Technology
The Decision Layer
The decision layer is where Swivel turns the ontology into judgment, scoring who is at risk, understanding why, and deciding which move fits which customer. It runs continuously as new events land in the model.
Risk, measured against the customer's own baseline
Every customer carries a churn-risk state that updates as new events arrive in the ontology. Risk is judged against each customer's own history rather than a population average. A customer whose usage fell from twenty hours a week to ten is flagged as declining, even though ten hours is high in absolute terms, because the System is watching the customer's trajectory, not a global threshold.
Separating habit from event
The decision layer decomposes each customer's engagement into a steady baseline and an event-driven component, distinguishing the customer who engages year-round from the one who subscribed for a single season, launch, or event. This is what lets a retention team see a churn risk weeks before the triggering moment ends, instead of after the cancellation arrives. It is one of the clearest things the System surfaces that a conventional dashboard cannot.
Behavioral archetypes
Our agent teams continuously research and analyze your at-risk customers, identifying complex behavioral patterns to group them into highly accurate archetypes. Instead of relying on static algorithms, this intensive, agent-led research ensures segments form around actual human behavior. This gives the System a stable foundation to reason from when deploying a strategy, while giving your team a clear vocabulary for what's happening.
How a strategy is chosen
For a given at-risk customer or segment, a multi-stage pipeline proposes and refines the move:
- A strategist stage drafts candidate strategies for the situation.
- A compiler stage turns each into a concrete offer and message.
- A ranker stage scores the candidates against fit and constraints.
- A composer stage writes the customer-facing copy.
Model-based validation runs before any candidate is surfaced, and nothing reaches a customer without passing governance (next section). The decision layer's job ends at a recommended move with the reasoning behind it; whether and how that move executes is governed separately.
Let the agents prove themselves on your accounts.
In three weeks, see the saves they worked, the reasoning behind each action, and what happened before you decide what comes next.