You may already have robust analytics that predict value, churn, bad debt & future customer interests. You may also have a robust customer database that may be used to glean additional insights and make the analytics actionable. That’s great, but how do you determine which offer a customer should receive in all inbound and outbound channels (CSR, Web, direct mail,..)?
Many companies are using analytics to possibly optimize one offer, but what about the full offer mix throughout all the channels? Through analytics it may be known that I may be interested in offer A, D & G but should I be presented all 3 offers? If I am presented all 3 offers will I be over contacted, potentially eroding offer performance?
As an analytic consultant, I find those questions tough, but fun. Several tools are available to partially answer those questions and fully answer the questions in some environments, but before evaluating tools evaluate your environment as it will determine the build versus buy dilemma.
If you do choose to build a custom or semi-custom offer optimization engine below are a few suggestions around an analytic data-driven offer optimization process:
- Use analytics to create profiles of the customer base. For example, high profit / high churn or low tenure / high bad debt. Start small (10 to 15 profiles max) and prioritize the selection process such that each subscriber only fits into one profile.
- One of the profiles needs to be a control group and in many environments that should be the priority1 group.
- Determine an initial set of offers that you would prefer to optimize. Again, start small if you have many offers. Possibly start with an offer category such as retention offers.
- Create a data driven process such that you may periodically calculate offer take rates within each profile and offer cell. An example shown below for one profile only (High Value / High Churn).
- Using the offer priority determined above execute your business rules to determine the eligible offers then the highest priority offer remaining may be used.
We only scratched the surface above so key considerations when building an offer optimization engine should also include:
- Once enough contact history and responses are received, start incorporating response models into the profile creation process.
- Use previous contacts within the responses, as predictive variables (ie. time since last contact, number of contacts the past month,..), to help minimize contact fatigue.
- Many environments need an override process as the best eligible offer statistically may not be best for business reasons.
- Statistical methods should eventually be incorporated, such as design of experiments, to better understand the master control group size for each offer as well as to analyze the results when the number of profile by offer cells start to become large.
- Many tools are available that could satisfy your requirements around offer interface needs and business rule maintenance.
- Reporting needs should be incorporated into the evaluation.
- Continue to evolve and add additional profiles once you are confident with your initial results.