I’m pretty regularly asked for my thoughts on how to go about applying machine learning (ML) within an organization. I think there’s much more nuance here to unpack that one typically thinks about when trying to apply ML to a specific aspect of an organization.
Within industry, and even academia, there is a bit of a misconception around what actually is possible today, given the constraints within a corporate organization. Constraints that aren’t necessarily always technical tend to provide more of a challenge than those that are purely technical. However, understanding what about your business you’d like to automate, potentially using ML, and what that means to your business from a staffing and financial investment perspective is arguably more important than understanding the technical details of methodologies used within the ML space.
The application of machine learning within a business looking to newly adopt this technology tends to run orthogonal to the traditional engineering practices present in many enterprises, and this is to be expected. New technologies tend to follow adoption curves, and for good reason, as it can sometimes take decades for abstractions to be created that allow these discoveries in science to transition into the realm of technology ― or to where a user isn’t required to have a deep understanding of the complexities hidden beneath the surface.
Exploration vs. exploitation
Following this thread of being a bit non-traditional, it is important to understand that strategy isn’t where you should start when trying to understand how ML will fit into your organization. Viewing this through the lens of exploration versus exploitation, traditional corporate strategies tend to err on the side of exploitation, especially when applying technology to the problem space. They look at their existing business, existing processes and existing staff and plan accordingly.
However, when applying ML, you are largely working within the exploration side of this framework. It is not intuitively understood whether the problems an organization is looking to ML to solve are solvable using ML to begin with. We have a limited understanding of what is possible in certain problem domains, such as computer vision and natural language processing.
So, if you’re trying to apply ML to some less-understood problem domain, then you, as a business, would need to understand whether it is a worthwhile investment, and understanding what constitutes a worthwhile investment for your business does not come easily. The best approach here is to understand that your data science/ML team is ultimately working to provide your business with new value that hasn’t previously been realized and that this doesn’t typically fit into a strategy-first approach to business.
Building a baseline understanding
To build on this observation, you may recognize that you have a bit of a bootstrapping problem in that you’re looking to apply this innovative technology that promises unimaginable gains. However, you don’t know whether the gains can be realized in your specific problem domain. Lacking the ability to intuitively understand how these technologies will perform within the organization, you can appeal to what looks and feels like a research and development effort.
Once an organization has decided they would like to understand if and how ML can help improve their business, they can then staff a team to help bring value to their organization. This team should look to build pilot projects that can showcase some of what is possible using ML, specific to the business, and how that fits in with the current operational model. It is important to note that your operational model shouldn’t just include stakeholders within the technology side of the business, but those throughout. Everyone from finance to marketing to logistics should be involved in building a baseline understanding of what ML could do for your business moving forward.
Now for strategy
Once this baseline is built, you can then look to scale some of the work done in the proof-of-concept (POC) phase and to measure the rate of return on the overall investment being made, from people through to technology. Scaling a POC requires investment from all areas of the business impacted by the results provided by the ML system. For example, if you’re modeling click-through rates for an ads business, then you will want to engage the accounts teams to help them understand how your ad-serving product is better enhanced for your customers. Or, if you’re modeling logistics for a manufacturing business, you’ll want to involve the procurement teams to help them understand what the output of an ML model will provide them from a business point of view.
Now that you have a thorough understanding of what it takes to investigate the application of ML somewhere within your business, what it takes to interact and integrate with your existing lines of business, and what it takes to scale a POC to a point of value for the business, you can integrate all your learnings into your overall corporate strategy.
It is important to understand that applying machine learning to your business isn’t something that will be “appliance oriented.” There likely won’t exist a black box you can purchase and install in a data center that you can then pour data into and expect to provide immediate value. Like most other areas of your existing businesses, it will require measuring and understanding the capabilities of the technologies in your problem domains, the impact across all areas of your business and, once this is understood, the investments that will need to be made to scale the solutions.
Investment in building a team― or discovering one within your organization that is already toying with applications of ML ― should be the starting point for helping your business to adopt this new technology. Once your organization has transitioned from the “science” side of the world to the “technology” side of the world, you can then work to scale these solutions and help machine learning to provide new, meaningful value to your business.
To learn more about unlocking the value of data with artificial intelligence systems, explore Dell EMC AI Solutions and Dell EMC HPC Solutions.
Ed Henry is a senior scientist and principal architect for machine learning at Dell EMC.