Does your business have a machine learning problem or a data problem?
May 22, 2020
Whether you’re in healthcare, retail, or the technology industry, artificial intelligence and machine learning are being touted as futuristic solutions for problems facing your business.
When thoughtfully deployed, these technologies can generate crucial, time-saving results. But your organization must look beyond the hype to determine whether these advances suit your needs.
Though AI and machine learning are often misunderstood, their core applications are built upon accessing and analyzing data. Before knowing for sure whether this technology suits your business needs, the key priorities are to recognize what AI and machine learning are, how they can be deployed, and, most importantly, whether the information side of your organization is prepared.
AI and machine learning: What’s the difference?
“Artificial intelligence” sounds like science fiction, but in reality AI is far less magical than its pop culture counterparts. In a broad sense, AI uses algorithms to instruct machines to complete a task in a way that mimics the logic of humans. Often misunderstood as a standalone system, AI is actually programming that can be implemented and applied to a system.
An advanced subset of AI, machine learning describes when a computer receives a large set of data and uses statistical probabilities to draw conclusions and alter those algorithms. While all machine learning involves AI, not all AI involves machine learning.
From a business perspective, when most people talk about AI they mean machine learning. But for all its promise, machine learning has very specific use-cases where it can be effectively applied.
In e-commerce, machine learning is often applied to analyze online shopping habits. As you browse for one product, machine learning analyzes your choices to surface similar products and those that have been often purchased with them. In this instance, so much data is being accessed and acted upon from multiple sources that the process must be operationalized.
Complex computational challenges like these are well suited for machine learning. But there are significant steps that must be taken beforehand to harness this technology.
Before exploring machine learning, check your data
While a machine learning system can look like a lifesaver in TV commercials, there’s a considerable amount of legwork involved before it can be implemented. Prior to evaluating whether you have a business need suitable for machine learning, you must ensure your dataset is organized, centralized, and governed. If you can’t trust your data, then you won’t be able trust any conclusions that may be drawn from it.
Is Machine Learning a Good Fit for Your Business?
Is someone in your organization asking about AI and machine learning? Before the conversation goes any further, answer these questions.
The majority of this effort is data engineering, which in most cases will offer more bang for its buck to your organization. By and large, businesses lack a single, centralized data source and the capability for descriptive lookback reporting, which allows an organization to analyze its data across a given date range. Plus, with data spread across different silos, non-technical people in the organization cannot access or analyze key business information. When your data is organized and easy to navigate, multiple teams can then answer their own questions to allow for self-serve analytics.
In the e-commerce example, generating product recommendations requires streamlined access to sales data, inventory data, and customer data to create a model useful to machine learning. Without the data engineering required to organize these multiple sources under a single umbrella, companies can struggle to quickly generate even basic information, such as sales figures from a single day.
Data engineering reconciles and standardizes these datasets into a single pipeline in an automated and reliable way while ensuring it has semantic meaning to the business. In this way, data transforms from endless spreadsheets with cryptic rows and columns to something easy to understand and explore across your organization. Consequently, questions about your business that were once hard to answer are made easy.
Structuring for these capabilities is complex, time-consuming work that must take place before machine learning can enter the conversation. If that sort of technology suits your needs, your business will have covered 80% of the ground required for outside experts to implement machine learning.
But even without exploring machine learning, your business will find an immediate return on investment from data engineering projects like these.
Machine learning suffers from a ‘black box’ problem
Machine learning can generate useful conclusions from large amounts of data, but the reasoning can be difficult to understand and explain. Statistical probabilities may be revealed, but those will offer little insight about your business or its customers.
For example, credit card companies use machine learning to guard against fraud detection. Machine learning will flag a specific transaction based on statistical probabilities. However, it may be difficult to understand on a human level the justification for that alert.
Machine learning models offer benefits in specific and often niche circumstances but often cannot help you gain broader insight into you. Machine learning models are often like a black box where data is input and predications are output, but the why and how these conclusions were reached can be difficult to explain and understand.
Streamlined data opens up new possibilities
Machine learning is an eye-catching buzzword, but it does not provide an out-of-the-box solution for every problem across your business. Whether it’s suitable for your needs depends on your goals. Are you looking to explore your data to reach an ambiguous conclusion? Then traditional analytics may be enough. If you’re looking to leverage large datasets in an operational use case with a clearly defined outcome, then machine learning may help. But in many cases, there are other solutions as well. With your data on a solid foundation, new possibilities become available.
Imagine that a company is running automatic ads for its product depending on search engine input. If the product is out of stock, the company hopes to pull these ads from circulation automatically. If the company’s data is centralized, a programmer can set a few automated rules and provide a cost-effective solution. With reliable and centralized data, more traditional data engineering and analytics can resolve a problem that may have initially seemed suited to machine learning.
Take that, science fiction.