When Considering Machine Learning

Whether it is through the lens of the investor or the consumer, expectations will challenge businesses to find new ways of interpreting and learning from available data.

At MOON, we are continually driven to better understand human interactions with technology within our diverse environments. What does the world look like over the next two, three or twenty years? What necessarily goes away? What stays with innovation? What new ideas give promise? These questions are a glimpse into what challenges inspire our own research and development. And while we take interests in many frontiers of human interactions with technology, none have been so profoundly promising toward humanity as the rise of artificial intelligence.

The particular area of interest for our clients is a defining mechanism behind AI, machine learning. From general inquiry to implementation strategies, over the years some commonly themed questions have continued to surface. Here we want to highlight and unpack two of those questions.

What are some examples of industries and businesses using Machine Learning and what kind of success have they seen?

This two pronged question surfaces shortly after some introduction into machine learning has occurred. It is a natural step towards better understanding the practicality, performance and expectations; the beginnings of framing it as a possible solution. At the time of this article’s publication there are two recent examples within Healthcare and Food Service & Retail industries that are yielding fantastic results and we will see a common model crediting their success.

Mercy Health
The healthcare sector is ranked among the highest for opportunities of improvement. And one will find no shortage of articles exploring the potential machine learning has to offer in healthcare. Mercy Health, is however, already putting that exploration into practice… with impressive results.

Inspired by their successes of applying machine learning to supply chain management, Mercy Health carried their practice toward defining integrated care pathways with hopes to better deliver and manage the quality of care. Experience, access to large data sets and leveraging the robust AI platform, Ayasdi Care, Mercy Health made the strategic decision to focus on specific areas of care. Among the criteria, frequency of the procedures and area for improvements.

From this application came new pathway discoveries, such as a previously unknown set of best practices for knee surgeries. Armed with learnings from these initial implementations, Mercy Health began to broaden and define more integrated care pathways. The organization is now reducing patient length of stay, standardizing care and helping physicians make evidence-based decisions. From a financial perspective, Mercy Health reported saving $14 million its first year, with estimations to save $100 million over the next 3 years.

Ocado
Another industry ripe for potential and opportunity to improve their client’s lives is retail. Ocado, an online supermarket with no physical storefront, continues to expand its utilization of machine learning. Ocado’s challenges have similarities with other markets, requiring accuracy and timeliness with their products and delivery services, as well as maintaining quality communication with their customer base.

Being exclusively online, with a growing customer base and the variability of disruptive elements such as weather and holidays, Ocado first put emphasis behind understanding and optimizing communication practices. Receiving thousands of emails each day, peaking over 5,000, the volume and variations of communication necessitates an organized prioritization. For example, scoring the importance of positive feedback (lower priority) as compared to time sensitive request (higher priority). Using machine learning with natural language processing, Ocado has been able to inspect and prioritize their customer questions and feedback, routing them appropriately.

Ocado has continued to expand their machine learning into delivery routes, tackling the age-old ‘traveling salesman problem’ for distribution efficiencies and began development of a vision system with Google’s Tensorflow- an AI framework allowing the user to user their own data- to replace the resource requirements that barcodes impose.

Though unique in challenges and implementations, the common discipline between Mercy Health and Ocado has been to embrace and appropriately tackle specific, potentially most beneficial, areas of their business. This focused, gradual introduction allowed these two organizations to learn a great deal before scaling into unrealistic expectations.

What should I be doing at an organizational level to take advantage of Machine Learning and what obstacles might we face?

When an organization progresses from initial discovery into the preliminary framing of how machine learning potentially integrates into their existing ecosystem, we encourage additional thought be put into company culture and business strategy. Let’s unpack a few of these points.

Company Culture
In one form or another, many of our clients are familiar with introducing a new tool, the activation process and all that entails. Signs of a well thought out activation plan includes pathways for employees to become educated. This tends to reduce speculation, uncertainty and unrealistic expectations. Provided the bountiful nature of speculation and uncertainty already surrounding machine learning, we strongly recommend the total company culture be one to include education aimed at embracing change. For most, this looks like a summarization of the program, addressing where the business hopes to improve and what impacts to expect. But for some this can be a radical challenge to presuppositions.

The gap between marketing, business, sales and engineering teams has been a persistent struggle for many organizations. The advice to have your teams working together is nothing new. However, the requirements of technologist and strategist working together here is not just a loose recommendation to run more effectively. In the practice of machine learning, these teams working and contributing to one another is unavoidable.

The importance of organizational education is to shore up misconceptions, a preemptive strike on speculation and uncertainty. The elephant in the room here, of course, is automation replacing jobs. Yes, this is the current case for a few jobs. For example, if your job is to scan or position a barcode on the conveyer belt at Ocado, the vision system in development will replace that job. However, for the majority, automation entails replacing tasks. Thereby, redefining the job while providing or inferring new discoveries to research. The lively debate of what impacts on organizations, economy and humanity is not something we should gloss over or summarize. However, it is certainly out of the scope of this article. For more reading on the topics, I recommend authors like Calum Chace and Jerry Kaplan.

Focus
It should come as no surprise when I say frequent, small victories motivate the continuation and expansion of programs. This concept should not escape the planning sessions and setting of expectations for a machine learning program. A lack of focus on specific areas of your business or focusing on areas with too many variables can quickly lead a team into overly complex and unachievable implementations.

Focusing and initially setting realistic expectations allows for the iterative production of answers to fundamental questions such as, “Was this the correct area of business to tackle? Did we use the right models? Do we understand the business challenges?

Definitions
Before a team begins to engage in creating models, selecting a platform or deciding if customized algorithms are necessary, the organization should have very clear definitions to form a strategy. Definitions will come from answering questions such as the following:

• What area would most benefit from analysis and objective predictions?
• How will your team ascertain that resulting prediction models are objective?
• Do we have access to the quality of data required?
• What new insights are we expecting to gain and what do we plan to do with these?

Additionally, defining strategy will continually drive towards the establishment of quality baselines to better define and execute future efforts.

Expectation of continuation
Because the concept of machine learning as a piece of software running in the background is often conflated with other software applications, the expectations of how much continued work is necessary is often underestimated. An organization should be prepared to continue analysis and development efforts after initial launch. The nature of your industry, how customized your implementation is (algorithms, etc..) and discoveries you’re making will impact the frequency of updates to elements such as models & data and the subsequent testing.

The point here is to never expect a ‘cruise control’ setting for your organization when adopting and planning for the practice of machine learning.

Concluding advice: Diligence with a hint of urgency.

As eluded to in the subtitle, organizations will be challenged by expectations to leverage data, influencing pivotal decisions. Each industry and organizations within, must determine how they will begin to roll out the practices of machine learning.

I hope to have provided or reinforced the idea that machine learning is of great potential, but not without investment in education, planning and commitment from all disciplines within an organization.