|
Background Although the intent of most planned CRM campaigns is very strategic in nature and they are expected to induce stickiness and incremental behaviour with customers in a longer horizon - still, ROI has always been the immediate lens to judge these campaigns. Marketers have so far focussed on following to keep the bottom line of campaigns low: 1. Reduced targeted base 2. Limiting the number of contacts 3. Choice of cost-effective targeting medium (e-mail in most of the cases) 4. Increasing the gaps between contacts 5. Etc. In most of the campaigns, once the target audience and communication is defined contact strategy is generally “blanket applied”. Blanket contact strategy although simple from an implementation standpoint, it fails to recognize the inherent response likelihood of a specific customer given a particular frequency, timing, gap in communication and choice of medium for communication. Advanced uplift models recognize this and not only can they bring the cost of communication drastically down but also boost the overall response to campaigns by further tailoring the communication strategy. Framework An advanced statistical uplift model can be built to gather past learning of a customer / similar customer and identify sub segments that require a specific communication strategy and also the communication elements around it. To illustrate, let’s take a case study of a flash deal agent where the as-is communication strategy is aggressive and 2 emails are sent followed by SMS over a week. Average response rate realized is 15 %. | M | T | W | T | F | S | S | | | email | | email | | SMS | |
Crafting the sub segments Statistical model learns the following from historical data before grouping customers (with healthy response rates) in similar focus buckets: 1. Month, day of the month, time of the day the message was sent 2. Month, day of the month, time of the day the message was opened 3. Month, day of the month, time of the day the message where response was returned 4. Number of contacts that were sent till the response was returned 5. First, intermediate and final media types that were used 6. Gap between various media types With this exercise, customers segment that require just 1 email communication and with a higher likelihood of response if it is sent on Wednesday is identified. Profiling the sub segments Etched-out sub segments that show alignment to a specific communication strategy are profiled, pen pictured and their likely response rates determined. Learning from these sub segments can then be applied to customers with similar profiles but not sufficient experimental historic data. This exercise gives a face to the sub segment which is trackable. For example, customers segments that require just 1 email communication have an annual income of INR 25, 00,000 -35, 00,000; they are in age range of 30-40 years, and have 1 kid. They have on an average 3 credit cards registered and internet is their preferred mode of transactions. Average response rate from these customers was 35%. Reducing the target audience base It also determines the base which can be removed from the campaign communication 1. Customers that are least likely to respond to the campaign 2. Customers where communication budget required to generate the response is too heavy 3. Customers that will anyway show the behaviour without any campaign / communication This exercise eliminates (a) sub segments where either likely response rate is less than 1% (b) sub segments that require over 10 communications over 6 months and still have likely response rate lesser than 3% (c) sub segments that have purchase propensity of 90% without any campaign A Case Study Exhibit A depicts the model result summary for one of the flash deal agents. Key directional impact summary of the model is highlighted below: 1. It lowered the communication budget by over 35% 2. It increased the overall response rate by 70% 3. Identified 15% of the base where any communication was not required
Exhibit A
Challenges
Insufficient historic experimental data Statistical models are dependent on past learning that not only covers exhaustive communication scenarios but also have their depth. Robust design-of-experiment are recommended to marketers to test possible communication strategies and to build a rich historic pool from which 1. Insights can be mined out via advanced statistical models 2. A tailored communication be effected
Key Outcomes 1. Differentiated communication strategy for a given target audience 2. Significantly reduced campaign costs 3. Higher response rates to campaign 4. Key Insights such as profile of sure responders and sure non responders which can be the subject of various deep dive studies 5. Reduced subscription opt-outs
|
Articles 
