Authored by: Mike Lampa, Julie Burroughs, GDM Advisors and Sharon Allpress, Guest Contributor
The benefits of implementing machine learning solutions range from operational efficiencies to product innovation. The opportunities are endless. We believe machine learning should be a consideration for all analytics product development. As we work with our clients to go from hype to reality we advocate three key building blocks for durable machine learning programs.
BLOCK ONE: MACHINE LEARNING PROGRAM BUILD
As always, proper planning is critical as the cost of rework is exorbitant. Strategic planning provides organizations with the ability to grow and support a citizen data science community. True to all data programs is ensuring the proper resources and strategy for success. The great news is we are seeing the formalization of data science programs in traditional higher education institutions, ML boot camps, and Massive Open Online Course (MOOC) sites such as EdX. Machine learning has also given rise to the importance of the data engineer who’ builds and maintains the data pipelines and data architecture necessary to deploy and operationalize ML products at scale. Unlike the data scientist, formal training for this role is limited, most learn on the job.
Critical to this planning exercise is selecting the right ML strategy and technologies to support your organization’s requirements and goals. There is a vast array of approaches to building a ML program. Options include open source ML libraries, purpose built industry-specific ML alogrithms that provide a jumpstart to ML product development, and low-code / no-code AutoML platforms that use embedded ML algorithms to enable self-service ML product development for the budding citizen data scientist. The route you choose will be driven in part by your use cases, the amount of IP you want to own, the specialty skill sets that are required, and level of vendor lock-in your organization is willing to accept. Because of this we recommend a formal strategic assessment and technology selection process.
Outcomes from this phase should include; Identification of business driven strategic themes, creation of an analytic product roadmap, determination of the ML build vs buy strategy, execution of a technology selection process, definition of reference architecture and engineering guidelines that encompass the full life-cycle of the analytic products that are developed, deployed, operationalized and eventually, decommissioned. And don’t forget the humans with education.
BLOCK TWO: MACHINE LEARNING PORTFOLIO BUILD
There are likely hundreds of ML use cases throughout your organization. Identifying the most impactful and properly prioritizing them for development is key. Based on your company’s strategic themes, follow the Lean-Agile product management practices to continuously explore, quickly test your hypothesis, establish the value expected from the ML products you decide to persevere and prioritize those products for continuous development, integration and deployment such that you are generating the optimal value in the shortest amount of time. This comes with a strong focus on ideation with the business, while developing ROI and risk models. We advocate for cultivating the role of an Analytic Product Manager who works with the data science team to ensure those products and ML and models are developed with a meaningful business value lens. The outcome from this phase is a strong portfolio of prioritized Analytic Products augmented with machine learning that is continuously shared with the organization.
BLOCK THREE: MACHINE LEARNING CENTER OF EXCELLENCE (COE) BUILD
Now that you understand the abundance of opportunities your organization has through the use of machine learning, you will need to develop clear operationalization processes, model and simulate how the business will integrate the ML products into the targeted operational processes and business decision making. You will also need to implement the appropriate ML product performance and efficacy monitoring approaches. Monitoring is critical to ensure the incoming features are not drifting from expectation and that the generated insights are at the expected level of accuracy. When drift occurs and / or accuracy erodes the monitor triggers the Analytic Product Manager, citizen data scientist and Analytic Product development teams to perform root-cause analysis and apply the appropriate ML product tuning. The building block is critical for ensuring the analytic products in the portfolio are continuously generating accurate insights that contribute to value generation.
If incorporating machine learning into your analytics practice seems daunting and is not part of your strategic goals, you should reconsider, as the benefits are abundant. The great news is, program management methods, techniques, solutions and services for machine learning are continuously maturing to help you more quickly deliver incredible value and enable your enterprise to truly differentiate in your competitive markets..