There's no doubt that Artificial Intelligence (AI) offerings are bound to change the way we work and do business. Thus, AI and its underlying technology is currently top of mind for many high-level executives. But, AI adoption still has an air of mystery about it. Should you build it from scratch? How can you trust the technology? What about proper implementation?
To share our expertise and shine light on some of the questions surrounding AI adoption, we attended the AI Summit in New York, a conference dedicated to uncovering the implications of AI solutions. Here, Melissa Boxer, Vice President of Adaptive Intelligent Applications at Oracle, spoke on the five practical considerations for AI adoption in the enterprise. Boxer's advice: Know your why, decide whether to build or buy, inject transparency for trust, go all-in on cloud, and start with smart data.
"AI is a journey and it's measured in purposeful steps." - Melissa Boxer, Vice President of Adaptive Intelligent Applications at Oracle
To help facilitate a strategic approach for AI adoption, let's break down the significance of these five considerations and reflect on the importance of each step.
The success of AI implementation is largely dependent on asking the right kinds of questions and focusing on business needs. To that end, it's important to put actionability first.
Don't ask what AI can do for you; instead, take a comprehensive look at your organization and ask questions about your top business needs. Do you need to optimize sales and marketing programs to increase conversion and close more business? Do you need increased efficiency across payables and procurement? Do you need to analyze supplier risk within your network? Do you need to reduce time-to-fill and cost-to-hire of new employees? Once you've pinpointed compelling business objectives and asked the right questions, assess the viability of answering these questions with AI solutions.
While every organization prioritizes their business needs differently, understanding your desired outcomes allows you to think of AI solutions in terms of actionability and business impact. If you're thinking in terms of ROI, it doesn't have to be monetary - it could be increased productivity or greater efficiency. Regardless, AI should be a means towards the same end, transforming your business with smart decisions to deliver better results.
Once you've understood why you need to deploy an AI solution, it's time to think about implementation. Essentially, this is a choice between building an AI application and buying a pre-packaged solution.
If you have a niche use case or are in a specialty industry with unique data needs, it might make more sense to build. Similarly, if you find there are no commercially available pre-built applications, then the default choice will be to build. While building an AI application is an immense undertaking, if you have a strong culture of development and a readily available team of data scientists, then you already have some of the foundational pillars needed to begin such a challenge.
Alternatively, if your use case is common and there is already a pre-packaged application available, you should buy. Purchasing a ready-to-go AI application provides a lot of advantages: lower barrier to entry, shorter time to value, lower risk, and lower cost. Many pre-built AI solutions even come with third party data to help you optimize outcomes and spend less time wrangling necessary data. Purchasing a pre-packaged AI application also means you're transferring risk -- maintaining the application, refining the data science, validating data quality, and following security regulations all fall to the vendor.
Not having to develop a platform and build the architectural components needed to maintain and integrate an application means you're seeing benefits in near-real time. However, if building an AI application is your preferred choice, then AI cloud platforms make it easy for you to get started.
Even though AI applications have the power to transform your business with smarter decisions and better results, people have an inherent tendency to distrust what they don't understand. Much of the decision-making involved in AI applications is usually too difficult for most people to understand; that makes the need for transparency a crucial aspect of any AI solution.
One way transparency creates trust is through explainability. While most end users don't want deep dives into the technical workings of an algorithm, insight into how the machine made a decision based on data inputs allows us to augment our intuitions with data science.
Let's look at a fraud detection use case as an example. There's no need for a user to know exactly how the machine determined fraud, but insight into the signals, clues, or anomalies that lead to the suggestion of fraud provides us with the understanding needed to refine our decision-making and trust that the AI outcomes are correct.
With much of AI still requiring some degree of human guidance, another method for transparency is allowing humans supervisory management of the machine. Being able to adjust desired outcomes to meet business needs, allows for some sense of control and trust in the machine making correct decisions. For example, as a merchandiser, you may be trying to get rid of certain inventory. While an AI application would not otherwise recognize this as a changing business need, human control over adjusting a specific product constraint ensures that the likelihood of recommendation goes up and ultimately inventory clears out. Control over this process and transparency into the subsequent outcomes ensures that AI deliverables are trusted and continue to align with your organization's needs.
AI use cases are naturally quite complex and require large volumes of high-quality data in order to perform successfully. The cloud thus provides a scalable, secure, low-cost, and high-performing infrastructure on which AI applications can be deployed. Paired with the availability of ancillary APIs and third party data vendors, the cloud makes it easy for AI solutions to consume large amounts of data and integrate across different business units.
Fundamental to the success of any AI solution is data quality. If better business decisions are the output and large volumes of data are the input, then you need to be certain your input data is pristine. Otherwise, any AI analysis will be rendered useless.
To elaborate on matching your data strategy with your AI strategy, we had the chance to listen to two more experts: Craig Muraskin, Senior Managing Director of Innovation at Deloitte; and Bastiaan Janmaat, former CEO of DataFox and current Vice President of Product Management for Fusion Adaptive Intelligence at Oracle.
As Muraskin pointed out, "Applying a specific AI technique - regardless of its use case - always starts with data and ends with insights." Data is, of course, infinite and almost always messy, which is why sourcing and wrangling data is so critical. That's why Deloitte partnered with Oracle DataFox.
"We've built a smart data engine that automatically structures data on millions of companies around the world. Focusing on breadth of company coverage, depth of data points, accuracy of data, and easy configuration; Oracle DataFox is a one-stop shop for your smart data needs," explained Janmaat.
Muraskin's favorite example of this synergy between AI and data is a document classification and interpretation use case - reviewing and deriving information and insights from documents (contracts, financial statements, emails, etc).
"We used to have experts reviewing 100,000's of contracts which themselves could be hundreds of pages. Typically, trying to find just several key terms or provisions. We then developed an NLP/ML enabled solution that allows us to automate the ability to review and extract insights from documents at scale," explained Muraskin. This solution has saved hundreds of thousands of hours and enables Deloitte to provide clients with much greater insights because this intelligent AI automation can now asses full populations versus just samples. Auditing a large company's contracts to confirm total revenue, fees, schedules, and/or terms becomes easy with the right AI deployment and the right data input.
The utopian state of AI adoption is achieving both scale and precision in solving your business objectives. While the road to AI may be windy, the considerations in this article provide guidance on where to begin: Know your why, decide whether to build or buy, inject transparency for trust, go all-in on cloud, and start with smart data.
Learn more by checking out Oracle DataFox.