Retention Offers Crowd Out High-Value Upgrades
A national telecom with 6.2 million prepaid and postpaid subscribers runs Pega Customer Decision Hub for inbound care and outbound SMS. The business has five propositions: a low-cost data top-up, a plan upsell, a retention discount, a SIM-only plan, and an add-on. The current NBA Designer has one issue, Retention, with one group, Base, containing all five actions. Suitability suppresses any action with propensity below 0.30. Arbitration selects the action with the highest propensity. In the last quarter, 78% of all outbound offers were the data top-up; the retention discount, which protects about $4.8M in annual margin, was offered only 3% of the time. High-value customers with churn scores above 0.65 received the top-up in 61% of cases. Contact policy limits customers to two outbound offers per seven days. The business wants to protect margin without increasing contact volume. Analytics has customer value bands and a churn model available as decision data.
1. Explain why the current design produces this outcome and identify the decisioning layer where business value should be introduced. Why is raising the propensity threshold not the right fix?
Read the worked answer
The current design puts all five actions in one group and one issue, and arbitration picks the highest propensity. The data top-up appeals broadly, so its propensity is high across the base. The retention discount appeals to a narrower at-risk segment, so its propensity is lower, and the 0.30 suitability floor often removes it before arbitration. The system is maximizing propensity, not business value.
Business value belongs in arbitration, not in eligibility or suitability. Attach expected margin or a strategic weight to each action and use a weighted score such as propensity × expected margin × weight in the Prioritize component, or use NBA matrix weights. Raising the top-up threshold to 0.50 would reduce offer volume and create more no-offer outcomes; it would not make the retention offer competitive. The threshold should suppress genuinely low-quality actions, while arbitration handles trade-offs.
2. Design a revised NBA structure that protects margin and respects the two-per-seven-days cap. Include issue/group/action structure, arbitration logic, and where the churn model and value band are used.
Read the worked answer
Create two issues: Retention and Growth. Under Retention, create groups High-value at risk, High-value stable, and Standard. Use customer value band and churn score from decision data in group eligibility: High-value at risk requires value band A or B and churn score above 0.65. Place the retention discount only in High-value at risk, and place the data top-up in Standard and High-value stable.
Use a low suitability floor for the retention discount, such as 0.15, so it is not removed before arbitration. In arbitration, use a weighted score: retention discount weight 3.0, top-up weight 1.0, and score = propensity × expected margin × weight. Keep contact policy at the engagement-policy layer with a two-per-seven-days cap, and add a separate service category so mandatory messages do not consume marketing contacts. This structure lets the retention action win where it matters, while the top-up still serves low-risk customers. Monitor offer distribution by value band and incremental margin, not just propensity.
3. The business asks for a guarantee that every customer with churn score above 0.65 receives the retention discount. What would you advise?
Read the worked answer
A guarantee conflicts with contact caps and with other eligibility rules. First clarify whether the requirement is eligible for consideration or must be delivered. If it must be delivered, model it as a mandatory service or retention action in a dedicated issue with priority and an explicit contact-policy exception. Otherwise, use a high weight and group-level eligibility so it competes strongly but can lose to a higher-value or compliance action.
Also verify that the churn score is available at decision time; a nightly score can be stale. Test with simulation to measure offer volume and contact-cap breach rate. A blanket guarantee would likely cause contact fatigue and regulatory issues. The defensible design is a prioritized segment with measurement and an exception process.