3 Steps For a Successful AI Deployment

More than a century later Mark Twain’s advice,“The secret of getting ahead is to get started” remains highly relevant for today’s business leaders. But let’s face it, starting on a project is a daunting task. Determining when and where to start investing your company’s limited resources holds strategic importance. Business leaders sometimes approach a project as one-size-fits all. No two projects are the same just like no two individuals ever are.

Still organizations need to act swiftly to meet stakeholders expectations in deploying Artificial Intelligence (AI), but they also need to proceed cautiously to avoid violating regulations or ethical standards, particularly in areas such as data privacy and bias. Below, I outline a few steps to get started.

Step #1 Begin With the End in Mind

Begin with the end in mind, or Habit #2, as highlighted in Steven Covey's influential book “The 7 Habits of Highly Effective People (Organizations).” Covey emphasized the importance of commencing a new project with a well-defined vision or goal in mind of the desired destination.

When organizations look to implement groundbreaking technologies such as AI, a common error is prioritizing capabilities over assessing how that technology fits with the overall business strategy. And it’s challenging to avoid getting sidetracked by the AI hype that we find in the media these days. McKinsey’s recent report predicts that Generative AI impact on productivity alone will be in the trillions of dollars, while Gartner places Generative AI at the top of most hyped technologies in its 2023 Hype Cycle for Emerging Technologies. As a first step, start by evaluating your organization’s business objectives, including revenue goals, how AI implementation would empower your workforce, your products and your end-clients. Some common use cases:

• improve customer-service solutions or customer journey,

• automate repetitive tasks,

• enable discovery of data-rich information (research, due-diligence),

• manage risk and prevent fraud,

• And many more

Once the business goals and the numerous use cases are laid out business leaders must assess AI capabilities within their own organization.

Step #2 Assess AI capabilities within your organization

As previously stated, launching an AI project is not a straightforward “plug-and-play” process. When evaluating capabilities, leaders should adopt a holistic approach to implementing AI throughout the entire organization, rather than running isolated pilots. Organizations should also take time to instill an AI Guild to ensure upskilling and knowledge that goes beyond a team or squad. This is a pivotal part, emphasized recently by a Chief Information Officer (CIO) of a large technology company in Silicon Valley: “We are all learning as we go”about where and how to create value from AI in our organization. An AI project needs to include the following elements:

• Buy-in from key leaders and stakeholders within and outside the enterprise (customers and partners) to assess the impact.

• Availability of the right talent, both technical and operational.

• Data Resources. AI systems operate on large amounts of relevant and reliable data. If no data is available organizations must obtain it from various sources such as client surveys, emails, public datasets, APIs and so on.

• Model Accuracy. Model accuracy is an important measure to keep a close eye on. The initial data set that was used to train the model does not preserve its accuracy when new data and more diverse data is introduced.

• Interpretability. An interpretable model allows us to understand how and why a model arrives at a certain prediction? The more interpretable a model is, the easier it is for humans to trust.

• Security, Risk and Compliance. Establish appropriate guardrails in place to ensure data quality, transparency, security, regulatory (EU AI Act) and compliance - especially for regulated industries such as finance and healthcare - within your AI system.

• Ongoing monitoring. A comprehensive monitoring system is necessary to track any gradual or sudden changes affecting the predictability of an AI model. The National Institute of Standards and Technology (NIST) suggests four high-level categories for tracking AI systems in practice: Govern, Map, Measure and Manage. The Measure category is further broken down into: Monitoring Metrics, Input/Output Monitoring, Model Monitoring, Review and Action Plan.

Step #3 Ensuring human-centered and responsible AI technology

Artificial Intelligence is not a technology of the future, it is already here. And it’s moving at a rapid pace. Business leaders that embrace the technology must also prioritize ensuring that AI is human-centered. This means that the development and deployment of AI systems should consider the impact on the organization, its employees, its clients and the society.

The recent Executive Order by the Biden Administration provided necessary guidelines for how the public and the private sector need to develop and deploy AI. Top priorities were given to safety and security, data privacy, promoting innovation and competition, equity and civil rights among others.

One aspect of human-centered AI technology is trust - the token of any human interaction. Trust is crucial for widespread adoption and acceptance of AI. Users need to confide that AI systems are reliable, transparent and trustworthy (accountable). When individuals can understand how and why AI arrives at a certain prediction or recommendation, they are more likely to trust and engage with the technology.

In conclusion

Based on conversations with a number of business leaders, it is becoming evident to me that ChatGPT and other Large Language Models (LLMs) are being actively experimented with and utilized across organizations and in various industries. This curiosity ensures business managers are better equipped to meet ever-evolving expectations of customers and employees by delivering value faster, efficiently and hyper-personalized products/services - think Spotify's AI-curated list of music, podcasts, and now audiobooks.

Deploying an AI system is a balancing act that depends on the company's size, business model, industry. Every organization has its distinctive capabilities and competitive pressures, therefore starting a project by reflecting on your own business goals, strategy and talent is the first step towards taking on an AI initiative. Everything needs to be examined within the context of the whole or start with the end in mind.