Rapid prototyping with Lean Labs
Many companies struggle to foster innovation within their existing teams. The challenge is usually cultural—how to marry the need for experimentation, freedom to "fail fast" and a creative license to throw away the "old ways of doing things" with compliance to corporate standards and security practices. Grid Dynamics has a service offering that helps companies try fresh ideas quickly outside of the corporate bureaucracy before bringing them in-house for productization once they have proved to be successful.
We call this program Lean Labs, and it works like this: a dedicated, self-sufficient team staffed by Grid Dynamics engineers works on a customer problem with the client's subject matter experts. The team moves fast, uses the latest technologies, and aims to solve the business use case to demonstrate measurable value to the business stakeholders. The project takes anywhere from 4 to 12 weeks and costs under $200K.
largest US retailers
largest US technology companies
world’s largest consumer goods companies
largest US financial companies
How rapid prototyping works
Emerging technology center
Many problems in supply chain, revenue management, and personalization are traditionally solved by handcrafting an optimization task that can be tackled using standard numerical or combinatorial optimization algorithms. Reinforcement learning is a game changer: it provides a generic multi-step optimization component that can learn optimal control policies from system logs or simulators. We put a lot of effort into the development of applicable reinforcement learning solutions that adapt the latest advancements in this field to enterprise use cases.
Many traditional models for personalization, recommendations, price management, and supply chain optimization are designed to use only one type of data, such as clickstream or sales transactions. In practice, it is almost always beneficial to combine multiple data sources to create more accurate models and gain a deeper understanding of the processes behind the data. We use deep learning and representation learning techniques to solve use cases like the following:
- Improve demand forecasts for slow-moving and new products using a product similarity graph that accounts for numerical and categorical attributes, macroeconomic and market signals, own and competitor prices, and textual descriptions.
- Improve product recommendations by combining customer behavior data, textual product descriptions and reviews, product images, and other data sources
The main idea of semantic vector search is to represent both products and queries as semantic vectors in the multidimensional semantic vector space. Products and queries have to be mapped to vectors in such a way that similar products and queries close in meaning would be clustered together.
This is achieved by training a deep learning model based on all available catalog data and customer engagement history mined from the clickstream. The model takes into account all available data about the products, such as attributes, images, descriptions, reviews, prices, and promotions, to find the best possible vector representation.
Advanced statistical and econometric models can provide deep insights into customer behavior, market demand, and equipment reliability. We have done many enterprise AI projects for our clients and learned that these insights are very important for getting tangible business results. Combining our domain knowledge with advanced technical expertise, we created a toolkit of interpretable models and decision support tools that helps to solve use cases like the following:
- Understand the structure of the market demand and outcomes of promotion campaigns, taking into account demand cannibalization, pull forward, and halo effects.
- Understand what drives customers toward conversion, churn, or in-app purchases by using clickstream, demographics, call transcripts, product reviews, and other data sources.
Modern computer vision techniques are extremely powerful and can be applied to a wide range of enterprise problems, including product recommendations, visual search, quality control, and traffic analytics. We have extensive experience with a number of computer vision use cases and have created a comprehensive collection of models, development pipeline templates, and production deployment components that enable us to build end-to-end computer vision solutions with unparalleled productivity.
As people become accustomed to Amazon Alexa or Google Home, they are learning that a conversational user interface (CUI) is an intuitive way of interacting with digital channels. According to data published by Voicebot, the number of Amazon Alexa skills in the US has more than doubled since 2018.
We use state-of-the-art NLP models to solve tasks such as intent classification, entities and relation extraction, and coreference to develop conversational agents.
Personalizing in-game experience using reinforcement learning
- Personalize in-game experience
- Reduce model development effort
- Increase long-term engagement / LTV
Grid Dynamics' solution
- Reinforcement learning based personalization platform
- MVP delivered in 8 weeks
Up to 25% dollar-per-user improvement compared with the baselines
Price optimization for video games using machine learning
- Optimize promotions across many channels and countries
- Forecast the demand 24-month ahead
- Properly handle new game releases
Grid Dynamics' solution
- Demand forecasting models
- What-if analysis tools for promotion scenarios
- MVP delivered in 6 weeks
- Manual process replaced by data-driven optimization
- Increased promotion efficiency compared with the baselines