How the negotiation engine decides when to offer a discount

The most common question we get during demos: "does the agent offer discounts to everyone?"

No. And that's exactly the point.

Offering universal discounts is the most expensive strategy possible: you give up margin for debtors who would have paid without one, and you still don't recover from those who needed a different approach. Dyvit's negotiation engine was built to avoid both mistakes.

The variables that feed the model

Before the agent sends its first message, the system has already calculated a payment propensity score for that contract. This score combines four groups of variables:

• 01
  Debt maturity

  Contracts 0-30 days overdue have a high propensity to pay without a discount. Forgetfulness and short-term cash flow issues are the most common causes. Discounts in this segment cost margin unnecessarily. Contracts 90+ days overdue enter a different dynamic: the debt has already been "psychologically processed" by the debtor as a loss, and the discount works as an anchor for re-engagement.
• 02
  Contract value and customer LTV

  For a high-LTV customer (history of previous payments, other active products), the agent prioritizes recovering the relationship. The tone is solution-oriented, not collection-oriented. For low-value contracts with debtors who have no prior history, operational efficiency weighs more. The agent is more direct and objective.
• 03
  Conversation behavior

  The agent interprets signals in real time: the debtor responded quickly (positive engagement signal), asked for more time (openness to installments), mentioned a specific financial hardship (trigger for a grace period offer). Each signal recalibrates the next message.
• 04
  Creditor rules (what you configured)

  The agent never offers terms beyond what the creditor authorized. Discount cap, available installment plans, contact windows, escalation criteria: all of this is configured by you. The propensity model decides when and how to use the space you opened, never beyond it.

The discount decision logic in three paths

Given the propensity score and the conversation context, the engine makes one of three decisions:

Installments: when and how the agent proposes them

Discounts and installments are not equivalent. A discount reduces the total recovered amount. Installments preserve the full value, distribute the cash flow, and significantly increase conversion on medium and high-ticket contracts.

The agent proposes installments when it detects two signals: the debtor has shown intent to pay but resists the total amount, and the contract falls within the ticket range where installments are viable for the creditor (configurable by you).

The installment plan is generated dynamically: the agent offers the most suitable option for the profile, not a menu of choices. "Would you prefer to pay in 3 installments of R$280 or settle everything now with a 10% discount?". The binary choice has a 34% higher conversion rate than menus with multiple options.

What the agent never does

Just as important as what the agent does is what it cannot do. By design, not by instruction:

It never offers terms beyond what's authorized. If the discount cap is 15%, the agent will not offer 16% under any circumstances, even if the debtor insists.

It never contacts outside configured windows. Consumer protection law establishes time-of-day limits. The agent respects this absolutely. There is no human oversight that could accidentally break the rule.

It never continues after an opt-out. A debtor who responds "don't contact me again" has contact stopped immediately and the record permanently flagged.