Insights Into How Stanley Black And Decker Achieves Commercial Excellence Through AI

0
199

Artificial intelligence is rapidly gaining acceptance in many enterprises, and proving its value along the way. Some organizations are putting their AI efforts front and center to deliver value internally and to their customers. 

At the recent Machine Learning Lifecycle 2021 conference Prabhakar Narasimhadevara, Director of Data Science at Stanley Black & Decker (SBD), delivered a keynote focusing on how the company implements AI for commercial excellence. He shared how the company has  experience delivering analytical solutions in many application areas including automotive, healthcare, industrial distribution and manufacturing, and digital marketing and services industries. Prabhakar oversees the implementation of AI-driven frameworks for the many brands of products and services under SBD. 

At Stanley Black & Decker, commercial excellence is defined as optimizing all of the commercial actions that are necessary to balance the needs of the end consumer while simultaneously maximizing the value that is generated for all parties along the way, explains Prabhakar. These commercial actions can be thought of as part of a pyramid, with business strategy at the very top supported by brand strategy and marketing and sales strategies, which in turn are supported by customer strategy, fulfillment strategy, and pricing, promotions, and terms and conditions on the bottom level. 

In order for successful execution, a critical component is measurement architecture, which is built to supplement all of the strategies in the commercial actions. This is a way of understanding how to take the notion of a market price and its realization across all of the steps leading up to the point of fulfilling consumer demand and looking at the final margin. You cannot build a measurement architecture without knowing all of the strategies for the commercial actions, says Prabhakar, so it is important to approach the problem of AI and ML formulation in small chunks and phases. 

AI/ML Solutions, Data Ecosystems, and Challenges The scale and complexity of the organization’s architecture and the strategies associated with each level are what drives the need for cognitive technologies, a topic that Prabhakar will share at an upcoming Data for AI online event in September. If a business such as SBD is looking to increase the value in one particular sector, the organization needs to perform a large number of commercial actions that need to be reconciled and kept consistent. This need for consistency means that there must be a digital strategy to approach the problem, explains Prabhakar. This is where AI and ML come into play. 

In order to take actions at the right level of detail, a business needs to define a solution ecosystem for incorporating the right technologies. The solution ecosystem consists of a system of records and allocations, strategy grids, a system of mappings, a system of analysis, a system of visualization, and a system of action. 

There are two main areas of challenges for building an architecture of AI-driven commercial excellence, notes Prabhakar. The first area consists of modeling challenges, since past models are not reflective of how businesses should be run. Past data used to train models often do not have enough variation to drive active customer insights, and data hierarchies are not sufficiently detailed or managed. 

The key challenge, however, lies in the fact that data hierarchies in systems of record are not aligned to business decision makers, because businesses evolve organically but data hierarchies do not. In order to prevent this issue, businesses must remember that data cleansing is a significant and essential part of building the right ecosystem. In addition to modeling challenges, organizational challenges often pose obstacles as well. For example, the existing organizational culture may not be conducive to AI/ML, or data systems may not be agile enough. The right decision-making and deployment structure may be missing, which can lead to decision fatigue or lower quality of execution. Because of these potential dangers, Prabhakar emphasizes the importance of clearly defining the strategies for commercial actions before attempting to adopt AI technologies. 

What is the data ecosystem that is needed to drive successful execution of the commercial strategies and build the right AI architecture? Basic information should include customer POS data, promotion, shipment, sales order, competitor, marketing, external driver data, market risks, customer RDC inventory, financial allocations, strategy matrices, and business hierarchies. Prabhakar shares that with the knowledge of the necessary data, we can understand the AI/ML problem formulation archetype. The formulation archetype consists of the business dimensions, strategy dimension, action lever, response variable, and the loss/margin function, and some of the best places to begin the formulation are in the sales or channel strategy, customer strategy, or service strategy areas. 

From the archetype, an AI/ML model portfolio should include various elements: price driver and response models, product basket and demand models, price waterfall and conjoint analysis, deal scoring, segmentation and promotion models, and whitespace analysis. All of this development does not become established in a day, reminds Prabhakar. However, when building a new data science team, it is necessary to take a portfolio approach rather than just focusing on the immediate use cases. And because teams will run into the problem of having more work than resources, Prabhakar recommends using a portfolio prioritization model that maps the degree of organizational transformation against the complexity of delivery to achieve the best ROI and investment. In terms of creating high level goals, organizations should consider three levels of maturity: first thoughtful action, then enhanced recommendation, and finally cognitive automation. 

Aftermath of Model Implementation and Key Considerations Even after models are successfully built, it is important to remember that this is not the end, states Prabhakar. Constant meta analysis measurement and comparison should be carried out to ensure that one set of models for certain commercial actions is not affecting others, and teams must work to solve any issues as one problem if there are dependencies across different parts. Additionally, model design should incorporate insight assurance to monitor every step of the insight manufacturing process to ensure that each area is well taken care of. 

Above all, understanding the best way to use a model and knowing its limitations is vital to reaching the end goal of commercial excellence through technology. Prabhakar explains that every model comes with the assumptions of its designers, and deploying the model is like learning the habits of other data scientists. Because of this, organizations should take the time to examine the levels of uncertainty with different AI/ML models before complete adoption.  You can hear more from Prabhakar Narasimhadevara at the upcoming Data for AI event on Sep. 2, 2021.

Source