In upcoming years, artificial intelligence is predicted to reinvent sales and pricing domains and become extensively adopted by sales and pricing professionals. Competitive organizations are already using AI to automate tasks, improve efficiencies, and drive more revenue.
AI allows sales reps to work smarter and faster, and identify the best next move. As an example, AI can enhance sales efficiency by standardizing the operations of sales teams based on the collection and analysis of a proliferation of data sources: historical transactions, win/loss analysis, competitive price levels, and contextual data.
An additional advantage is enhancing customer experience, as studies show that 75% of business buyers expect companies to anticipate their needs and make relevant suggestions.
The business problem: automating special pricing
A recent project involved creating a machine learning model for auto-approving and auto-denying a large volume of special pricing deals. Special pricing is traditionally led by a team of pricing professionals, and their role is to support pricing requests that exceed a smart price floor.
Smart price floor is a discount limit assigned to Dell products based on deal size, the customer history, as well as the deal richness. The floor is the highest discount possible the salesperson may offer the customer without seeking further approval.
The process of approving or denying is a manual and labor-intensive process, causing significant delays in the approval of deals. Not responding fast enough might lead to lost opportunities and revenue.
Applying machine learning to solve the problem
An initiative to automate special pricing within Dell with machine learning started in early 2019. As is typical in many data science projects, it took several iterations with the customer, in addition to relevant data explorations, to come up with a solution and model that could be implemented and generate real value.
Matching the best solution to a given business problem requires a mutual learning of both the business unit and the data scientist. Through the process, the business unit gets a better grasp of what can be done using data science, and the data scientist gets a better understanding of the data and the opportunities that lie within.
The engagement started with the intent of creating customers’ segmentation and to use that to define the proper discount for the customer. Yet after pursuing this direction, the data showed that a discount given for a certain customer cannot be based solely on the characteristics of the account, but rather on a combination of different aspects, such as the products the deal is composed with. So, a score was generated using statistical methods for different aspects of the deal: the account, the products, and the deal. In addition, an aggregated score was calculated for the three different aspects.
Realizing the benefits
The feedback for the overall solution was good, the individual scores for the account, deal and products were informative, however, the aggregated score for the three aspects of the deal was confusing for users. The final solution was a scoring of a deal as a whole, accounting for all the different aspects.
The deployed solution includes a machine learning model which reduces the volume of deals that need to be manually reviewed by the special pricing team. By doing this, the special pricing team can focus on deals that are more complex.
For each new deal that comes in, a score that reflects the likelihood of approval is generated. Considering the history, the false positives, and the false negatives, a threshold for auto-approving and auto-denying a deal is set. Deals scored above the approval threshold will be automatically approved and deals that were scored below the denial threshold will be automatically denied. The rest of the deals will be manually reviewed.
Since each region has a different set of rules, a different model was developed for each region. The development process was mainly focused on creating features that would capture the attractiveness of the deal based on the purchase history of the account, the richness of the deal, as well as the revenue generated from the deal.
Since the process of denying a deal also included providing an explanation to why the deal was rejected, we incorporated a reasoning mechanism in the solution. For the reasoning mechanism we used SHAP values, which break down a prediction to show the impact of each feature in the model. Using this mechanism, we can provide an explanation to the salesperson quickly and allow him to adjust the deal accordingly.
This solution is currently being deployed and integrated into Dell’s systems and processes. The integration of the solution facilitates a more productive workflow. It allows us to identify patterns of deals that can be automatically approved or denied, and enables the special pricing team to focus on more complex deals.
This solution also improves customer experience and reduces lost opportunities, since in cases when deals should be automatically approved, the response will be faster. Lastly, it allows the decision makers to base their decisions on rational factors that stem from the variables of the machine-learning model. This leads to in a more confident, data-driven pricing policy.
Special thanks to Shiri Gaber for editing and proofreading this blog, to Amihai Savir for consulting throughout the project, and to Dhev Kollannur, our business partner and a big AI promoter within the organization.
To learn more
Whether it’s special pricing or other business challenges that your organization wants to address, check out Dell EMC Ready Solutions for AI & Data Analytics.
Copyright © 2020 IDG Communications, Inc.
Credit: Google News