My name is Bashir Umudov. I am a product manager with experience leading startup products and mature core products in BigTech companies, as well as ML/AI solutions. Today, I would like to talk about an interesting business challenge that can be solved using ML algorithms - dynamic pricing.
Many have probably noticed how, after refreshing a page while buying airline tickets, the price suddenly changes, or how the same taxi ride costs different amounts at different times of the day, or how product prices may differ depending on the device operating system.
This is how dynamic pricing works in the modern world. Technology companies actively integrate dynamic pricing into their products to grow key business metrics. According to McKinsey’s research “The Hidden Power of Pricing,” revenue growth from dynamic pricing averages 6-9%.
This is a significant uplift for businesses, which is why companies are willing to allocate resources to build their own dynamic pricing systems.
One of the most interesting and complex implementation cases is the e-com/e-grocery market. The complexity comes from tens of thousands of SKUs across multiple product categories that require price prediction. Additional complexity comes from promotions and cashback programs typical for this market. In this article, I want to highlight practical aspects of such a system: how it should work, what capabilities it must have, and which metrics it should influence. This information will be useful for managers implementing and developing dynamic pricing systems.
First, let’s outline the problems dynamic pricing solves as a product:
Automates SKU pricing;
- Simplifies and scales the addition of new SKUs, stores, regions, etc.;
- Forms dynamic prices based on product attributes, inventory levels, transaction activity (demand), weekdays, seasonality, holidays, regions, stores, competitor prices, etc.;
- Creates beneficial prices, promotions, and cashback offers for consumers, including through personalization and segmentation;
- Increases margin and revenue (GMV);
- Provides transparent analytics on prices, promotions, cashback, margins, GMV, etc.
Otherwise, these tasks are handled manually. There is a base price, then promotions and cashback are applied. Often these live in separate systems and require manual uploads for tens of thousands of SKUs across regions with different purchasing power and competitors. Manual management consumes significant human resources and increases the risk of errors. As a result, businesses lose revenue and negatively impact customer experience.
How should a system be designed to support hundreds of thousands of prices, cashback, and promotion decisions:
1. Pricing must include base prices, promotions, and cashback managed through an internal UI admin panel.
2. The UI admin panel must allow rule configuration (input parameters) for SKUs and enable price, cashback, and promotion setup, planning and launching campaigns with segmentation by regions, store formats, product categories, customer segments, etc.
3. Dynamic pricing must consist of a base price engine, promotion engine, and cashback engine that automatically generate price, promotion, and cashback values based on rules.
4. Dynamic pricing should operate in the following order: first the base price is formed, then promotions and cashback are applied. Promotions and cashback reduce margin but positively impact GMV. Minimum margin flags must ensure total discounts do not fall below a defined threshold.
Here’s the dynamic pricing scheme.
5. Engines should be powered by ML algorithms adjusting base prices, promotions, and cashback for SKUs. The base price engine uses:
- Product attributes;
- Transactional activity/current demand (views, clicks, purchases);
- Inventory levels;
- Time of day, weekday, seasonality;
- Competitor prices;
- External factors (weather, events, holidays);
- Minimum margin, maximum discount, etc.
The promotion engine represents the second layer. It works on an uplift principle: what discount percentage maximizes sales growth and which products should receive promotions. It answers: Should we apply a discount? What discount? For how long? In which region? Will it increase GMV?
The cashback engine can also be implemented using ML.
6. The UI admin panel must include analytical tools (dashboards) for margin, GMV, and promotion results monitoring, engine control, and system alerts.
Which product metrics should be tracked when implementing dynamic pricing?
I would highlight several groups.
Key metrics:
- Margin and GMV. A pricing product cannot be evaluated by a single metric. Margin may grow due to higher prices, but GMV may fall due to reduced sales volume. Looking only at margin can be misleading. It is best to analyze margin and GMV together. The ideal scenario is margin growth without GMV decline.
Proxy metrics:
- Average Order Value (AOV). Shows how prices and promotions influence basket size and affects both GMV and margin;
- View-to-order conversion rate. Reflects user response to prices and influences GMV.
Health metrics:
- LTV;
- Retention;
- Share of SKUs, stores, and regions covered by the automated system. Growth in coverage indicates scalability;
- Time-to-market for adding or changing prices, promotions, and cashback.
A metrics tree may look as shown below, demonstrating the relationship between price management and key business metrics.
For businesses, dynamic pricing can become a source of competitive advantage due to faster adaptation to market demand. Price becomes a product that can be A/B tested and optimized under different pricing strategies to generate measurable business impact. For this reason, dynamic pricing should be one of the core business directions for large technology companies.