There’s no two ways about it. Artificial intelligence (AI) is the wave of the future and one of the hottest areas of software development today. Companies recognize the competitive advantage it gives them: the ability to generate greater efficiencies, provide better customer service, and predict outcomes that can lead to better business decisions.
In the last decade, the availability of tools to aggregate vast amount of data was the hot thing. Now, this proliferation of data is the biggest reason that AI has taken off. AI gives us a way to do something with all that data. Today, analyzing raw data has become important to understand customer behavior, trends, and future outcomes.
Traditionally, companies would custom-develop solutions in-house. But eventually, big providers realized that they could meet a broad need by mass-marketing solutions that would be cheaper, faster, and more readily available than customized ones. This is how off-the-shelf solutions took off, and has since sparked huge debates about whether to build or buy.
So, given this historical trajectory, as AI becomes more mainstream, will users be able to pick up their favorite app right off the shelf? Unfortunately, not—that’s just not the way that AI works.
AI, at its very essence, is dependent on huge amounts of custom data running through algorithms to find patterns that recognize images or predict outcomes and behaviors.
AIaaS takes hold
Recently, though, some big companies and tons of start-ups have started offering artificial intelligence as a service (AIaaS), which make AI algorithms available to the public, letting them create powerful new cloud-based AI applications fast, without building the tools, infrastructure, or expertise in-house.
AIaaS offerings implement APIs for a variety of operations, such as data sources, statistics, models, and aliases. The programmatic interfaces guide nonexpert users to start using machine learning best practices relatively quickly, without the users having to expend a lot of time and effort on fine-tuning models, or on learning AI techniques. They claim that professionals can customize the parameters or settings for specific machine learning tasks, such as data input, processing, model building, execution, etc.
Amazon offers machine learning services through its Amazon Web Services (AWS). Google offers a number of homegrown AI capabilities, such as predictive analytics, speech recognition, translation, and image content identification. Microsoft offers the Distributed Machine Learning Toolkit so users can run multiple and varied machine learning applications simultaneously. And IBM’s Watson Developer Cloud lets developers incorporate Watson intelligence in their apps and provides its Watson AI engine as an analytics cloud service.
Most of these AIaaS offerings provide a preconfigured infrastructure and pretrained algorithms, which they claim reduce development time, as well as internal resources that are needed to train algorithms.
While there is certainly an advantage to having ready-made tools that can help companies get a leg up on AI implementation, I’m not convinced this can be effective for many companies because the data—as well as the algorithms—needs to be specific to a company to get the best outcomes. And the more unique the data, the more accurate the algorithms will be.
Each business has different needs, infrastructure, processes, and customers; and each different business is unique and can’t fit into a one-size-fits-all approach to AI.
For example, if a bank wants to determine the type of customer that will default on a loan, the training data to create the algorithms must be unique to that bank because its customer profiles are unique. The same applies in retail: How can a company apply the same reasoning to the buying habits of generic customers when it doesn’t consider their customers generic, but unique individuals?
While AI is fundamentally data-hungry and strong AI algorithms like lots of it, it can’t be just any old data. The data needs to be specific to the organization and its data sources.
There still will be a place for off-the-shelf—just not in AI
Despite the prevalence of AI, off-the-shelf software for most business functions, such as customer relationship management (CRM), marketing automation, and cybersecurity, will continue to be needed, but they will be integrated with cognitive capabilities (that is, AI). AI brings cognitive capabilities to software. For example, a core system, such as an enterprise resource planner (ERP), may be running several apps at a bank, and integrate through an application program interface (API) with an AI-based fraud-detection module.
There are still many great off-the-shelf applications for a variety of tasks and by using them companies can get up and running with best-of breed solutions more quickly and at lower costs. But just don’t expect them to be smart—that comes from your unique data and how you train AI algorithms.
AI is clearly making inroads into businesses of all types. Yet unlike other aspects of the computer industry, there’s no quick fix to getting there. It still requires customization, rigorous training, and on-the-job experience to deliver real value.