Your “unfair” competitive advantage in a world of unlimited data, quick feedback, and countless opportunities.

About us

We are a technology consultancy helping businesses to overcome competitive challenges and to develop new ways of delivering value through predictive analytics, APIs, and data science.


As our client you'll find in us a long-term partner with domain knowledge, technology expertise, and proven methodologies to deliver the outcomes you envision.

We are effective; by listening carefully and thinking clearly, we solve the right problem.

We are efficient; our experience means we pick the right tools for the job and use them well.

We are consistent; our tried and tested methodologies lead to predictable results.

Linas Juškevičius

Linas Juškevičius

Director and Software Developer

Linas obtained his BSc in Software Engineering from Vilnius University. He has been coding for as long as he can remember. But his passion for programming doesn't stop there. In his spare time Linas has also been voluntarily teaching school children about Computer Science and the art of programming.

In Linas' world nature and technology have a symbiotic relationship. Taking long walks in the forest is his source of inspiration, a way of unwinding and letting fresh ideas breathe.


Vidas Pažusis

Data Scientist

Vidas got his BSc and MSc degrees in Physics from Vilnius University and breathes it every day: “Physics is a way of life when you find wonder in everything around you”. Yet, his professional passion is programming, fueled by the beauty and elegance one could achieve in code – constantly striving for perfection. In life, he's a modest idealist who craves learning and values family and fairness.


Kostas Feruliovas

Equity Analyst

After getting his BSc in Economics at Vilnius University Kostas acquired an MSc in Finance from VU University Amsterdam. He is passionate about the discipline of Finance and he is working towards expanding his practical and theoretical knowledge every single day. In his spare time he enjoys spending time with his family, traveling, or playing sports.

Services

For Marketing and Ecommerce

1. Create a Single Profile for Each of Your Existing or Prospective Customers

Why Would You Need This? The most prominent use for customer profiles is as an input to other analytical processes. Consolidated data from web analytics, e-mail campaigns, transactional data and other sources allows you to correctly identify customer status and value. Opportunities to capture cross-channel shopping arise. Customer preferences can be leveraged across all the stages of customer interaction. Costly errors of a bad treatment caused by lack of information (e.g. offering a discount after a customer already bought the item) can be avoided.

Example: A single customer profile may include data from web analytics, subscription mailing lists, transaction logs from the cart and from the payment processor. Also, Facebook likes and comments, traffic origins from search and display ads, support phone and chat logs, emails.

Technology: Ingest data from different data sources. Apply de-duplication and matching techniques to gather all customer data in a single location. Optionally use "householding" (sometimes your implicit customer is a household, so you want to look at spouses, kids as a single unit). All this data can be used directly (by customer support for example). Use the data when training predictive analytics models and performing analytic queries. Supplement single customer view with the calculated data from the models (predictions).

2. Predict Which Potential Customers are Most Like Your High-Value Customers

Why Would You Need This? Discover potential high-value customers among prospects and new customers early on. You can reward behaviours indicative of high-value customers to positively affect customer evolution. You can plan upsell opportunities for prospective high-value customers.

Example: Phil's online store sells party tricks, weird souvenirs and other peculiar things. Most of his customers are one-time buyers. He's tried selling collectible items, but only a tiny part of his customers became regulars. Phil decides to get some external help and finds Emily, a consultant that specialises in predictive analytics. Emily aggregates all available customer data and builds a statistical model for analysis. Exploring the model Emily discovers that customers buying on Mondays correlate strongly with them being repeat buyers. She also notices that some of the best customers come searching for keywords related to being bored. Phil listens to Emily's advice and increases advertising for "bored people" and fine tunes timing. As a result, the pool of repeat purchases grows. Profit and recurring revenue increase while marketing expenses are lowered.

Technology: Label your high-value customers (using your own criteria such as profitability, frequency, time together). Train a classification model to predict which customers are high-value based on external features (cookies, demographics, behaviour, etc). Evaluate prospects and customers with your model and take corresponding actions like upselling.

3. Discover Customer Clusters that Represent Distinct Buying Personas

Why Would You Need This? Use customer personas to find specific customers to target with explicit marketing campaigns. Other applications range from making relevant product recommendations to offering a personalised experience. Behavioural clusters can be used as implicit segmentation when predicting engagement probabilities.

Example: After discovering how to predict who his best customers will be, Phil wants to get more insights about his customers. He agrees on another gig with Emily. Emily uses a clustering algorithm to find groups of his customers with similar "phenotypes". Phil's intuition tells him that people mostly buy his products as presents and amusements, usually to impress friends. After all, his business is about bringing surprise and wonder with unusual gifts. Strangely, the clustering model shows a different picture. A very distinct and profitable customer group is repeat buyers, males, buying on weekdays (again!) with preferences for unique mechanical toys and meme-related souvenirs. Most of them have even left positive reviews and invited new customers! Trying to find out more, Emily uses a geocoding service and finds out that most of these customers have delivery addresses in financial districts. His best customers may be bored office workers! Phill plans completely new marketing campaign targeted at the office guys and contemplates ordering a batch of threbuchets.

Technology: Create several customer clustering models (behavioural, demographic, product-based) using available data. Use the resulting insights to guide mass marketing campaigns. Improve targeting for personal marketing campaigns. Use clustering models when recommending products.

4. Predict Which Prospective, Existing, or Past customers are Most Likely to Buy (Again)

Why Would You Need This? Handle abandoned browse, search and cart cases with higher success and lower costs by targeting only the promising subset of the prospects/customers.

Example: Phil is using re-targeting advertising to target his customers based on their previous actions on his online store. His usual re-targeting strategy is to target customers that abandoned the purchase process at some point. However, Phil has a relatively small re-targeting budget and this budget does not generate a lot of purchases. That's why Phil decides to ask Emily to use her expertise to optimise the use of the re-targeting budget. Emily builds a logistic regression model that lets her distinguish a specific group of people that abandoned during purchase process but their probability to buy in the future is much higher than the average. Further analysis shows that men between thirty and forty years old who visit the online store on weekdays in the first half of the day are almost a certain buyers in the future. Emily's conclusions let Phil reorganise his re-targeting campaign budget. He estimates the number of customers that belong to the specified group (men between thirty and forty years old who visit the store on weekdays in the first half of the day) and spends the largest part of his re-targeting budget on them.

Technology: Create predictive models (separately for prospects, inactive/active customers) using customer data with transactional data as labels. Use models to predict likelihood to buy and increase acquisition/activation efforts where necessary.

5. Predict Which Existing Customers are Least Likely to Buy Again

Why Would You Need This? Use low likelihood-to-buy prediction to keep customers with appropriate incentives (e.g. free shipping, discounts). Low likelihood-to-buy can also be a predictor of churn and trigger proactive churn management techniques.

Example: Phil observes that about half of his registered online store visitors are one-time buyers who never return or return but never buy again. Phil asks for Emily's help to identify this type of visitors in the future in order ti take some actions to avoid churn or encourage more than just an one-time purchase. Emily constructs a random forest regression model using demographical customer data. The model shows that a significant part of one-time buyers are students. Probably they create an account in Phil's store to buy just a specific product, possibly a present. They are only interested in that specific product but not in Phil's online store as a whole. Probably that's why they never come back. Emily advises Phil to apply a new marketing strategy for these customers to gain the maximum possible benefit from one time buyers. Phil starts sending a specific newsletter to them offering large discounts for specific products and occasionally free shipping. This gives a couple of benefits at the same time. Firstly, customers get more information about Phil's offerings and are perhaps encouraged to come back to the store because of these products. Secondly, Phil maximises the profit that could be gained from these customers. While not high-value customers, they still generate some profit.

Technology: Create predictive models similar to the high likelihood-to-buy models with bias (tuning) on low-likelihood-to-buy cases. Use incentives (discounts/free shipping) as a last-ditch effort.

6. Predict Which Products or Content an Existing or Prospective Customer Might Be Interested In

Why Would You Need This? Increased relevancy of recommendations increases value both for the customer and the business. Optimise inventory by selecting more relevant products.

Example: Phil's existing system recommends products that are from a similar category as those viewed or purchased. However, Phil understands that his recommendation system is not perfect. He asks Emily's help to create a new recommendation system that takes customer interests into account. Emily uses demographical information and purchase history to construct a support vector machine model that serves as a base for a new user-user recommendation system. This system shows personalised recommendations according to customer profile information and in-store behaviour. This leads to an enjoyable experience for visitors. Every time they come to Phil's store the main page features previously unnoticed products that are interesting to them. Also, Phil's weekly newsletter now promotes products that were produced using the new recommendation system. Additionally, customers getting better recommendations during the purchase process are encouraged to add more items to their cart which in turn increases the average cart size.

Technology: Use available customer and product data (you can also employ customer clustering data) to create user-user, user-product and product-product recommendation systems. Use relevant product recommendations during shopping and as targeted offers during reactivation/retention campaigns. Apply customer-product prediction models with transactional data to predict future sales. Derive a product's value model for a specific kind of a customer (using a personas model).

7. Predict the Lifetime Value (CLV) of a Prospective or Existing Customer

Why Would You Need This? Discover potential high-value customers among prospects and new customers to apply customer value segmentation early. Reward behaviours indicative of high-value customers to positively affect customer evolution. CLV comparisons can be used to derive Upsell LTV and accelerate customer propagation to higher tiers.

Technology: Create a financial model to calculate lifetime value (CLV) for existing customers. Use the calculated CLV as a label to train prediction model(-s). Apply prediction models to customers at all stages of their customer lifecycle. Use the predicted CLV as input to gauge marketing effort, upsell value, and tiering possibilities.

Example: Emily suggests to Phil to use all the data they have collected in the last couple of years, including purchase size, purchase frequency, and retention rate, to create a metric that shows what value each customer brings to Phil's store. This metric is called Customer Lifetime Value (LTV or CLV). Emily constructs a linear regression model to predict CLV using customer profile information. This model lets Phil project what value each existing or prospective customer will have in the next couple of years and adjust his marketing strategy accordingly. For example, offering discounts to high lifetime for customers to ensure that they continue using Phil's store. It also changes Phil's marketing approach towards new customers. The model states that 30 to 40 year old men generate the highest CLV. Therefore Phil creates a new long term marketing campaign for such customers.

For Bespoke Systems

API Design and Implementation

We will help you define a roadmap for your internal or external company or product APIs. We will then design and implement your APIs to the highest standards ensuring longevity and robustness.

Custom Data Products/Platforms

We will work with your business and technical stakeholders to lay out your business objectives. We will help you formulate your strategy, devise your technology stack, and use proprietary or public data to build internal or external products or platforms, allowing for future growth while minimising risk.