Customer Targeting


Customer Targeting

Customer targeting using predictive models facilitates increasing the success rate of marketing and CRM campaigns and improving customer experience.

Targeting customers using predictive models can increase the efficiency of your campaigns. They help concentrate your energy and resources on customers who are more willing to buy a specific product or service.


Predictive models created using statistical methods are able to identify customers with higher willingness to buy (Propensity to Buy). Based on the use of these models in cross-sell and up-sell campaigns, priority lists can be created that the company can use to concentrate its resources on the most valuable customers, for example. Using the Net Lift approach (incremental response), we ensure that client targeting has a visible impact and that marketing costs are spent efficiently.

By deploying a Propensity to Buy model on a pension fund product we increased the conversion of a campaign of a medium-sized bank by 48 %.

Benefits for the company

  • Increases the success rate (conversion) of cross-sell and up-sell campaigns;
  • Increases the profitability of the organisation and marketing activities;
  • Improves the quality of customer experience by optimising offers for customers in a way that reflects their current needs;
  • Allows more efficient allocation and capacity use of communication channels.


Our starting point is the CRISP-DM methodology. First, we define the business problem, select models that will solve the problem, specify their inputs and outputs and determine their role in the business process. Then we focus on data that has potential for production operation of Propensity to Buy and we build an “ETL process” (extraction, transformation, load). The underlying data is transformed into the form of the so-called predictors, i.e. quantities aggregated at customer level for a given moment in time. The key variable is the “target” – indicator if the customer did or did not buy the relevant product. The table with predictors and the target serves as an input for the modelling stage, where we use advanced analytics algorithms (see below) to develop a model, i.e. predictors and a functional relationship, that explains the target as best as possible. We iterate the model several times and assess its performance, stability and inner logic. In the last stage, we implement the model in production, set up reporting, integrate in the business process and thus close the circle.

Areas of use of Customer Targeting

Uplift modeling

Do you feel that your PtB scores high almost always the same customers? That the same group is always hit by campaigns and the rest of the portfolio is unexploited? Then you probably do not have Uplift PtB models. The uplift concept (net lift, incremental response) is based on the assumption that our target is not the customers who will buy the product, but the customers who will buy the product as a result of a marketing contact. The campaigns then purposefully avoid the segments “sure thing” (will buy even without being contacted”) and “do not disturb” (contacting decreases the likelihood of purchase). Uplift modelling increases marketing ROI, but it requires a very high quality of CRM data on customer interaction.


Target is a variable at customer level that the PtB model tries to explain. Example: the customer has a new credit card in the month of M+2 that has been activated (M is the current month). The correct definition of the target is key for the business use of the PtB model and also fundamentally determines its performance. We have already seen several models where the incorrect definition of the target led the business to refuse the model, even though the model itself was crafted flawlessly. We therefore give appropriate care to target definition, which has paid off many times already.

Machine learning

Customers often ask us: and what model do you use? The answer is not easy. We use a relatively wide range of models, because each problem requires a different model and there is not a universal one. This is confirmed by our experience and it has even been scientifically proven (refer to the “no free lunch theorem”). We often solve the dilemma of model performance and transparency – the highest performing models may be so complex that they cannot be validated from a business point of view. We decide according to the use case model together with the user. Our favourite binary classifiers include logistic regression, decision trees, random forest, gradient boosting, SVM.

Correct use of the Propensity to Buy models

There are many models that ended up closed in a drawer, because a good model on its own does not guarantee good customer targeting – it needs to be used correctly. Our team therefore includes a campaign specialist who is able to connect the result of the model to the business and thus elevate our customer targeting to an end to end solution unique on the market. This change provoked an unexpectedly positive response from our clients and we are now more secure in the knowledge that we bring our clients real added value.

Predictor Factory

Our development of PtB models is quick and efficient thanks to Predictor Factory. Predictor Factory is a specialised software for the preparation of predictors for modelling. It works with a database that contains input data in many interconnected tables. For a particular table with observations and a calculated target (the target table), it calculates all the possible predictors that can be calculated based on the other tables (based on an expandable set of patterns) and selects a reasonable number (approx. 1,000) of the best ones for modelling. In comparison with the traditional preparation of predictors (manual SQL coding), predictor preparation using Predictor Factory is:

  • Faster (saving approx. 20 man days per model);
  • Of better quality (the resulting model performs better);
  • Flawless (no coding errors);
  • Documented (predictor catalogue including the SQL code); and
  • Flexible (if input data changes, it simply launches again).

Predictor Factory is a unique technology, we have so far not encountered a similar tool of a comparable quality in the world. We have developed Predictor Factory in cooperation with the Faculty of Information Technology of the Czech Technical University in Prague.

Real time marketing

Today’s highest goal for marketers is real time marketing. This attractive but at the same time feared discipline is now a solution that we have tested and can recommend. This demanding project requires the involvement and coordination of technology experts, data specialists, programmers, testers, business users, modellers and marketers. But the result is worth it. Real time communication and orchestration of many channels is something that will perfectly set you apart from the competition.


Filip Trojan

Filip Trojan

Senior Manager

Filip is a Senior Manager in the Advanced Analytics deparment. He has over 15 years of experience in analytics, machine learning, mathematical optimisation and data science. He has an extensive variet... More

Veronika Počerová

Veronika Počerová


Veronika is a manager in the Advanced Analytics department. She specialises mainly in analytical end-to-end solutions for clients from the finance, energy and retail industries. She focuses on predict... More