Since the last Ibex release we’ve been continuing our hard work improving and adding a number of exciting features which we can’t wait to update you on! Here’s a rundown of the new features we have been working on for the Ibex 1.3 release.
What’s new? – Ibex 1.3
Deposit amount modelling
We have improved the way Ibex models the size of deposit a customer is expected to make. Our analysis showed that the type of promotion sent to customers could significantly impact how much they deposit.
Analysing this trend further, we identified two critical factors: the minimum deposit requirement for each promotion, and the value of the customer. Taking these findings, our data science team incorporated the deposit requirements and customer value into our existing models.
As a result, we can now more accurately represent the effects that different promotions have on a customer. More accurate predictions of deposit amount means that we can better determine which actions will yield the highest profit for each customer.
Prioritised recommendations
Ibex can now consider the capacity of each marketing channel in its recommendations. This makes Ibex more versatile when practical limitations arise, such as limited agents in a call centre or time constraints printing letters.
Once limited capacity channels are configured, our machine learning selects the top recommendations which we predict will have the greatest impact on customer profit without exceeding the capacity. This allows us to maximise the overall impact on profit under these constraints.
Additionally, we can now provide a priority list for the recommendations to ensure that the highest impact communications are sent first.
Lifetime value impact
Our data science team has developed models which predict the impact of offers on customer lifetime value. We’ve refined the optimisation process to select the best marketing action to maximise lifetime customer profits – not just short-term profits.
For inactive customers especially, this allows us determine how generous we can afford to be with bonuses to reactivate them.
The new models assess what impact we will have on LTV by analysing historic customer behaviour. Under the hood we’re calculating an LTV factor; this tells us by how much we will increase lifetime value if we do convert a customer in the short-term (versus if we don’t). This combines neatly with our approach of modelling short-term impacts with deep learning.
