The client is a digital marketing company at the forefront of lead generation in South Africa.

They provide safe, reliable and convenient channels for individuals to find and buy

products and services best suited to their unique lifestyles. They make use of cutting-edge

digital marketing tools that produces satisfaction for both their clients and the end-user

making use of the products on offer.

The Problem

The client makes use of a sophisticated rules-based system to decide which end-user should receive

which marketing message based on POPIA and other legal requirements. The vast proportion of

marketing messages sent to consumers do not generate a lead, however there is a cost associated

with sending out a message, irrespective of whether a lead is generated from that consumer.

Revenue is only generated once an end-user subscribes to a service being marketed to them.

The client would like to reduce the number of marketing messages being sent to customers without a

lead being generated, hence saving costs. However, at the same time, ensuring that they do not miss

out of any potential leads by not sending marketing messages to an end-user who would have

subscribed to a service.

‍

Our Solution

Elucidate developed a series of models to accurately identify whether a scheduled marketing

message is likely to generate a lead using sophisticated tailored gradient-boosting techniques and

feature engineering on all known campaign and end-user characteristics. The system learns and

improves over time and is becoming more effective as it sees more campaigns. The system is able

to respond in real-time and provide reliable results far surpassing human performance.

‍

Our machine learning algorithms far surpass the human-intense methods currently employed. The

system is automatic, operates in real-time and most importantly improves over time by learning

from all campaign data.

The Results

1. Reduce cost per lead (cpl) by 30%

2. Increase profit by 10%

3. Less than 0.3 seconds response time for decisioning

‍