Churn and Next-Best-Action Evaluation
A churn model estimates which customers may leave under current conditions. It does not show which customer can be retained, which action will change that outcome or whether the retained relationship will create enough value to justify the intervention.
Marketways separates four questions: who is at risk, what may have caused that risk, what the relationship is worth and which feasible action creates an incremental benefit. This separation prevents a high churn score from becoming an automatic discount.
Define churn and the decision horizon
Customers can close an account, stop using a service, move one product or reduce their spending without formally leaving. Each event represents a different commercial problem.
Marketways defines the churn event, observation date, outcome period and unit of decision before modelling begins. We also identify when management must act. A prediction delivered after the customer has committed to another provider cannot support retention.
Use risk to direct attention
Usage, product holdings, payment, network experience, service contact and tenure can identify patterns that precede departure. The resulting probability helps the organisation decide where to investigate.
However, risk does not identify a suitable treatment. A customer affected by repeated network failure may not respond to a discount. Another customer may accept the same discount despite having intended to stay. Applying one offer to every high-risk customer can waste margin without changing retention.
Diagnose the condition that management can change
A customer may reduce usage because of price, service failure, product fit, relocation, a competitor or a temporary change in need. Similar behavioural signals can therefore arise from different causes.
Marketways connects behavioural data with complaints, incidents, billing, product and research evidence. We treat the diagnosis as a testable explanation rather than a label inferred from correlation. The analysis identifies which conditions the provider can address and which remain outside its control.
Estimate the incremental effect of each action
Management needs to know what will happen because the provider takes an action. That effect cannot be observed by comparing contacted customers with uncontacted customers when staff selected the contact group using earlier risk or value judgements.
Randomised experiments create a defensible comparison between an action and an alternative. Where experimentation is limited, causal and uplift methods require explicit assumptions about treatment assignment and comparable customers. Marketways defines the offer, eligibility, outcome and measurement period before estimating heterogeneous effects.
Distinguish customers by likely response
Some customers would stay without an offer. Some may leave regardless of the intervention. Others can be retained, while a poorly timed contact may reduce trust or encourage a customer to demand a concession.
Uplift analysis estimates how the action changes the outcome for relevant customer groups. This distinction directs scarce retention capacity towards relationships whose outcome can change. It also protects customers who should receive service correction rather than a commercial incentive.
Calculate relationship value after the intervention
Current revenue can overstate the value of a relationship that requires a large discount, repeated technical work or expensive support. A retained customer may also change products or usage after accepting the intervention.
Marketways estimates expected margin, likely duration, treatment cost, service cost and wider product relationship under each policy. We include the cost of contacting customers who would have stayed and the cost of remedies that fail. The objective is incremental relationship value, not the number of accepted offers.
Represent operational capacity and eligibility
The most effective action may require a specialist technical investigation, a plan change or a service-recovery team. These resources cannot reach every high-risk customer at once.
Marketways includes eligibility, consent, channel, contact limits and operational capacity in the decision policy. The policy may prioritise one customer for technical correction, another for a plan review and a third for no contact. This makes the recommended action executable rather than theoretically attractive.
Evaluate the policy rather than the prediction alone
A churn model can improve its ranking accuracy without improving the retention programme. Management must measure the customers selected, actions assigned, outcomes caused and value created.
Marketways evaluates incremental retention, margin, treatment cost, complaints, opt-outs, service outcomes and capacity use. We compare results across products, channels and customer conditions. A policy succeeds only when it changes outcomes that matter after all material costs.
Use Agentic AI to execute an approved policy
An Agentic AI workflow can retrieve authorised customer and service evidence, apply an approved policy, prepare the permitted communication and record the action and outcome. It can also escalate conflicting evidence or an exception to an employee.
Marketways limits the tools, data and offers available to the agent. The agent cannot invent commercial policy, infer consent, change contractual terms or conceal uncertainty. Management retains authority over eligibility, offers and exceptions.
Monitor treatment effects as conditions change
Competitor offers, network quality, prices and customer expectations change over time. An action that once improved retention may later have little effect or may attract customers who would have stayed without it.
Marketways monitors risk calibration, treatment mix, incremental response and value. We preserve control evidence where practical and retest material policy changes. Declining uplift triggers investigation before the organisation increases contact or discount expenditure.
A service failure illustrates the distinction
Suppose a high-value customer reduces usage after repeated network disruption. A churn model assigns high risk, and a conventional campaign recommends a discount.
The complete policy would first connect the usage change to the unresolved service condition. If evidence shows that technical correction improves retention, the customer should receive that remedy before a price concession. The provider measures the outcome against a suitable comparison and includes technical cost in the value calculation.
How Marketways designs the retention decision
Marketways combines Customer Satisfaction and Experience Research, Operational Performance Diagnostic and Causal Inference and Experimentation. We separate churn risk, cause, relationship value and incremental treatment response, then incorporate capacity and authority.
The result tells management which relationships require attention, which action can change the outcome and whether the expected benefit justifies the intervention.