Companies that have invested heavily in big data solutions want to know how to make smart, strategic investments that will distinguish them from the competition and enable the best possible return before making the decision to go all in. In the past, not all enterprise big data initiatives went as planned. These failures are not usually published, but the big data failure rate is unusually high.
According to Gartner, only 15% of businesses make it past the pilot stage of these projects. Our fear, as leaders of technology companies, is that with so much attention surrounding AI, the pressure is on to apply the technology or risk falling behind the many decision makers who are adopting technologies without first establishing clear business goals and understanding the differences between AI and ML and how they should be applied.
It’s easy to get caught up in the allure of artificial intelligence as well as its hype, including breakthroughs like deep learning, but those looking to make an outsized impact should instead focus on its more practical counterpart: good old-fashioned machine learning -- or “cheap learning,” as my colleague Ted Dunning and Ellen Friedman explain in their guide Practical Machine Learning: Innovations in Recommendation.
The distinction is simple: Cheap learning is about leveraging basic machine learning techniques on straightforward data sets en masse to generate a large number of small, incremental improvements. Deep learning, on the other hand, is a specific subset of machine learning. Deep learning is a collection of sophisticated and highly intensive machine learning approaches that make business decisions based on highly complex data sets possible.
For tasks that involve analyzing raw data, such as images and voice recordings, deep learning is best. But when it comes to working on simplified, structured types of data, we’ve found cheap machine learning will do the trick. When you consider that the majority of data flowing through enterprises falls into this second category, it’s clear which tool makes the most sense.
As you chart a course forward, here’s what you should be doing today to set your company up for success tomorrow:
Capture More, Better Data
Artificial intelligence is fueled by data. Pick an approach, and you’ll find data at the center. Why? Because large volumes of complete data sets are needed to accurately recognize significant patterns of behavior with people, events or other characterizations, and that’s what AI is all about.
Having access to more data -- especially a range of contributing or related data sources -- is usually an advantage. This is why companies like Google (a leading investor in our company), Amazon, Facebook, Alibaba and Baidu are so powerful from an AI perspective. These companies have enormous data sets that they’ve been capturing for decades across a wide variety of trended patterns. This data has fed into their algorithms for years, making them increasingly more refined, accurate and targeted.
For most enterprise companies, the big challenge is that it’s not always clear at the time data is collected what’s going to matter down the road. This makes it hard to know what to measure today and if that measurement will be valuable in the future. This line of thinking represents the old-school way -- it presumes there is only a finite amount of data one can feasibly capture and store, but that’s no longer the case with the advent of new technologies. Furthermore, the ability to connect this data, at a meta-schema level, allows a completely new perspective on previously unrelatable data sources. In addition, big data has seen its fair share of innovation in recent years with storage becoming increasingly smarter and cheaper.
Establish Clear Business Objectives
Successful machine learning isn’t just about choosing the right tool or algorithm and feeding it tons of data. Context matters. Putting machine learning to work on large data sets will yield little value without clear objective goals guiding the efforts.
Do you know what success looks like today? How about five or 10 years from now? Machine learning can help you get a clear baseline today and empower data scientists and engineers to point it in the right direction based on data visibility that is continuously being reviewed and refined.
There’s a sense that AI techniques like machine learning will offer businesses a magic bullet that turns everything into a smarter, more efficient version of itself. This is wishful thinking. Today, these tools work best in narrow frameworks; in the long term this will not be true, but it’s today’s reality. The more specific the objective, the more effective the tool and the higher likelihood of success. Operationalizing a vast number of simple but powerful techniques can deliver enormous business value with relatively short development times and ease of deployment and maintenance.
The path to real business value is a well-crafted strategy. Once you have a business roadmap with goals and well-defined objectives, the application of AI techniques will make more sense and align with the overall business strategy. There is no worse feeling or decision more career-limiting than using advanced techniques and technologies that are not aligned to your business goals and strategy. These projects are, typically, the most strategic and have the greatest visibility and highest expectations.
Every business wants demonstrated improvements based on hard data to support the results. The bottom line: Use the appropriate technique for the assignment given. Truthfully (and based on our practical experience), deep learning will come in handy and may be the right strategic technological choice. But for most applications in the enterprise, cheap learning will offer a more practical -- and effective -- solution. Don’t be afraid to recognize the difference.
Source: All the above opinions are personal perspective on the basis of information provided by Forbes and contributorTom Fisher