The foundations of machine learning in mid-sized organisations
- 28 August, 2018 06:30
Learn from others but avoid following competitors
As global firms with deep pockets invest in machine learning and launch new AI-backed products, services and processes, how should mid-sized organisations with less time and technical expertise respond?
Up to now most have chosen to wait and see. But many who have invested have achieved results which show that across industries and company sizes machine learning solutions can slice cost from internal business processes and inspire new product and service possibilities.
We’ve seen voice and image recognition come of age, mapping platforms identify traffic delays and recommend new routes in near real-time and e-commerce platforms find and cluster similar products helping customer find what they need and what they didn’t know they needed. For mid-sized organisations, machine learning can lower costs or opens up new markets by offsetting the scale advantages available to larger organisations, and much more.
Software-based prediction and classification are not new concepts. They have been worked on for as long as personal computers have been around, however the last five years has seen a combination of factors enabling much quicker and simpler solutions to emerge. The drivers are expanded use of sensors to capture data, easier access to external data sources, greater use of scalable cloud computing and new frameworks and platforms which perform complex calculations while keeping the development workflow simple. Together this progress has made the creation of machine learnt solutions quicker and cheaper.
Data is the lifeblood of machine learning solutions
Despite the progress made, it’s not a simple path to navigate. Many of the tools available are far from mature, people with experience are scarce and determining which data is needed and available is difficult. But these challenges shouldn’t be a deterrent, by the time they’re fixed many of the most valuable opportunities will have been explored, and firms will be well on their way to launch solutions and change market expectations. Firms that begin now with a solid approach and learn from those who have trod the path so far will have the opportunity to shape their market rather than play catch-up later.
As you begin to explore machine learning possibilities, follow these five points to achieve meaningful results quickly:
Get to know your data
Data is the lifeblood of machine learning solutions. Organisations that manage financial assets, physical devices and inventory well often lack any data asset catalogue and have no visibility of the value potential of their data assets. It’s important to get an understanding early on as to what data is currently captured, what the condition of that data is, where there are gaps and what external data is available. In the past, data was treated as a passive participant in business processes, something created and used within applications. Data smart organisations have learnt that its value potential increases when combined across applications or when augmented by external data sets. Likewise, data from one business process or organisation can have predictive power in optimising processes elsewhere. Understanding data relationships across the business value chain is powerful.
Identify a small number of high potential machine learning concepts
Find specific concepts where machine learning can provide valuable input to a existing business process or help in creating a new solution. By focusing on specific business problems you’ll ensure effort to test concepts and learn technology is closely aligned to where there is highest value potential. Avoid progressing too far playing with tools and learning technical topics without a solid prototype concept. There are many tools available with different approach and possibilities. It’s easy to become tool-focused and lose sight of the destination so ensure that the goal is clear from the outset.
Understand the types of problems machine learning addresses well
Identifying high potential use cases is difficult as lack of familiarity with good examples results in many ideas being suggested that don’t fit the strengths of machine learning. Study past concepts from similar and diverse industries, involve individuals from across the organisation with awareness of current issues and what is valuable to customers, document current and possible data assets, study the relationships between data assets and jumpstart progress by involving people from outside the organisation with more experience in machine learning concept ideation and to fill technical skill gaps.
While growing experience, aim for a series of small successes
When evaluating high potential machine learning concepts early on, it is best to achieve a series of small successes while learning and gaining experience. It’s usually best to favour internal process streamlining over creation of new customer products and services which will often require greater development rigour. As experience is gained, more ambitious ideas can be attempted. Several common concepts that works well early on are:
Streamline a task to classify free form questions or issues into standard question and answer pairs. This can be used to help customer support staff or later as it matures provided directly to customers as part of a chat bot.
Where many factors may lead to a complaint or customer loss, create a score to identify high frustration risk customers. Later, this work can be extended to identify which factors have the highest link to frustration and through addressing the issues be used to improve customer experience.
Positive outcomes can be predicted such as identifying customers with higher likelihood to spend, which marketing levers are most effective, or identify and recommend complementary products more likely to be appreciated by customers.
Learn from others but avoid following competitors
Copying what others have created or taking a generic capability benchmark and improve approach dulls differentiation. Focus instead on what your current and future customers value, and on what aspects of your organisation are unique and possible to be leveraged. If you have individuals or teams that shine, look at how this can be scaled further through extending their reach with better insights and efficiency. Running behind competitors to solve customer problems that have already been met results in generic products leaving only cheaper prices left to attract customers.
In some industries, machine learning will have small benefits, such as enabling a minor increase in productivity enabling workers to achieve higher output with less, or reduced excess inventory through more accurate prediction of demand. However, in other industries the impact will be more intense—we’ll see the development new products and services which become expected by customers and step-changes in cost efficiencies that erode the margins of slow movers. In markets that experience bigger changes incumbent organisations will struggle to stay profitable and compete.
It takes time to develop a data-driven culture and build machine learning capabilities. Mid-size organisations have a lot going on and responding to customer requests, keeping the business running and coordinating suppliers and production is demanding. Be that as it may, the possibilities presented to mid-sized organisations by machine learning are significant; it’s worth watching, exploring and for many companies pursuing solutions to deliver greater product differentiation and customer value.
Derek Papesch is an independent data consultant experienced in planning and executing data initiatives to achieve organisational results. He has wide industry experience in New Zealand and overseas with a focus on rapid industry change and driving market growth. To discuss or seek assistance with any of the topics in this article, contact via Linked In.
Get the latest on digital transformation: Sign up for CIO newsletters for regular updates on CIO news, career tips, views and events. Follow CIO New Zealand on Twitter:@cio_nz