Very few companies achieve pricing goals, with the majority missing these targets, therefore having a negative impact on profit margins and overall profitability.
Pricing strategies are a smart lever that businesses can use to increase profitability. A 1% improvement in pricing can increase net margins up to 11%. In a Bain & Company survey, 85% of global C-Suite leaders, and Sales VPs “believe their pricing decisions could improve.”
Despite the need to improve pricing decisions and outcomes, why are so many companies struggling?
The majority of companies operate using cost-plus, or a pricing strategy that is heavily influenced by competitors/market forces, or price banding and discounting is used to retain long-term accounts. Neither of these three models is as effective as customer-driven pricing, a way of determining what customers are happy to pay, rather than what companies think they should discount to keep them as customers.
One of the reasons companies struggle is that the technology to implement real-time customer-driven pricing hasn’t existed until recently.
In this article, we look at how actionable data-driven insights are generated for businesses to solve complex customer-centric pricing challenges, based on interviews with the team that have made Bubo.AI possible.
What is the business impact of Bubo.AI?
“Too many businesses know they’re leaving money on the table. Automatic discount price banding, depending on what a customer spends and other factors, often has a negative impact on profit margins,” Alan Timothy, Bubo.AI CEO said in an interview.
“We have come up with a working, proven solution, that captures lost margins without the risk of losing market share.”
“With Bubo.AI, when we work with a client, we start by analyzing customer spend history for three years. Next, we analyze the spending patterns and behaviors of similar customers over the same 3-year period,” Timothy explained.
“Once we have that data, the Bubo.AI generates a price or price band for a product or range of products, for each individual customer. In on-the-fly pricing negotiations, such as when a customer steps into a branch, calls in an order, or meets with a field sales rep, our solution shows them what each customer is actually wiling to pay, based on the data”, Timothy further explained.
How does Bubo.AI work?
The customer-driven prices that Bubo.AI generates starts with Behavioural Customer Network Analysis (BCNA).
Behavioural Customer Network Analysis (BCNA) is the process of representing complex data sets as nodes and links in a network to be able to employ graph theory on them.
As Dr. Huseyin Seker, Bubo.AI’s Senior Data Scientist explains, “Graph theory enable us to reveal hidden patterns, social structures, and cliques in the underlying data. This is a useful way to utilize and extract insights from multiple data sources to enable informed and insightful decision making.”
“BCNA is achieved by means of spatial, transactional, behavioral data analytics and results in a social network representing an interaction between data points (e.g., customers and products); as you can see in Figure 1. It can help discover every aspect of social structures and cliques inside the network”, Dr. Seker explains.
Bubo.AI “utilizes a cutting-edge BCNA technique that is based on multi-source data sets and ensemble hybrid data analytics and artificial intelligence techniques. This unique tool can be further enhanced by integrating the company’s own customer profiles to uncover hidden patterns inside customers’ behavior and their transactions.”
Dr. Seker further explains, “For example, we can identify which products are preferred with individual customers and how we should take this information into consideration when we make a price suggestion, or we can segment customers based on their behavior and purchase history.”
Figure 1: A typical Behavioural Customer Network Analysis analytics process
Next, we look at predictive models and how these impact clients’ results.
Predictive modes are built using machine learning methods upon historical data sets to be able to predict likelihood of the outcome of actions and what it may happen in the future.
Figure 2, below, illustrates the development of predictive models which involve several algorithmic phases. These follow the typical machine-learning-based predictive model regardless of binary (e.g., classification of an object) or continuous (e.g., prediction of actual price) outputs.
Dr. Seker explains that, throughout these phases: “feature extraction and selection from data sets are carried out to characterize the data sets. As the data will be represented by several hundred features, one of the most crucial steps is to identify the most significant sub-set(s) of the features. This allows us to capture the most significant data about interaction between the features.”
Doing this manually would take hundreds of years (e.g., for n=20, there are over 1 million combinations of the features, and in most data sets there are more than 20 features). Even doing this using traditional computing techniques and processing power would take decades.
With machine learning algorithms (e.g., genetic algorithms, deep learning architectures) and cloud-based computing power, running billions of calculations is cut down to a matter of weeks. Machine learning algorithms “play an important role in iteratively selecting the most significant characteristic features of the given data set. As multiple data sources help further enhance the predictive models, this approach also helps identify the validity and robustness of a particular data source”, Dr. Seker said in an interview.
“These algorithms are iteratively trained using historical data sets for pre-defined tasks (e.g., a price or price range that could be offered to a particular customer, along with suggested product(s)). Given the complex nature of the data sets and their (selected) features, several predictive models, each of which is developed using a particular machine learning and/or sub-set(s) of the features, are required to be combined to devise the most accurate and reliable predictive model.”
Figure 2: Main algorithmic phases used to develop a robust predictive model
In summary, Bubo.AI uses a machine learning workflow that continually optimizes the price through the process feedback loop. This way, the predictor becomes a business KPI to measure the performance of the process in correlation with profitability impacts. It helps businesses use it to manage business performance, with the ability to tailor the value offered to the customer within operational models.
How is Bubo.AI different from CPQ solutions?
Known as the Configure, Price, Quote (CPQ) software market, Gartner estimates it will continue growing at a pace of 20% annually, currently worth over $1 billion. In the CPQ market, most of the big tech players have a solution, including SAP, Oracle, IBM, Salesforce, and a number of SaaS and cloud-based solutions.
“In the right context, CPQ solutions and software are incredibly useful. If a big client wants a discount on 50,000 Mercedes left front-wheel break pads, then a supplier needs to adjust the margins on other product ranges to keep profitability the same, or improve slightly”, According to Timothy.
“Bubo.AI, on the other hand, is far more useful for those on-the-fly, last mile pricing negotiations that happen hundreds of times a day, in thousands of branches, on calls, and in meetings between managers and reps and their customers.”
“Instead of them working on ‘gut’, discounting for the sake of winning the business, or discounting because a banding model told them to, Sales reps and managers can quickly generate a price that accurately reflects what each customer is really willing to pay.”