AI Disclosure

If you received a message or used a cancellation page

Businesses use Lokuna to keep their customers happy and subscribed. If one of those businesses is yours to deal with, here is what you should know:

• Some messages are AI-written. Emails, text messages, and the text on the cancellation/retention page may be composed by an AI system acting for the business, in the business’s name.
• Offers are real. Any discount or pause you accept is applied directly to your actual subscription and is honored by the business.
• The AI cannot block you from cancelling. The cancellation page always lets you complete your cancellation; retention offers are optional.
• Offer decisions are automated. Which offer you see (if any) is selected automatically from your subscription details and the reason you give, within limits the business sets, and may take into account how similar offers worked before. Occasionally a different allowed offer is shown at random so that results can be measured fairly. You can ask for human review of an automated decision by contacting the business, or us at info@lokuna.com.
• Opting out works. Unsubscribe links and replying STOP to texts stop those messages.
• Your data. The business remains responsible for your information; Lokuna processes it on the business’s behalf as described in our Privacy Policy.

If you are an operator (a business using Lokuna)

• What the AI writes: email and SMS copy, cancellation-page copy, and reply drafts are generated by large language models accessed through OpenRouter (Anthropic Claude by default, then OpenAI GPT, then Google Gemini, only if the earlier ones are unavailable), run on Google Cloud (Vertex AI), Amazon Web Services (Bedrock) or Microsoft Azure endpoints that OpenRouter lists as zero data retention. Output is grounded in your connected data and brand settings, and filtered by a brand-safety check that blocks guarantees and risky claims.
• What the AI recommends: on the cancellation page, the model recommends the single retention offer (discount, pause, call, or none) most likely to keep that specific customer. It can only pick from the offer types and ranges allowed by your guardrails, and deterministic rules clamp every amount to those limits before anything is applied. If the model is unavailable, a rule-based offer is used instead.
• How it learns: Lokuna records each cancellation-page offer and its outcome (accepted, next invoice paid, still subscribed at day 90). Your own outcomes improve your own offer choices. Across businesses, only aggregated, de-identified statistics are used (at least five businesses and thirty outcomes per statistic, no business above a quarter of it), as starting estimates, never as proof for your business and never visible to anyone. From the first visit, about one visit in ten may be shown a different allowed offer that suits the customer’s stated reason (never a discount for a technical problem or a missing feature), chosen at random, so results are measured fairly. An offer is only dropped once your own outcomes show it clearly works worse. Learning can only narrow the offers your guardrails allow, never widen them.
• What the AI never decides: churn-risk scores, whether a message is sent, sending caps, approval holds, spending limits, and billing are decided by deterministic rules, never by the language model.
• What the agent can do autonomously: send retention emails/SMS within your caps and quiet hours, apply offers within your ceilings, schedule end-of-period cancellations requested by customers, and escalate high-value or uncertain cases to your inbox instead of acting.
• Your controls: approval modes (autopilot, review high-value, review everything), daily and monthly send caps, per-customer frequency limits, offer ceilings, brand-voice and never-say rules, and a pause switch. In autopilot, messages go out without human review — choose the mode that fits your risk tolerance.
• Accuracy: the agent references only data you have connected and degrades gracefully with less data; it is designed not to fabricate customer facts. AI output can still contain mistakes — review modes exist for a reason.
• No training on your data: your data and your customers’ content are not used to train or fine-tune AI models, ours or anyone else’s. The outcome statistics above are counts and rates, not model training.

Questions

Contact info@lokuna.com. See also the Terms of Service (Section 4) and Data Processing Addendum.