So you want to get started with AI in your enterprise? Join the club. Just over half of companies say they are investing in AI deployments or pilots, according to recent Forrester Research survey information. What that means may be different for each individual business. Some may choose to invest in data scientists dedicated to machine learning efforts. Others may choose a more tentative step -- looking at the AI-as-a-service options available from vendors.
If you are in that more tentative group, what's the best way for you to get started? What sorts of AI-as-a-service options are available for you? How do you choose your initial project?
The first thing to acknowledge is that AI is really a collection of complementary technologies including machine learning, natural language processing and text analytics, and computer vision. All those capabilities are available via algorithms and APIs from giant as-a-service cloud vendors Amazon, Google, Microsoft, and also from IBM. Smaller vendors may also be offering some configurations of AI-as-a-service options. The public cloud is acting as a first testbed for AI in for many organizations. It can enable you to start small, without making a big commitment.
You may want to make those first tentative steps very carefully. Forrester has said that 55% of organizations have not yet achieved any tangible business outcomes from AI. Gartner is predicting that through 2022, 85% of AI projects will deliver erroneous outcomes due to bias in data, algorithms, or the teams responsible for managing them.
Even if you are starting small, there are some best practices for success. For instance, before you get started, it's a good idea to decide on your first use case, according to Gartner VP and distinguished analyst Bern Elliot, who spoke with InformationWeek in an interview. Enterprise organizations will need an initial project use case that leads to an initial project success, as with all first-time advanced analytics efforts.
"You don't want to have the technology drive your roadmap," Elliot said. "You want your business objectives to drive your roadmap."
Elliot said that organizations should look at existing use cases and consider software-as-a-service and professional services to help them accomplish that first project. Gartner expects many to do just that.
"It's a ratcheting process," Elliot said. "We are still in the early stages for a lot of machine learning implementation." Gartner's survey data for AI in the enterprise is more conservative than Forrester's. Gartner's 2018 CIO survey showed that only 4% of CIOs have implemented AI, but that 46% have developed plans to implement it.
Gartner has further noted that most organizations aren't really well-prepared for implementing AI because they don't have the internal data science skills. They plan to rely on external providers in many cases, which is probably a good idea, considering how far they have to go. The CIO survey also revealed that 53% of organizations rated their own ability to mine and exploit data as "limited."
The right quality data is the crucial ingredient for these early AI projects to succeed and produce value. Whether organizations outsource the projects to cloud providers or not, organizations will need to make sure their data is of good quality. Training algorithms with bad data will lead to bad models and erroneous results.
"Data is the fuel for AI, so organizations need to prepare now to store and manage even larger amounts of data for AI initiatives," said Jim Hare, research VP at Gartner, in a prepared statement. "Relying mostly on external suppliers for these skills is not an ideal long-term solution. Therefore, ensure that early AI projects help transfer knowledge from external experts to your employees, and build up your organization’s in-house capabilities before moving on to large-scale projects."
Structured data to get off the ground
Indeed, external suppliers won't have the specific data that can offer the most value to your business, noted Forrester Research Senior Analyst Brandon Purcell in an interview with InformationWeek. For instance, Amazon's image recognition solutions will not be trained to your organization's specific needs right out of the box.
"If I'm L.L.Bean, and I want to use Watson Computer Vision to recognize my product catalog, well, Watson may not have been trained to recognize comfortable slippers."
That example concerns unstructured data, however, and most organizations will find the biggest value for their initial project if they apply machine learning to structured data.
Purcell said those projects tend to provide the biggest impact. For instance, looking at it from a customer insight perspective, an organization can take historical data on existing customers and then determine who has bought a product or service and who has not bought. Purcell said that projects like that can help companies look for signals in their current customer base and use the models to target customers. These projects deliver measurable positive results.
Meanwhile, startups in the AI-as-a-service space have started to crop up, too. Noodle.ai, for instance, is offering a platform with pre-built applications, five designed to help companies make demand-side decisions and five designed to help companies make supply-side decisions.
CEO Stephen Pratt told InformationWeek in an interview that the offerings from big vendors such as Amazon and Microsoft Azure are designed more like tool kits than applications.
"One of our philosophies is that a lot of companies understand the power of AI and machine learning," he said. "They've seen it in action, in how Amazon itself has used it to make good decisions. They want to do something like that. They want to 'Amazonify' their businesses."