The article considers the tariff as a controlled system of cash flow, where each level of the package affects revenue per subscriber (ARPU, average revenue per user), contribution margin and lifetime value (LTV, lifetime value). A framework is proposed that connects the “cleaned” gigabyte economy, behavioral models (cohorts, survival) and causal experiments to distinguish cannibalization from real inflow. It is shown that a “quiet” plan with a guaranteed base speed, a night window and speed limitation protect the margin better than distributing additional gigabytes. Personalization of retention is effective only when tied to the identified customer problem and strict limits on individual discounts. For content bundles, the key is payment “upon consumption”. The result is a portfolio of 4-5 levels with “handrails” that prevents self-stratification and reduces churn without overheating the network. Success metrics are an increase in the contribution margin per gigabyte, stable ARPU and a decrease in the share of downgrades.