AI, unlike any other initiative is a business transformation enabler and not another technology system implementation that business users need to be trained on. Traditionally, businesses choose either the classic waterfall approach of linear tasks, or the agile approach, where teams review and evaluate solutions as they are tested out.
In contrast, implementing AI technology requires a different approach altogether. AI requires that you look at a problem and see if there’s a way to solve it by reframing the business process itself. Instead of solving a problem with a 10-step strategy, is there a way to cut it down to six steps using data already available or by using new types of untapped internal or publicly available data and applying AI to it? A study by IDC last year found that 60% of organizations reported changes in their business model that were associated with AI adoption.
But implementing AI successfully is proving difficult — perhaps in part because it requires such a different approach. The IDC study found that most organizations reported some level of failure among their AI projects. A quarter of the organizations surveyed reported a 50% failure rate due to unrealistic expectations.
To ensure success during a period of transformation, enterprises need to embrace three concepts.
1. Understand the human impact
Introducing AI is not like introducing a new software program; it will impact how employees and customers work, behave, and make decisions. AI technology in the workplace will create new opportunities for employees to learn new skills. Employees’ domain knowledge is key to getting AI right, and employees are also valuable in edge case scenarios, where AI doesn’t have the correct context, capability, or parameters to respond appropriately. It’s important to properly define the human inputs that successful AI projects depend on and to adjust employee roles to provide that support.
For example, there are medical call centers designed to respond to patients in medical trials and discourage them from discontinuing drug treatment. Employees in the call center transcribe calls for managers so they can follow up. However, natural language processing (NLP) in AI can transcribe calls and highlight “problem phrases” that indicate an individual will stop their treatment.
With this new transcribing technology, the business no longer needs employees that can type quickly and accurately. Instead, it needs empathetic workers who can intervene and provide real-time guidance to patients who are going to stop treatments. This is not simply plugging in a new technology; it is altering the business model and requires more in-depth change management.
2. Lead with design thinking
Understanding the human impact is just the beginning of how relentlessly one must focus on the end-user in AI implementations. Companies need to understand the goals they are trying to achieve at the human level instead of just at the business processes level. How will the application improve an individual’s experience, whether that of an employee or customer? In the past, engineers built new technologies alone at their desks based on a set of requirements. This linear approach to development is no longer sufficient. Due to the complexities of AI, a new, non-linear approach must be deployed.
Design thinking is an iterative process of observation, ideation, prototyping and testing, that ensures the end-user is central to all decisions related to the technology. The product is tested (either formally or informally) throughout, which allows the engineers to not only pick up on the explicit feedback given but can gain an understanding of their unsaid preferences and behaviors. Through this design thinking process, AI technology can blend in seamlessly and intuitively to the user experience, becoming almost invisible.
In one example, a cloud computing company’s service portal lacked human-centric design, which led to a poor user experience for its clients. When auditing the application, the company discovered that the portal was dated and had an inconsistent user interface, a fractured information architecture and poor navigation features. The cloud computing company first approached its client’s employees and used design thinking workshops to come up with an employee-centric approach to the application, instead of rolling out an updated application that was function-specific. The employees wanted information that was centralized, contextual and easy to find. The cloud company rolled out a personalized dashboard with an AI-driven virtual assistant to help these employees quickly find information. By reimagining the platform’s applications from the employee point of view instead of only focusing on user functions, the company was able to ensure a successful launch and 81% adoption rate. If the company had rolled out a function-specific AI solution, it’s likely the adoption rate would not have been as high. And what’s the point of investing in a new AI solution if nobody uses it?
3. Learn to embrace failure
Implementation of a traditional CRM is almost guaranteed to be successful. AI requires a higher tolerance for failure. Companies need to expect to fail and learn from the consequences rather than abandoning their efforts ahead of a new fiscal quarter. Each failure leads to new insights that will ultimately create value for the company.
This type of optimism is especially important because businesses will be working with unstructured data when they work with AI technologies. Unstructured data seems chaotic, but it can be an opportunity across the enterprise. For example, in the healthcare space, the population that uses Medicare is the most vulnerable segment of our society. Right now, the medical community only uses structured data like embedded notes to track patient concerns and progress. With AI, the solution can look through the unstructured notes for social determiners of health, including access to clean water, reliable shelter, and access to food through food stamps. It’s not mandatory for doctors to track social factors in their notes, but AI can do this by scraping details gleaned from data. After, AI can put this information back into the structured notes and provide a new level of context.
However, it takes time for AI to parse unstructured data and organize it into sets that are useful to employees across business lines. Engineers and programmers will be needed to help AI figure out how to make this data useful for other employees. This can take multiple brainstorming sessions, and it might not result in a profitable or useful outcome. It’s important for all shareholders to understand that these are exploratory and experimental solutions that may not yield immediate ROI but are essential for preparing the business for the AI revolution. Executives and shareholders looking to see results in a short time period will be disappointed and tempted to scrap the implementation project all together.
To combat this impulse, companies need to accept failure and start the implementation with an experimentation mindset. Because AI is such a revolutionary technology, full implementation is too challenging of a goal. This experimentation mindset must start with investors and executives and continue down to the rest of the employees.
One example of an AI misstep is in healthcare. In early applications of AI solutions, they’ve suggested unsafe recommendations for patients undergoing treatment. The advice that AI provided was unsafe because most of the data fed into the machine was hypothetical medical data rather than real patient data. Instead of viewing the poor performance of these early AI solutions as failures, the lesson learned is that hypothetical data can’t be used in place of real data.
AI implementation will be a true learning process for every employee across the organization. By understanding the human impact, leading with design thinking, and fostering a culture where innovation becomes mainstream, setbacks are embraced, organizations can ensure that the AI will reach its promised potential.
All Rights Reserved for Bret Greenstein