Managing Machine Learning: How to Bridge the Man/Machine Divide
The idea of "learning computers" that become smarter with more time and information used to be confined to the realm of sci-fi or the most cutting-edge technology projects. But fact has caught up with fiction; machine learning is transforming both large and small businesses that want to take advantage of reams of data at their disposal.
Machine learning allows the computation of patterns and outcomes across massive data sets at a scale far greater than traditional analytics could ever achieve. And because the computation is completely independent of manual processing, companies gain the ability to effectively collect and analyze even more data, utilizing the inexpensive storage, computing power and distributed database technologies now available. Simply stated, machine learning begins to present itself as a retailer's best tool in a time where data doubles every two years.
Despite these clear benefits, machine learning still arouses the suspicion of those in organizations who stand to benefit from its adoption. Perhaps the idea of machine learning crunching complex algorithms on an enormous scale lends itself to the belief that the technology spells the end of human analysis in a zero-sum scenario. However, businesses can help employees to more receptively embrace machine learning when guided by the following principles:
1. Honestly assess your company's readiness. Before rushing headlong into machine learning nirvana, determine if your company comfortably accepts new technology in general. The adoption of machine learning necessitates changes in the processes and methods of a company, which may meet with resistance from teams that aren't typically early adopters or are more inclined to adopt technologies after the first movers in their industry have already done so. After analyzing your company's culture, ask yourself if you have a competitor making visible gains with machine learning. If industry rivals are already seeing return on investment from the technology, you're more likely to get a positive response to the idea of machine learning adoption.
Don't underestimate executive support; machine learning necessitates major cultural shifts, so top-level buy-in is crucial. Don't focus on the title of the adopter — CEO, CTO or CMO — the most important criteria for this internal champion is acceptance of the vision and a problem that machine learning will solve.
Determine early on whether your organization even has the capacity to recruit and support data scientists or analysts. These people are the key to translating domain experts’ needs into workable machine learning algorithms. But they're a scarce resource. To entice data scientists on board, your company needs solid support technology and compelling projects. Be prepared to invest heavily in recruitment; competition for talent in data science is at an all-time high.
2. Point out that automation doesn't mean elimination. Machine learning naturally instills the fear of job loss due to automation. While it's understandable for employees to think that a machine might replace them, that's simply not the case. Machine learning adds to the workforce in the same way that adding a new employee would. Roles may shift and some pre-existing processes may need to be altered, but overall focus and productivity increases.
Communicate openly with those team members most directly affected, and demonstrate how machine learning is a tool that will enhance, not obviate, their role. For instance, one database marketing professional we worked with was initially very concerned that the adoption of machine learning would push him out of his job. By the end of the project, he had become a critical domain expert and the resource for all machine learning-related questions within his organization.
3. Build trust in the machine over time. Users will be hesitant to let machines analyze, make decisions and fully implement those decisions right away. Complete acceptance is a tall order, so I recommend certain trust-building techniques in the early stages of machine learning adoption to help manage the change:
- Begin with small steps to build confidence. Identify a tangible, specific business problem that responds to a data-driven strategy. Continually refine the algorithms through trial and error. You'll find that as the successes grow, so will comfort with the technology.
- Set manual checkpoints. Allow employees to both interact with the data and confirm its conclusions. Let the confidence levels in the algorithms guide the direction of your check-ins — e.g., a 95 percent confidence level likely doesn't need human intervention whereas an algorithm with a lower level of confidence may benefit from more frequent checkpoints.
- Analyze the data; trust your instincts. Machine learning provides unexpected insights, but it can be difficult to accept seemingly counterintuitive findings. For example, if your inventory management system automatically recommends ordering hundreds of units of a seasonal product that you expect to drop off in demand, listen to your intuition. Even if the machine may recommend certain actions, a human expert can overrule action. However, be sure to go back and check to see if the demand was there. If the machine was right, you'll have more faith in future recommendations.
- Define key performance indicators. Consistently track how successfully your machine learning algorithms are performing. Tracking performance will not only quantify your successes, but also increase confidence and provide the data necessary to continually improve.
As a machine learns project by project, so will your organization. As trust in the technology grows, machine learning will uncover ways that your company can succeed today and into the future!
Dan Darnell is the vice president of product and marketing at Baynote, a provider of personalized customer experience solutions for cross-channel retailers.
