Nearly all business leaders on the planet, 94%, believe AI will be critical to success over the next five years. Yet, as Deloitte’s latest research on the state of AI reveals, many companies are still not achieving the value they anticipated – there was a 29% increase in the share of respondents who s identify as AI “underperformers” this year compared to last year.
Issues that diminish the impact of AI include difficulties in improving its business value and lack of full management commitment, according to the Deloitte survey. Industry executives and observers in the trenches agree that it is these organizational issues, rather than technical ones, that are holding back progress.
An important point is that AI should serve the customer and help the business put the customer first.
Most AI projects fail to generate value “because they don’t start from business realities, such as the benefit of correct prediction, the cost of incorrect prediction, or constraints such as size of the marketing budget,” says Arijit Sengupta, CEO. and founder of Aible. “If your AI project looks and feels like a lab experiment, and your experts talk about things like lost logs instead of revenue, profit, and cost, your AI is almost guaranteed not to produce results.”
The key is to make decision-makers more comfortable and knowledgeable about AI, to create organizational support for AI, and to focus directly on how it can help the customer. “Organizations are more likely to embrace AI-based approaches when they tie directly to demonstrable customer value,” says Rajesh Raheja, director of engineering at Boomi. “For example, a recommendation engine that can show you what is the next step to implement in a business process based on the proven best practices it has learned will be much more useful to a business than the same engine that only presents more products the business can buy. Both require sophisticated AI models, but the former clearly creates value for customers.”
We have the tools, we don’t know how to apply them. “Adoption of AI depends on the value and ROI it generates versus the effort of training a model, which needs the right resources and skills to build a pipeline of solid data,” says Raheja. “Machine learning results change with data, algorithms and their evolution. The fear of an unknown decision leading to unforeseen liability is another factor that makes traditional businesses cautious.
As an example, adds Raheja, “an AI-based loan application refusing certain sections of the population after a model update could be an inherent error or bias introduced into the data and the model.”
It also has to do with the timeliness of AI data. “AI is actually a perishable good,” says Sengupta. “If your AI takes months to build, it’s trained on months-old data and the world has changed in the meantime – so the AI is no longer relevant. You need to build and deploy the ‘AI in days, not months, if you want to get value, then iterate as the world changes.
There are many innovative business cases that can be developed to drive AI adoption. Examples cited by Raheja include the use of AI “to analyze customer data entered into the system for formatting and semantic checks. Natural language processing and automated chatbots for customer relationship management are other use cases.
Keep in mind that “AI adoption is limited by the availability of data scientists, and we can’t get out of this problem,” adds Sengupta. We are trying to teach business users to speak AI instead of teaching AI to speak business. Before sending business users to learn Python, stop and say, “Why can’t AI understand my business needs and automatically generate Python code?” The Internet revolution didn’t happen because everyone learned to code to interact with the World Wide Web; it happened because the Netscape browser could be used by almost anyone.
#Artificial #intelligence #speak #language #business