How teams replace spreadsheets with billing that runs itself.

A fixed retry schedule treats every decline the same, recovering under 35% of failed charges and generating support tickets from confused customers. A card declined for insufficient funds and a card declined because it expired need completely different responses, but a blind retry schedule sends the same attempts at the same intervals regardless of why the charge failed in the first place.

Support ended up carrying the weight of a billing problem — a customer whose card had simply expired would get retried against the same dead card before anyone reached out, by which point the account had usually already lapsed and the customer was annoyed rather than helpful.

Kribana's agent reads the decline reason and matches it to a recovery playbook — a retry timed around payday for insufficient funds, an immediate update-method email for an expired card, and a different path again for a card flagged as lost or stolen. The playbook runs in autopilot, approve-first or alert-only mode depending on how much the team wants automated versus reviewed.

Because the response is matched to the actual reason a payment failed instead of a generic retry timer, most recoveries happen without a support ticket ever being opened — the case is resolved before a human account owner needs to look at it.

The recovery rate almost doubled without the team writing a line of dunning logic themselves, and the support tickets tied to failed payments dropped along with it, since most declines now resolve through the right playbook instead of escalating to a person. For an observability company billing recurring usage, that's involuntary churn recovered automatically rather than involuntary churn that shows up as a support queue.

The team also stopped treating dunning as a settings screen they'd configured once and never revisited. Because the playbooks are matched to decline reason rather than a fixed schedule, adjusting recovery behavior for a new payment method or a new region is a configuration change, not a rewrite — a distinction that mattered once the customer base stopped being concentrated in a single country.

Alert-only mode gave the team a way to trust the system before handing it full autopilot. Early on, recovery actions surfaced as recommendations for a human to approve; once the playbook's calls consistently matched what a person would have chosen anyway, switching specific decline reasons over to autopilot was a deliberate, low-risk decision rather than a leap of faith.

“The recovery rate almost doubled without us writing a line of dunning logic ourselves.”