4.5% uplift in gross profit after using Competera’s demand-based engine

Omnichannel electronics giant sees 4% growth in key business indicators from price recommendations powered by Neural Networks.

Dear reader, this is a non-standard case study. On the one hand, due to legal arrangements, we can't provide our client's name. However, such an impersonal structure allows us to reveal detailed and real figures of the project as well as to show the work of our algorithms in all aspects.

Our client is one of the most successful omnichannel electronics retailers in Asia and Eastern Europe. With a versatile range of products, the retailer sought to optimize its pricing process, and as a result, to increase the margin using the portfolio pricing strategy. After implementing Competera’s bespoke deep learning algorithms for price recommendations, the client was able to achieve several key business objectives, including a cumulative 4.5% uplift in gross profit.

4.5%

uplift in gross profit after 8 weeks

4.4%

increase in total revenue of the test category

42 000

applied price recommendations during PoC

The challenge

Since our client belongs to the major market players, the assortment of any of its categories can be described as a variety of brands of any model, color, or shape. Motivated to streamline the pricing process, the pricing team required a solution that could assist in solving a number of challenges.

Price changes for analogues of direct competitors affect sales more than changes in own prices

Different reactions of demand to price changes of different products

Sales movements within the portfolio from more profitable to less profitable products

The need for frequent repricing of large quantities of SKUs

The Solution

To meet these challenges, our pricing architecture team suggested using a demand-driven pricing engine powered by Neural networks. It can consider price elasticity of demand, cross-elasticity, competitive environment, and other crucial factors to recommend optimal prices. Category historical data for 2.5 years (sales, stock, geography, margin, motivation, promo, etc.) was taken as a basis for calculation and design of ML algorithms.

Solution

Execution

The whole process of work with Competera can be conventionally broken down into several independent milestones. Let's go over each of them.

The PoC Design

The project’s success was determined by the growth of the target metric – the gross margin. Using algorithmic pricing from Competera, the client expected to see the metric grow by 5% or more.

Revenue retention was chosen as the metric to protect. In addition, the algorithms had to automatically accept and take into account the client’s existing pricing rules, recommending changes only to shelf prices. The decision to conduct promo remained on the client’s side as well as the list of potential promo models.

For the Proof of concept launch, we chose a method of comparing the test and the control groups across two different regions with similar sales history and customer behavior.

Execution
Execution

The implementation

Full deployment of the platform to recommend optimal prices goes through a multi-stage process.

1 month

Setting project goals

Integrate, set up and check continuous data flow

Training of ML models

2 month

Using price recommendations for one test category

Process debugging and model improvement

3 month

Scaling

The fine tuning (key learnings)

If you are preparing to implement such a solution in your business, be sure to pay attention to several factors. Your data output, namely new prices, directly depends on data input. During current project, we faced several interesting moments
  • ML algorithms can automatically round prices to get a number that could be put on the price tag. However, it can lead to a significant impact on products with low price elasticity.
  • Some goods are systematically returned by customers. For instance, due to promos like Ignoring this fact leads to inaccurate forecasts and, as a result, dropped target metric.
The fine tuning
The fine tuning

Results

While finding optimal price points across the range, our algorithms most often make two sorts of decisions. The first situation is when a product’s demand is elastic to its price. It reacts well to price decreases and badly to price increases. This situation is quite typical for the economy price segment. Typically, algorithms recommend decreasing prices after facing such behavior. You can see the reaction on the charts below.

Note: Each chart displays the historical data of one particular SKU and contains three metrics. They are the total revenue from sold products per day, a shelf price set by the customer team, and a price recommended by Competera. If the shelf price index correlates with the price recommendation index, it means that the pricing team accepted the suggested price.

headphones chart
This basic headphones belongs to the medium price segment. Two recommendations to reduce prices resulted in revenue growth both times.
headphones chart
In the second case with earphones, you can see the period before and after Competera’s technology was implemented. The algorithm recommended decreasing the price that category managers haven’t revised for a long time. As you can see from the chart, price decrease led to an increase in revenue and sales.

Let's consider the second frequent type of situation when the product reacts well to a price increase. In such a case, it's logical that an algorithm would recommend setting a higher price. We will observe an increase in profit without a fall in revenue. For example, we can see it on these charts.

headphones chart
As you can see, Competera immediately recommended increasing the price of headsets shortly after implementation. However, the client team did not accept the recommendation for some time. After the price was increased, the revenue didn’t fall, but on the contrary, the gross profit increased up to 15% compared to the same period with the lower price.
headphones chart
On this chart for the pro headphones, you see how the price was regulated both upward and downward. After the first recommendation, we saved the revenue but managed to increase the profit. After the recommendation to reduce the price, the revenue began to show growth.

Of course, these are just a few options for situations taken for specific products. If you look at the picture more globally, our work with each SKU separately and in relation to each other, as the true portfolio pricing approach assumes, showed cumulative growth in general. Thus, the growth of the two targets was 4.4% in revenue and 4.5% in gross margin. Thanks to effective cooperation with the client, all the project goals were achieved. The company has moved to the next stage of its pricing journey.

«Today, portfolio optimization and artificial intelligence are still fighting for retailers' trust. However, proper teamwork and trust in each other lead to incredible results, as this case shows. I think that close collaboration between the teams is a major part of the success of implementing innovations in modern retailing».

Alexandr Galkin
Alexandr Galkin
CEO of Competera

Competera Pricing Platform helps retailers to craft optimal offers

Discover how Competera helped other retailers to thrive
Consumer electronics
Foxtrot
Maximazing Sales Without Risking Profit Margins
Sporting Goods
Wiggle
Market-driven pricing for over half a million SKUs
Apparel & footwear
Intertop
Markdown optimization: saving 200 b.p. in profit margin

Want to know more or leave a comment? Email us at [email protected]

Best Analytics / BI Solutions

E-commerce Germany award

Price Optimization Solutions

Constellation

#1 eCommerce Solutions Software 2019

Crozdesk

Top 3 startups at the AI Summit

London Tech Week

Great User Experience

Finance Online

High Performer 2020

G2 Crowd