Five Practical Considerations for AI Adoption in the EnterpriseFive Practical Considerations for AI Adoption in the Enterprise
Five Practical Considerations for AI Adoption in the Enterprise
December 10, 2018
by Melissa Boxer
NEW YORK - With AI, the first step is the hardest. Many enterprises feel compelled to jump on the AI bandwagon, but do so by trial and error and fail to derive the expected benefits as quickly as they would like. Others are reluctant to invest in AI, waiting to see how it plays out for early adopters.
In conversations with customers and prospects, it often becomes clear that customers don’t know where to start. My advice to organizations: Start with “the why”, think about “build vs buy”, consider transparency and move to the cloud. Once these steps are in motion, invest in smart data – the lifeblood of any AI solution.
Know Your Why
AI is too often a shiny object that organizations want. But they haven’t fully thought through why they need it. AI adoption is a journey that requires measured and purposeful steps. Each step needs to be approached with a clear objective in mind and with a clear return on investment goal in a set timeframe. Remember: ROI goes beyond monetary value and can include productivity gains and other benefits.
Take a realistic view of your business. What area needs improving? Do you need to improve conversion and increase revenue per customer? Do you have the right employees? Do you need to improve employee productivity, or increase morale? Do you have a leadership succession plan in place? Is your supplier network healthy?
Every organization will prioritize the use cases above differently. Having a clear reason why you are implementing AI will help to rank your AI needs and properly stagger deployments.
Decide Whether to Build or Buy
The next logical question is “how do I implement?” This essentially boils down to one of two paths: either you buy a pre-built AI application, or you build your own.
If you choose to buy pre-packaged AI applications, your implementation time will likely be shorter, costs lower and maintenance easier. For example, you won’t have to employ data scientists or pay for development platforms and architectural components to learn, buy, maintain and integrate.
Buying a ready-to-go AI application provides the lowest barrier to entry, near-immediate benefits and they are often bundled with third-party data sources. This also transfers risk from your organization to the application vendor, who is tasked with maintaining the application, securing and validating the data, and ensuring it is in compliance with local and global security and privacy regulations.
Cloud software vendors, like Oracle, have already created out-of-the-box AI applications for common use cases and business processes (e.g. Sales, Marketing, Finance, HR, Supply Chain and Manufacturing) that align with industry best practices. The best AI application to achieve your goal might already exist.
Alternately, building an AI application might make more sense for an organization with a unique use case. Sometimes niche industries require bespoke processes, which aren’t included in standardized applications that cloud vendors offer. Sometimes organizations have already invested in data scientists and resources to enable them to develop their own AI apps.
If your organization fits in this category, it might make sense to build your own AI application. Luckily, AI cloud platforms offer organizations an easy way to get started on building your own application, with tools and modules that remove much of the heavily lifting in building an AI app.
Related: Where AI Is Going - And How You Can Get There
3. Inject Transparency for Trust
AI and machine learning gives people better insights so they can make smarter decisions but, in most use cases, still requires human oversight.
A major issue is trust. The machine provides an output, but how can the user trust the machine made the right decision, or is recommending the right action?
To establish trust, a machine learning algorithm needs to show its working and what data was important for the machine to make a particular output. For example, in fraud detection use cases, a human must be able to see how the machine determines fraud. What were the clues and anomalies that led to the suggestion of fraud? This will become increasingly important as findings in AI applications become evidence in legal activity.
Human guidance and supervision is also important to ensure users can trust the output. Supervisory controls provide users with the ability to boost or constrain outputs from the machine learning algorithm. End users are not developers and need to be able to fine tune or adjust the outputs without knowing about the complexities of the machine learning algorithm.
Insights into the data inputs the machine learning application leveraged and Supervisory Management features help organizations spot errors, anomalies and bias and enable algorithms to be adjusted to improve the quality of outcomes. Without these features, it’s hard for the humans using the applications to trust the accuracy and validity of results. Transparency is a difficult problem, but proper design of an AI use case should make transparency and insights a high priority.
Go All-in on Cloud
The complexity of AI solutions and the performance of infrastructure needed to run a continuously learning application makes cloud a necessity for AI deployments.
Cloud provides organizations with a cost effective, easy to maintain infrastructure that is quick to deploy and easy to scale. AI apps and intelligent platforms for AI development are only as effective as the infrastructure that they sit on.
Another benefit of running AI in the cloud is the ready availability of AI services through standard APIs. Enterprises need not worry about choosing the right machine learning algorithms or training custom models. And, exposure of voice and text bot services through the cloud enables enterprises to easily develop popular digital assistant and chatbot capabilities.
Cloud also allows an enterprise to ingest data easily and rapidly across different platforms and business pillars – something not easily accomplished through major integration projects. The availability of 3rd party data through the cloud, and access to always available and coordinated data makes cloud an attractive environment for AI adoption.
Start With Smart Data
To be successful in AI, you need a lot of data, but it needs to be the right quality and type of data. After all, bad data leads to bad decisions. Structured, clean data leads to smart decisions. Businesses need structured and clean first-party data, as well as high-quality third-party data (which adds additional context) to drive smarter outcomes.
The addition of data from sources like Oracle Data Cloud – the world’s largest pool of third-party data – enables AI applications to improve accuracy and focus. However, flaws in first party data (company-owned, such as data in CRM/CX, ERP, SCM and HCM systems) can also inhibit the intelligence and effectiveness of AI applications.
Not all data is suitable for machine learning and at Oracle we focus on Smart Data, which entails gaining a precise and comprehensive view of data, which continuously updates in real time and connects across systems. Smart Data provides dynamic signals and flexible classification, as well as being embedded in AI applications which leads to the production of intelligent outputs.
Oracle recently acquired DataFox, whose cloud-based AI data engine provides the most current, precise and expansive set of company-level information and data insights to optimize business decisions.
According to Bastiaan Janmaat, CEO of DataFox, “Data is the single most important thing that determines the value of your AI results. It’s easier to keep bad data out than it is to clean it once it’s in.”
Janmaat recommends that companies protect points of entry, clean existing data, verify data enrichment, maintain ongoing data refresh (static data goes out of date quickly), continuously flag data irregularities, and then connect the cleaned data across systems. It may seem counter intuitive, but there is a significant human element in data enrichment, which leads to more intelligent AI.
It can be hard to know where to start when it comes to deploying AI in your business. Most organizations know that AI is going to be an essential ally in improving productivity and business agility over the next five years and beyond.
Knowing how and where to get started is one of the biggest hurdles to overcome. Start with “the why”, think about “build vs buy”, infuse transparency and go all in with the cloud. Once these steps are in motion, invest in smart data – the lifeblood of any AI solution.
Melissa Boxer is the VP of Adaptive Intelligent Applications at Oracle. Oracle’s Adaptive Intelligent Apps are a series of cloud-based applications with use cases in finance, HR, supply chain, manufacturing, commerce, customer service, marketing, and sales.
About the Author(s)
You May Also Like