A Framework For Creating Massive Impact in the Enterprise With Machine Learning
This document outlines our framework at SensAI Group for approaching and executing enterprise machine learning projects.
This has been an effective process for working with companies that have established products, and who are interested in unlocking the value of their unique, domain-specific data to create massive results .
There are two main phases of this process: Strategy and Development. These are further divided into two smaller feedback loops. Strategy is composed of Analyze and Plan, and Development is composed of Implement and Measure.
The Strategy phase is a two way conversation. We work closely with decision makers and stakeholders within the organization to align on common goals and begin to formulate a roadmap to achieve them. The initial Strategy engagement generally takes about one month, but can vary depending on project scope.
First we do a thorough analysis of the company, domain, and existing technical infrastructure. At this point, we are trying to answer questions along the lines of:
- What are the goals of this project?
- Who are the stakeholders in this project, and what are they most interested in?
- How robust is the existing product and data infrastructure?
- What types of data are missing that we should begin collecting now?
- What is the minimum level of machine learning performance improvements that would be useful in this project?
We keep an open mind and use a breadth-first approach to get a high-level understanding of the entire problem space. Our ideal take-away is an appreciation of the status quo and the general direction of our overall strategy.
Next, it’s time to get concrete. During the Planning stage, we turn the results of our analysis into a roadmap for implementing and deploying machine learning within the product line or organization. Here is what is on our mind at this point:
- In what order should we prioritize the tasks uncovered during analysis?
- What are the simplest, fastest solutions that could potentially meet our criteria?
- What are the short- and long-term implications of pursuing a machine learning strategy at the organization level?
- What types of resources and timelines are realistic for the different approaches we might take?
We strive to get as low-level as possible while maintaining our flexibility. We will produce a document for both the executive-level and product teams with our detailed findings. This will both lay out a long-term vision and strategy for where machine learning could take the organization, as well as a concrete set of next steps to take, focusing on the quickest and most impactful wins first. This will also include estimates on achievable model performance and, where applicable, concrete financial ROI from implementation.
At this point, our team has all of the specific tactics, risks, and rewards for an organization to implement machine learning, either by deploying in-house resources or hiring development contractors. Of course, with all of the domain expertise and context we already have on the project, it often makes the most sense for us to continue our journey with the organization into the Development phase.
It’s time to start making things happen. With a plan in place and agreed upon by all parties, we will charge full steam ahead to turn the company’s ML vision into a reality. The length of this phase will be completely project-dependent.
This is the most straightforward step. Our team will put all of the models and infrastructure in place for a successful machine learning application, as well as all of the necessary product and visual design. We have collective decades of experience building products in Silicon Valley and it all comes together to create the best possible experience for a company and its users.
Tightly coupled to the implementation is, of course, the measurement or evaluation. We go back and forth between these two steps as many times as necessary to achieve the level of performance that we know is possible and will be satisfactory in a production setting. The metrics we optimize towards and evaluate the models on were laid out during the Strategy stage, and it is our pride as engineers to be able to meet and exceed the expectations. Have we mentioned how powerful some of these algorithms are?
When this phase is complete, our client will have a holistic machine learning implementation that solves one or more of their key business goals.
We begin every conversation with a client with this model in mind, and we’ve found that it provides a useful common reference point for all parties.
It is important to note that we must strike a balance between using these proven processes, and being open to change and adapt based on the unique needs of each organization. If you are interested in how machine learning can create a massive advantage for your organization and would like to open a dialogue, please reach out to us at firstname.lastname@example.org