Humans have to test AI solutions to judge their efficacy and benefits
There is much consternation nowadays regarding the inevitable loss of millions of jobs to AI. Often the doom and gloom revolve around robotics and automation of repetitive tasks taking over everything from warehouse operations to decision making. Our experience tells us something different. Although it is true that robots in warehouses don’t run out of energy and process automation doesn’t “do lunch,” there are still many touchpoints in any AI-driven business process that will require AI/HI (artificial intelligence/human intelligence) collaboration. While the former offers speed, scalability, and quantification, the latter is equally valuable for its leadership, teamwork, social skills, and creativity.
To make our point, we offer the example of a price optimization project with a wholesale food distributor that highlights many essential AI/HI touchpoints necessary for a successful outcome—from project inception to AI deployment. In other words, incorporating AI into any business process will require collaboration between the technology and those using it, or it may not be successfully adopted.
Wholesale distributor pricing strategies have gotten more complex since the landmark 1992 study (1) by McKinsey showing that a 1% price improvement could result in an 11.1% increase in operating profit. That price was determined not by invoices but by actual transaction prices. Today, there are so many considerations beyond invoice or even transaction pricing data to optimize daily product pricing, including external data sources, complicated technologies, compliance issues, and vertical-specific business goals.
Additionally, fast-evolving AI technologies that can parse through unimaginable amounts of data at near real-time speed have facilitated accurate prediction and prescription functionality at scale, so distributors no longer have to rely solely on the intuition of their category managers and salesforce—drawn from less data
1. Identifying a Pain Point That AI Can Fix
To begin any business AI initiative, a gap or pain point must be identified. Who better to identify gaps in a process than those using that process within the business? In our example, C-suite category managers, procurement, and sales teams all identified the need to handle copious amounts of data quickly to make the best pricing decisions for themselves and their customers. They knew it would require the use of data and artificial intelligence to get there. They also knew their process of calculating base prices and discounts was only part of their problem. They had unique requirements because of the diversity in customer types, large number of SKUs, and a well-oiled process for delivering price quotes each day to their sales reps. They needed to maintain at least part of that process which an “out-of-the-box” pricing solution would not allow.
Food distribution has the special complications of product shelf life (perishability) and faster delivery times required to preserve it. Salespeople act like traders, ensuring that the buying price from suppliers allows a margin between it and the selling price to customers. In between, they must be cognizant of market trends: Is the market going up or down? If they anticipate an upswing, can they sell at a price to advantage the buyers and their company? Here, the workforce helped to identify where they were having difficulty in accurately and reproducibly considering all of the necessary data to make the best decisions quickly. The distribution industry is dynamic and volatile, “a pressure cooker,” noted our project manager. The pricing leaders at the company worked with our analysts to create a roadmap of existing processes and helped identify where data and AI would most benefit them.
2. Finding the Right Approach
Humans have to test AI solutions to judge their efficacy and benefits. To begin to apply AI to the distributor’s pricing process, the company and A2Go formed an integrated team composed of our domain and data-science experts and their pricing experts using their existing methodology. This would be the company’s first AI project. Helpfully, KPIs had already been established to increase the top line at the firm.
In AI, the devil is always in the details, and our team found some hiccups of missing and incorrect data that had to be sorted so the algorithms would produce accurate outcomes. “Every day, we got new data from purchase orders and inventory positions,” said our manager. “We created a process whereby we could digitize the category managers’ insights as well.” During the experimentation phase, we were able to have AI augment the human process with automation at piloting scale and have humans augment our AI models with their insights into the nuances of the process.
3. Creating the Solution
Finally, our data scientists got to work doing what the project manager calls the “engineering and piping.” They determined that not one, but two algorithmic models would be needed to optimize the company’s pricing process. The first was designed to help category managers predict supplier pricing trends and optimize the base selling price for each of the company’s thousands of SKUs. The second model was designed to define base selling prices for each customer. A clustering technique was used to simplify this process, grouping customers based on their buying behaviors. All models were tested and evaluated with humans controlling the inputs and judging the outputs.
Our work resulted in an AI-driven price optimization application built to be invisible within the company’s existing workflows. The application is scalable, fast, and easily adoptable by the workforce. This food distributor experienced increases in revenue and margins by 3 – 4%.
4. Adopting the Solution
After deployment of the AI pricing solution (Price-Right AI for Wholesalers), category managers and the salesforce adopted it with confidence, knowing it was there to augment their work and not replace or overrule them. What we learned is that oftentimes human intelligence has to prevail, especially when there are intangibles that cannot be quantified for the models.
This collaboration, automation, and augmentation frees up time for the workforce to grow their customer base and focus on their relationships with existing customers. This company is now engaged in a continuous improvement process with A2Go aimed at optimizing AI/HI collaboration going forward.
Reference: “Managing Price, Gaining Profit,” by Michael V. Marn and Robert L. Rosiello, Harvard Business Review, September-October 1992.