Revenue’s
night shift.

An AI billing agent is software that watches every customer account and acts on revenue-affecting moments — low balances, failed payments, usage spikes, churn signals — without a human deciding case by case. Kribana's agent runs on the same metering, wallet and invoicing rails the platform already operates, not a bolted-on integration reading stale data. It watches every account continuously, and the moment a signal crosses a threshold it runs a matching playbook: sending a top-up link, retrying a failed charge with timing suited to why it failed, flagging a churn risk, drafting an expansion offer, warning a customer before bill shock, or tracking a commitment that's running under. The point is closing the gap between when a revenue moment happens and when someone actually responds to it.

On autopilot, the agent can send wallet top-up links when a balance crosses a threshold, retry a failed payment with timing and messaging matched to the decline reason, flag an account trending toward churn to its owner with context, draft an expansion offer when usage runs ahead of a commitment, and warn a customer before a usage-driven bill surprises them — all without a person clicking a button first. Every one of those playbooks can also run in approve-first mode, where the agent prepares the action and a person confirms it, or alert-only mode, where it only notifies a human and takes no action itself. Which mode applies is set per playbook, not platform-wide, so a team can automate the low-risk moments fully while keeping a human in the loop on anything higher-stakes.

These are the three modes available for every playbook the agent runs, chosen independently per playbook rather than as one global setting. Autopilot means the agent executes the action itself the moment its trigger condition is met, with the action and its trigger recorded in an audit log afterward. Approve-first means the agent prepares the action and queues it for a one-click human confirmation before anything happens. Alert-only means the agent only notifies the responsible person that a signal occurred; it takes no action on its own. Teams typically start higher-risk playbooks in alert-only or approve-first mode, watch how the agent behaves, and move a playbook to autopilot once they trust the pattern — rather than handing over full control from day one.

The agent doesn't introduce a new way to charge a card — it operates the same payment processor connection (Stripe, Razorpay, Adyen or Braintree) and the same billing logic Kribana already runs, just with better timing and diagnosis. On autopilot, the actions it takes on payments are retries of charges the customer already owes, timed to the specific decline reason instead of a blind fixed schedule, not new or unexpected charges. Every action it takes — what triggered it, what was sent, and what happened — lands in an audit log, so nothing it does is opaque after the fact. And because each playbook's mode is configurable, a team that wants a human to confirm every retry before it fires can set payment recovery to approve-first instead of autopilot.

Traditional dunning software runs a fixed retry schedule — try again on day one, three, and seven — regardless of why a payment actually failed. Kribana's agent reads the real decline reason before it acts: insufficient funds gets a retry timed around a plausible payday, a bank block gets different timing, and an expired card skips retries entirely in favor of an immediate update-payment-method email, since retrying a dead card just burns attempts and annoys the customer. It also isn't limited to payment recovery — the same agent watches wallet balances, usage against commitments, and churn signals, and runs the matching playbook for each, all on the same rails as the rest of the billing platform rather than as a separate dunning tool bolted on top.