will need to learn how to recognize AI opportunities and how to make the most of these opportunities from the standpoint of their own role.
Not only should AI training occur at the start of your AI journey — such as in short sprints to enable AI strategy development — but you must think about long-term shifts for employees over time. Using your strategic roadmap of AI use cases, inventory any AI capabilities that your people and organization need to be ready. Your use cases will determine what the potential impacts will be, and by working with HR leaders, you can minimize these impacts by upskilling, hiring, or partnering with outside help to prepare teams.
To fully leverage the narrow yet powerful abilities of AI, you’ll also have to organize people in new ways. One of the considerations is how to manage AI itself in either a centralized or decentralized manner. There isn’t a universal answer for every business. An AI center of excellence concentrates resources and decision authority, while in the other, AI is lead by multiple parts of the business simultaneously. Rather than deciding on a structure accidentally, figure out which approach makes more sense for your AI roadmap and vision.
Crafting a deliberate organizational strategy is also critical for people who don’t work on AI, but do work with it. We know that 133 million new jobs are being added to the economy in the next few years due to AI adoption, while 76 million jobs are declining. You’ll need to figure out what that means for your business in terms of reskilling or reshaping teams, hiring under the new paradigm, and designing completely new jobs. Many jobs will need new technical skills, but many more will need non-technical skills including innovation, active learning, and emotional intelligence.
Whether moving fast or slow, you need to be proactive about designing and implementing Governance for AI, including new policies, procedures, and principles to ensure safe, ethical, and trustworthy AI. This isn’t just about compliance. The foundation of every business interaction, with or without technology, is trust, and the same is true for AI adoption.
The challenge businesses face is that current approaches to governing IT systems are not comprehensive enough for AI applications. Plan to work collaboratively across business, technical, risk, and compliance teams to define mitigation plans for each use case in your roadmap.
For your first use cases, we recommend identifying a broad list of potential risks for every stage of AI application development. For example, because deep learning models rely heavily on data for both development and production use, you’ll need to monitor data for issues like bias at both stages: during initial training, and as the system is used on new data over time. Your use of specific facets in data may also be governed by regulation such as the GDPR in Europe or newly announced Digital Charter in Canada.
Your risk management won’t start from zero. After identifying risks along your AI roadmap, focus first on any new governance that’s needed above and beyond existing practices. Over time, these strategies can be synthesized into more top-down guidance that generalizes to more use cases.
Ideally, implementing your AI governance should extend beyond a simple checklist. Designers and developers should also be involved in crafting solutions—for example, by leveraging best practices that are now emerging in the field of AI explainability. An active area of scientific research, AI explainability lends AI systems the ability to demonstrate the chain of reasoning that lead to a prediction. In the future, Explainability by Design may become the standard, and your governance should reflect that approach.
Starting with the end in mind
The four pillars of Strategy, Technology and Data, People and Organization, and Governance guide you systematically through intelligent AI adoption. By taking each in turn, your progress builds on the insight from previous steps: first setting strategic and realistic goals; then assembling technology and data to achieve those goals, leading people to succeed in the new world they create, and finally installing the policies, procedures, and design principles to guide safe and trustworthy AI. On subsequent iterations through this cycle, new conditions and capabilities help you build and refine the progress you’ve already made.
Among all the activities that make up this journey, we’ve found the AI roadmap to be the single greatest accelerator of AI success when working with clients. When executed correctly, it guides and explains every decision that comes after. In our next post, we’ll dive deeper into how you can create one for your business.
Please contact firstname.lastname@example.org for more information.