When implementing any strategic initiative, it’s important for organizations to build a considered plan upfront taking in a number of variables. Getting started with machine learning is no different! This principle definitely holds true in machine learning and the fast emerging area of artificial intelligence (AI) too.
Since machine learning in particular is garnering significant attention for its ability to create efficiencies, improve customer experiences, and provide competitive advantage, it’s only natural that many executives are looking for ways to apply it within their businesses.
However, a hasty approach is unlikely to produce the desired results, or it may stretch out and become costly if the right parameters are not in place. Ultimately, without a defined ML and data strategy that has senior executive buy in, potential for success will be limited.
With all the buzz around the topic clouding the issue, it’s understandable that technology and business executives may have trouble discerning what is real and what is hype, let alone figuring out how machine learning might best help their organization. So what’s the way forward?
Identifying the problem
The key is to identify a real business problem. When you are first getting started with machine learning, you want to start small initially, while nurturing the broader vision to build out and scale the right kind of project iteratively from a foundation– learning, adapting, and growing as you progress. What, precisely, that business problem or issue is will depend on every organization.
Start by considering your own most pressing issues. Danny Lange, VP of AI and Machine Learning at Unity Technologies, believes one thought process in particular is helpful in eliciting beneficial machine learning use cases (specifically, asking the question, “If only we knew _____”). With that in mind, is there mission-critical information you are currently lacking?
Business Collective points to the example of solving customer satisfaction issues using machine learning by avoiding a general question (“How can we improve customer satisfaction?”), and instead being more specific: (“What factors are most closely associated with negative customer reviews?”). This example illustrates how being clear on inputs and outputs, can help a predictive engine to produce more accurate results.
Several of the organizations we have assisted are top global technology firms with teams running machine learning projects. Often these teams have begun a project, identifying a problem or an opportunity, before requiring assistance in scaling out the project. Here, the ability of the client team to scale effectively is dependent on a strong machine learning data strategy. Appen has helped advise on this strategy, and then supported it through data collection, annotation, evaluation, and transcription of various data inputs, preparing image, speech, text, and video data for use in the machine learning and AI applications these firms have created.
In the book, Applied Artificial Intelligence: A Handbook For Business Leaders, the authors provide insights into prioritizing the right opportunities, and combining data, technology, design and people on machine learning projects to solve real business problems. If you’re looking for a practical framework and playbook, it’s a great starting point.
It’s important to note that not all machine learning projects need take place in the domain of startups or Internet-generation businesses. Capital One, an established financial institution based in the US, has created a machine learning innovation lab to develop new data strategies and customer service chatbots, looking to progress practical solutions to several identified business problems. Its lab has the twin effect of helping the bank recruit top technology talent.
Finally, there are a number of more visible trends where machine learning has proved its worth. AI market research firm, TechEmergence, has collated a number of well-established business use cases for machine learning that could act as a starting point for zeroing in on the right initial application. These include: face detection, email spam filters, product recommendations for customers, speech recognition, real-time bidding in online advertising, and fraud detection.
Building the right data strategy
Alongside selecting a specific problem to solve when you are getting started with machine learning, it’s also critical to ensure you have the right data strategy in place. That can mean lots of different things to different companies, depending on industry, customer types, internal structures, the degree to which data is structured or unstructured, and so on. Appen has recently published a white paper focused on this topic. Generally speaking, machine learning requires large volumes of training data, and this data must be of the highest quality to ensure that machine learning-based products can interact effectively with humans. Our paper also explores which data sources should be considered – you can download the whitepaper here.
In many instances of machine learning, the key is making sure you have new data, and clean data. While not every application necessitates this principle, new data (as opposed to historical data) is going to be of most importance in rapidly changing fields, for example mobile eCommerce, new delivery apps, and omnichannel customer experience (CX) environments.
The process of cleaning data will be required in many cases, given most organizations store a range of structured data (like spreadsheet data, or simple information from sensors), which is easier to input directly for machine learning purposes, and unstructured data (for example human speech, audio, video, images, emails, text documents, and social media content), which needs the time and resources to be made ready for machine learning.
For machine learning applications, data must be selected, cleaned, and tested – as mentioned starting small for iterative learning, before diving into a full, live production system. The process will often start with data collection and will then move to the cleaning process – detecting and removing errors, inconsistencies, and unwanted noise. From there, companies can proceed into interpreting and analyzing their cleaned data, feeding it into a chosen “business problem” or identified machine learning use case.
Thinking about starting an machine learning project? Appen assists a range of organizations to create and improve products using data for machine learning and AI – visit our solutions page to learn more about the types of machine learning-based solutions we help to improve with our data services.