Who are we?
We’re a global data innovation lab,
trusted by our clients to define, design and deliver tomorrow’s
products, services and businesses today.
Praxis is not just our name, it’s what we do and build our reputation on.
Clients: Our Digital Partners
Case Studies
Client: An extremely large, globally known, American stock exchange.
Task: Stock Exchange Fault Monitoring
Challenge
A large stock exchange was needed to
predict system failures. When these failures occurred, trading could be halted, causing damage.
Given the extreme complexity of the exchange, there were innumerable possible causes of failure, and these could not easily be monitored to predict failure ahead of time, meaning resources couldn’t be deployed to avert a failure before systems completely shut down.
Our task was to predict failures before they occurred so that technical teams could respond before systems shut down.
Solution
To predict when the system would fail, we produced several physics- informed mathematical models of how the complex system would work. We then built a machine learning tool that used the models to make predictions about what could happen to the system.
With this Machine learning tool in
place, we could consistently make predictions about system problems before they began affecting performance. These predictions were then fed into an easy to read dashboard that could be understood not just by technical staff, but also by executives who had a stake in monitoring the performance of the system.
Results
After passing testing the fully functional system moved onto production servers where it runs on a live feed of data, allowing for constant system monitoring and early failure diagnosis. Since installation we have not been made aware of any system issues being able to progress undetected to the point where they triggered a market outage. Client renewed contract several times and offered several jobs as a permanent part of their operations.
Client: A well known UK bank
Task: Retail Banking Attrition Prediction
Challenge
Customer retention and support is a necessary feature of any bank, with customer attrition being a particularly key aspect. Customers who go on to leave the bank not only remove capital but, may also spread negative sentiment about the retail bank. To avoid this extensive customer support, monitoring, and outreach efforts are needed.
By monitoring complaint logs timely intervention can stop customer attrition, however this is time and skill intensive, requiring competent operators working full- time. To ameliorate this issue the client company, sought to develop a scalable solution that could predict which customers were likely to leave via the ingestion of customer data.
Solution
Working on behalf of the client, we constructed and supervised-AI approach that ingested customer complaint data in order to predict which customers would leave and highlight them for intervention.
This required a system capable of ingesting inconsistent and sometimes erroneous fields from a variety of sources, all while maintaining data security and avoiding outside exposure. These systems are a core feature of our business and are designed to be repeatable and fast.
Project is still in-progress, with the aim to be completed in late 2021.
Results
Client was left with a fully functional system that could fulfil all the requirements of the NGO mandate, as well as additional features such as the ability to detect water, electrical coverage, and vegetation.
Client continued our contract for further work in processing unique data-sets for a series of hedge funds.
Client: A European FinTech working on behalf of an extremely well-known NGO
Task: Satellite Monitoring
Challenge
Large NGO’s spend tens of billions of dollars a year on developing infrastructure and essential services in Africa. As the continent is massive, dangerous, and generally difficult to monitor, it is difficult to establish the effectiveness of NGO spend in helping develop this infrastructure.
To help ameliorate this issue the client
company, a FinTech based in Europe, sought to develop a scalable solution that could provide the consistent monitoring required for this NGO.
Solution
Working on behalf of the client, we
constructed a hybrid AI-Satellite based
approach. We began by developing a system that streamed worldwide satellite coverage.
The satellite data was then processed, with clouds and atmospheric disturbances being removed, before the data was then sent to the AI.
The AI we produced developed an understanding of what buildings and roads looked like from thousands of labled examples, and then applied that knowledge to the satellite images, identifying all the buildings and roads over the entire continent. It then used ne w satellite images to see how development changed on a month by month basis.
Results
Client was left with a fully functional system that could fulfil all the requirements of the NGO mandate, as well as additional features such as the ability to detect water, electrical coverage, and vegetation.
Client continued our contract for further work in processing unique data-sets for a series of hedge funds.
Client: A service using AI pricing strategies for setting optimal prices
Task: Price Discrimination
Problem Background
Retailers and service providers spend an extensive amount of time and money determining the perfect price point for their products in order to maximise profit and reach. By harnessing the power of artificial intelligence, our code can find the optimal price point not only for individual products, but for individual customers, faster and with less risk than ever before.
Solution
Our newest service is a dynamic pricing tool, powered by state of the art reinforcement and machine learning. Our tool can predict the prices that individual users are willing to pay for identical products, as some clients may be more price sensitive than others. By using an individual pricing tool, any company can provide products with a higher margin to those individuals that are more willing to pay, while still retaining individuals that would otherwise seek cheaper services.
This system works at speed, with calculations performed in fractions of a second and prices changed as fast as the client can receive a signal from our server. This means that prices can be customized directly for a customer before they even reach the landing page. A customer would be entirely unaware that prices were individually set, as the prices do not change once the customer reaches the site. This system works out of the box, but improves further as more customers purchase your product and this information is fed back into our AI system. The tool is safe and secure as our servers are encrypted and don’t store any personal information.
Our dynamic pricing tool works by embedding a javascript snippet into the client’s website, which quickly acquires a customer's information as they access the site. This information is then sent via API calls to our secure servers, which is then enriched and analysed to build a user profile. With our state of the art reinforcement learning methods the profile is then used to predict the ideal price margin for that customer. When customers purchase products, this information is fed back into our tool in order to further refine our margin prediction models.
Results
Systems have modified prices on over a hundred million dollars worth of product, with average increases in profit per user of 9% for our customers. Our solution is free to install and charges based on performance. It is accessible at www.pricehoney.com
Contact Us
Contact us