Spearheading a machine learning initiative, however, is often not a simple or quick process for several reasons. On the business side, it can be difficult to convince non-technical business executives of the measurable value of a new technology like machine learning (ML). On the tech side, it can be expensive—and in some cases infeasible—to build a data science team that can deploy ML, due to a shortage of talent in the marketplace. Finally, combining the business knowledge and data expertise of these two groups can be a challenge, given that they speak very different languages. Organizational sources of friction like these explain in part why the adoption of machine learning in the enterprise context has been slower than the hype might suggest.
Here are the four considerations we recommend to enterprises who are just starting their journey into machine learning:
1. Prove value with real-life business cases.
CIOs and IT leaders have the technical knowledge to assess the value of ML, but acquiring budget for an ML project to get off the ground typically requires business buy-in. Forty-seven per cent of CIOs say they struggle to source the budget needed to fund the new skills and technology required. The key to doing so is to spend time with C-level colleagues across the business to find high-value issues that ML can be shown to solve. CIOs and project leads must then create a use case around one of these issues, quantify the expected results, and articulate the measurable business value.
2. Maximise your talent.
The success of an ML project is influenced by the people who can source useful data and make sense of the results and recommendations offered by various ML techniques. Not every organisation has a team of data scientists on hand to lead ML projects, but recently-developed automated machine learning platforms can help here: existing data analysts within the business, who should already have a deep understanding of its data, can often achieve successful outcomes with ML thanks to these tools. Expanding the set of users who can pilot ML projects expands the potential of ML to drive ROI.
3. You need good quality data—and lots of it.
The quality of a machine learninginitiative fundamentally depends on the quality of the data. Training an ML algorithm requires lots of high-quality data, integrated and formatted appropriately, so that the algorithm can ingest the data and return meaningful results. The quality and quantity of data is crucial, which means that fast and accurate data preparation is the foundation for any algorithms success. The preparatory steps—accessing, structuring, cleaning, combining, enriching and delivering metrics and variables to feed a model—can determine the project's success. Intelligent data preparation solutions now exist specifically for this purpose.
4. Take a fail-fast approach when expanding models.
Machine learning is an iterative process. An analyst will typically start with a small set of available data and use that to build a first model with the aim of it answering a specific question: for example, what is our predicted sales volume for next month? Typically this first rough model will not be accurate enough. Improvements can come in a variety of forms: the analyst may need to find additional data sources to train the model, or to they may need to transform the data in different ways to provide the right features for the model to learn from. Within a single project, dozens of iterations can be made until high quality results are produced. To optimise the chances of success, it’s critical to reduce the overall iteration time and adopt a ‘fail-fast’ approach. Given the need to constantly acquire and transform data, the ability to accelerate data preparation and integrate it with a machine learning framework is critical. Agile data preparation allows for faster time to results, and more opportunity to engage with key stakeholders.
Forward-thinking organisations are already using machine learning to their advantage. Hubspot recently acquired ML firm Kemvi to hone its predictive lead scoring tools; Twitter uses ML to automatically curate its content timelines to serve people the best content; while Deutsche Börse Group has been using it to reduce fraud in financial services.
Figures suggest that enterprise investments like these will nearly double over the next three years , reaching 64% adoption by 2020. International Data Corporation has quantified this as a $47bn spending figure.
The race is truly on to take advantage of the business benefits of ML. But there is no reason to get left behind. By proving the value of ML projects, maximising talent, ensuring lots of good quality data and a fail-fast approach, even organizations just starting out on their journey can realize the business potential of this new technology.