An Executive's Primer On Machine Learning
Machine learning is no longer confined to the realms of science fiction. It’s an integral part of everyday life. It’s the reason Google can deliver scarily accurate search results, Facebook’s ads are far more appealing to you than they used to be, and your emails aren’t full of spam.
As the field continues to develop, businesses that want to stay competitive need to keep up with both the potential and limitations of this tech. In this post, we’ll take a look at what machine learning is, how it affects you, and examine different use cases.
A brief introduction to machine learning
Machine learning is the idea that a machine can make logical decisions and have interactions based on previous experiences in a way that mimics human decisions.
Let’s backtrack a little and take a look at how most computer programs work. A developer creates a set of rules that tell the machine what to do in a specific situation. It’s a lot like a series of complex behavioral guidelines that preempt various possibilities. The problem with this is that the program relies on the developer’s ability to accurately predict a series of future situations. This limits the potential outcomes.
Just think of the current generation of commonly used chat-bots like Slack’s ‘slackbot’. This bot is designed with a specific purpose in mind and responds to set inquiries.
While this is useful, it’s no longer immersive enough. This type of bot is completely reliant on the development team’s ability to predict what kind of questions the user will ask and how they’ll ask them. They are usually designed to only handle the most basic cases and scenarios. Once something more complex happens, the system falls back to the human customer support team who understand the larger context of the interaction. As a result, these type of bots have failed to scale because they can’t really meet today’s needs.
They can still work relatively well if you know your audience. When you ask Siri who her favorite Game of Thrones character is, her responses are remarkably on point. It makes her seem larger than tech and this trend is seeping into other human-on-tech interactions. Consumers are starting to expect seamlessness and want tech that understands them. To create these experiences, we need access to a lot more insights. Luckily, machine learning can help us get them.
Machine learning makes in-depth encounters possible by dispensing with the rules. Instead of creating a set of predefined behavioral rules, machines are given sets of data and use that data to drive future decision making in the same way we use past experience to inform what we do next. The more data a machine has access to and the more decisions it makes, the smarter it becomes.
Machine learning comes in three flavors: supervised learning, unsupervised learning and reinforcement learning.
1. Supervised learning
Supervised learning is a lot like learning how to talk for the first time. Our parents spent a lot of time pointing out various objects in different settings and telling us what they were over and over again. That’s how we went from using a few words to describe every person or object we came in contact with, to developing complex linguistic skills.
Machines learn in a similar way, but instead of incredibly patient adults, they use labeled data. Let’s say you want to teach a machine to recognize different animal types. You’ll feed it thousands of labeled pictures of ducks, honey badgers, platypuses and various species. Each image is labeled so that the machine can learn what they look like in different contexts. Based on these labels, it will be able to develop its own idea of what a platypus is and learn to recognize it.
Facebook’s facial recognition algorithm is the perfect example of this in action. The algorithm had access to thousands of tagged images. By using the tags it was able to develop an understanding of what different people looked like and create its own rules for recognizing them based on that experience.
Just like most of us need access to a dictionary to keep building up our language skills, a machine needs access to fresh data so it can keep improving. If it only has access to a limited amount of data, its interpretations (and their validity) will be limited as well.
The challenge is that most organisations aren’t prepared. While a lot of data has passed through, it hasn’t been tagged and stored in the neat data sets that would allow for rapid adoption. While we’ve got clear use cases for machine learning in industries like insurance, the data to use the technology effectively just isn’t there.
Gartner, a leading research and advisory company, argues that organizations starting to use ML tech need to develop their own internal core competence for preparing and tagging data so they can create unique models that work for their specific needs. This allows companies to take ownership of ongoing model training, increasing the results’ accuracy.
2. Unsupervised learning
If supervised learning is the equivalent of learning how to talk with the help of others around us, unsupervised learning is a lot like finding yourself in a foreign country where no one speaks your language. By listening to others, you gain access to the raw data but you have no way of interpreting it. However, with enough time, you can start grouping certain sounds and words together and infer meaning.
Machines follow a similar protocol. When given access to unlabeled data, machines work to discover patterns and group information together. However, because all this is done without a guiding hand, understanding the context is harder. Still, data clustering has really interesting applications in marketing. Marketers can study the characteristics of different groups and use that to create extremely targeted ads and supporting materials.
3. Reinforcement learning
Reinforcement learning is the idea of learning through trial and error. A machine uses past actions that brought success to determine future actions that are likely to succeed.
This type of machine learning is used to teach machines how to play complex strategy games like Go and Chess. Go is a profoundly complex game with more potential moves than the numbers of atoms in the universe. Yet AlphaGo, a computer using machine learning to master the game, was able to beat a world champion by using reinforcement learning. In every game it played, AlphaGo learned from both its moves and those of its opponent and applied that to the next game until it reached mastery.
Search is another major application of this type of machine learning. While Google is pretty tight lipped about how its search algorithm works, we know that it uses reinforcement learning on some level. It learns about ranking based on click through rates, time spent on the page, bounce rates and other factors. It then uses that information to make decisions about a page’s relevance and ranking. This process isn’t static. As users interact with different search results and provide additional information, Google uses that to reinforce machine learning.
Why machine learning matters
Machine learning allows us to use data to understand patterns, interpret behaviors and make decisions in a way pre-programmed algorithms can’t.
The digital revolution brought heaps of data with it. We have access to more raw information than ever before. Knowledge is growing so quickly that even your best people can’t process it fast enough.
Machine learning allows us to gain insights from the data and use them to improve every facet of life. Its usefulness goes beyond knowledge gathering. Machine learning can help automate repetitive and low value tasks and thus decrease costs across the board. At the same time, it allows your team members to focus on higher value tasks so that you can create more and increase productivity and earnings giving you the best of both worlds.
Let’s take a look at some recent uses of ML tech that have had a profound impact.
Machine learning in practice
Machine learning plays a role in many everyday tasks, from your inbox and movie choices, to the way data is shared in companies.
Streaming giant Netflix uses machine learning to power its recommendation engine and keep users glued to the screen. By studying user behavior on the platform, the Netflix algorithm can make informed guesses about what you want to see next. And the more information we give it, the better it gets.
Of course, machine learning has powerful applications beyond entertainment. Voice recognition software Dragon Speech, uses deep learning to transcribe and recognize human speech in various forms and pronunciations with an increasingly low margin of error. This is a huge step toward using natural language in all sorts of instances, from talking to virtual assistants like Siri all the way to transcription.
Machine learning can go beyond the customer facing side and make it easier for your team to work internally. That’s exactly what Uber’s internal ML platform Michelangelo does. Michelangelo is a machine learning as a service platform that “enables internal teams to seamlessly build, deploy, and operate machine learning solutions at Uber’s scale.”
Instead of building separate algorithms across departments, Michelangelo gives the entire team a common platform to use and build on. This doesn’t just save time- it democratizes the use of data and puts cross-department teams on the even footing. It’s an effective way to break information silos and help elevate every team.
Yet while some teams have jumped in and embraced ML tech, hundreds of large organisations are struggling to create successful strategies. There are a lot of reasons for this and before your team can really jump into ML, they need to be addressed. Let’s take a look at some of the major issues.
- Siloed data: Keeping data sets in silos doesn’t just limit accessibility- it decreases awareness so that teams have no idea what they can work with.
- Limited cloud storage: Storing the majority of data on cloud services allows for better access and decreases silos. However, a lot of companies are still in the process of shifting to the cloud making it harder to implement ML on a large scale.
- Unclear ownership: Who is in charge of data processing at your company? The CMO, CDO, Head of Analytics, CTO, CIO or someone completely different? Because big data is still a relatively recent development, many organizations aren’t sure who’s meant to handle it but without ownership, there can’t be effective and strategic leadership.
- Unclear use cases: Many large organizations aren’t certain about how ML can really contribute to their business. While machine learning is used in industries from retail to insurance, we need a better way to share these use cases.
- Internal resistance towards automation: While tech has opened a lot of doors, it’s changed the way things are. Many employees are worried about automation and the impact it would have on job security. Executives need to meet this head on and talk about it, discussing what automation will look like and assuaging fears.
To make machine learning really viable for businesses and organizations, the major challenges need to be addressed before we can make significant headway.
What machine learning can do
We are just scratching the surface of what’s possible. However, there are very real limits to what machine learning can and can’t do at the moment and we can’t make present day business decisions based on conjecture and possibilities. So what can machine learning actually help you do?
Understand your customers better
Most companies are swimming in customer behavior data. We know how long users spend on a web page, how they move around our site, how they use products and how they consume content. Pixel tracking and heat maps make it possible to really get to know your customers’ behavior.
Machine learning can help us process this data and use it to gain valuable insights that help businesses create better experiences. For example, if you know that an unusual number of customers clicked on a specific style of CTAs across your site, you can use that to create even stronger calls to action.
Studying behavior data from thousands of consumers also allows businesses to segment people into different groups and create hyper-personalized experiences. This is especially effective for B2C businesses. Instead of hitting everyone with the same general offer, we can micro-target niche groups with offerings specifically designed for them. This can simultaneously decrease customer acquisition costs and increase sales.
The insights we can gain from machine learning go beyond better marketing and can touch every aspect of the customer experience. This data can inform product decisions and help business create more of what people really want.
However, in order to use these insights, we need to give the algorithm context about the data. Feeding it behavioral data from the holiday shopping period will not give it insight into customer behavior for the rest of the year. Factors like holidays, special promotions, seasonality and geolocation have to be accounted for.
Plenty of smart SAAS products that address these issues exist already. The biggest challenge is that the upfront integration cost can dissuade companies from adopting these platforms without proof of concept. However, Gartner highlights that the majority of POC’s in this space never move to production creating a very high attrition rate.
Alternatively, it’s possible to hand all your data over to an internal analytics or data science team and ask them to build equivalent products. The challenge is that deploying a model to production is very different than creating it. It requires a lot of communication between the IT, DS and architecture teams which is tough especially because of the huge effort over recent years to outsource core IT functions to external parties.
It’s a Gordian knot of a problem that, nonetheless, needs a solution. The executives within each organization have to decide what makes the most sense in their case. Putting off such a crucial decision for too long can have devastating effects on your ability to use customer data to increase customer understanding, sales and productivity.
Better decision making
Relevant information can help businesses make informed decisions by studying patterns on a much larger scale. Machine learning algorithms can handle thousands upon thousands of data points and create various potential interpretations.
IBM’s CEO Virginia Rometty is a firm believer in the power of data to affect decision making. In a recent interview with HBR, she said “Our belief is that you’ll make better decisions if you can unlock that data and that there’s a $2 trillion market around better decision making.”
Machine learning can be particularly powerful in helping decision making across global corporations. Leadership and behavior styles differ around the world and what’s right in one location may not work in another. Democratized access to data can be used to justify and support different approaches that are uniquely suited to their environment.
There are a lot of potential areas of growth in this field. Smart recommendation engines can rank tasks in real time based on multidimensional data modelling and seamlessly deliver them via modern comms interfaces such as Slack or into core operational systems like CRM and ERP.
We’ve personally seen this in the e-commerce and insurance space.
Efficiently obtain, learn from and share knowledge
IBM’s Watson is one of the top of the line examples of machine learning available at the moment and it’s made knowledge sharing possible at a whole new level. According to Rometty, Watson can handle speech, visual data and audio data. It’s had access to information from some of the brightest minds in medicine, finance and multiple other fields and can access it in seconds.
This training makes it an incredibly useful partner for processionals around the world. Humans, even the smartest among us, can’t retain the level of information that a machine can. Having instant access to this kind of insight can catapult companies forward.
A high level of insight can help businesses in every industry really learn from past performances at a faster pace. Sports data is being used to help performers understand the opposition and make smarter, informed decisions. It’s used to help with recruiting fresh talent as well as improving mental and physical performance. Access to the metrics behind winning performances allows reverse engineering success and failure on a large scale.
It’s not just football teams that are using tech to improve performance. Banks are also exploring the potential of machine learning. US banking giant JPMorgan Chase designed COiN to analyze legal documents and obtain the most crucial data points. While doing this manually takes approximately 360,000 hours a year, COiN can review them in seconds. Even though this is tech still at a relatively early stage of development, applications like this can increase efficiency and allow employees to focus on innovation and spending more time with individual customers thus providing a higher quality service and increasing brand loyalty.
Beyond our current capabilities
Machine learning isn’t the answer to life or business. While it can be an incredible asset, it’s not magic. In can only help you solve problems that are clearly defined. It’s a tool that can help support clear goals and a strategy but it can’t help you create one.
Right now, machines need clean data sets and context, and they need us to provide it. In most cases, this means we still need a data scientist to structure the data correctly so the machine can understand it.
Data isn’t pure and a lot of it carries biases. There’s a reason scientists use several different data sets to account for various effects and only test one thing at a time. The way data was collected affects the results. The time period, the location and the method used to collect it all matters. So before we can feed them into an algorithm, anomalies and errors need to be removed. At the moment, we still need smart humans that understand the context to do this.
At the the same time, we need to counteract the bias that’s introduced through the data scientist. We all carry our own biases and can pass those to the numbers through the way they are selected and organized. Until we perfect unstructured learning or teach machines enough so that they can discover their own context, companies need to be aware of and create systems to protect data from these biases.
How to apply machine learning to business problems
Machines can crunch data much faster than humans. They can make interpretations and discover pattern but ultimately, at least at the moment, humans need to be the end decision makers.
That’s why as with every new technology or the adoption of a new plan or strategy, machine learning needs to be linked to clear goals and ROI. You need a plan about how you are going to use machine learning to achieve ultra specific goals.
1. Start with a clearly defined problem
We need to be clear about what we want to achieve in order to get there. Machine learning is best applied to complex problems that need a lot of data and can’t be solved with traditional if/then statements.
2. Test it on non-essential processes first
Machine learning is a work in progress. Just like a high school physics major is unlikely to be able to complete a specific job to the same standards as a PhD student, a fledgling algorithm isn’t going to suddenly know how to recognize complex patterns. It needs time to study and it needs as much clean, contextualised data as you can give it.
That’s why we need to start using machine learning in situations where a margin of error is allowed. Facebook didn’t launch their facial recognition algorithm straight away. It was given time away from the public eye so it could learn and improve. Even then, it took years to take the margin of error down significantly. Putting too much pressure on early attempts will only lead to frustration and isn’t the best use of your capital.
However, while the process shouldn’t be the be all of your company it does need to be a real problem that deserves the investment. You’ve got to give the algorithm data that’s relevant to the problem. This means providing plenty of context and fresh data. Historic data sets that are irrelevant to the problem at hand won’t help you understand what’s going on and will lead to inaccuracy in interpretation.
3. Set clear, trackable goals
Just like with everything else you do in your business, setting clear goals for machine learning is absolutely essential. Decide which part of your business this project will help, whether that’s increasing revenue, improving retention rates or something completely different. Next, break it down into smaller measurable goals and decide how they will be tracked and what counts as success.
Machine learning isn’t a one size fits all magical solution. You must set clear measurable goals that you can share with your team, with the rest of the C-suite and with investors.
4. Create room for growth
Machine learning is a process that should grow with your organization. Google’s search algorithm took years of development to get to its current level of accuracy and it still has plenty of room to grow.
While machine learning doesn’t have to be a foundational building block to be a viable move for your business, it does need to have clear uses. It needs to make all the investment you are going to put into it worthwhile. This means having enough resources to get started and projected resources to help you reach your end goal.
Using machine learning in your company
Machine learning is changing the way we use data. As the world becomes increasingly digital, we are going to see more and more data become available. According to recent data from IBM, the world generates 2.5 quintillion bytes of data every single day.
Businesses of all shapes and sizes, from various industries can benefit from this data. The insights we gather can help create better products and target niche markets leading to more business.
Executives play a key role in embracing machine learning and making it a part of the company. New technology carries risks and fears of getting displaced. Team members worry about getting replaced by AI and there’s general resistance every time we introduce something new.
This slight anxiety towards change is completely understandable, yet it must be managed. Executives play a key role in helping their teams embrace machine learning and see its potential in making the company stronger as a while.
Machines aren’t here to replace knowledge workers, just like computers have not replaced talented office managers and other skilled workers. Instead, machine learning can support your team by giving them access to better information. This means higher level of customer service, a deeper understanding of the customer, of the product, and more access to crucial info to give your business a competitive edge.
Embracing machine learning is a smart move if you make sure that you take it armed with a solid plan and your eyes wide open.