Approved knowledge only
Every answer grounded in the business map you approve before launch.
03/5
Guardrails
It learns. It never freelances.
Learning changes recommendations, scores, and proposed content — never your pricing, offers, claims, or qualification rules. Material changes stay reviewable, traceable, and reversible, and buyers are told up front they are speaking with an automated assistant.
Confidence threshold
Low confidence triggers a handoff — never a made-up answer.
Approval on what matters
Claims, offers, pricing, and qualification rules change only with your sign-off.
Evidence behind every suggestion
Each recommendation keeps the interactions and outcomes that supported it.
Every ad dollar gated
Campaigns launch only with your approval — platform, budget cap, time window, and stop conditions spelled out in advance. Budgets never increase on their own.
Channel-safe by design
Buyer-side assistants reach the same approved knowledge through a permissioned channel — no browser-only actions, no unapproved claims.
How the boundaries work in practice
What it will never do
It never contacts anyone who has not reached out first. Outbound cold outreach is not in scope.
It never invents prices, offers, claims, or qualification rules. Those change only with your sign-off — learning may propose updates, but nothing material goes live without approval.
What happens when it does not know
It says your team will respond, captures the question with full context, and never guesses — no invented service, pricing, legal, insurance, or policy claims.
Low confidence is a handoff trigger, not a cue to improvise. The unanswered question becomes part of what the system learns from next.
What the buyer sees
You’re speaking with an automated assistant, not a person. Tell us how we can help, and we’ll get you to the right person.
That disclosure is shown up front. The same approved knowledge and boundaries apply to supported website engagement. Instagram support is assessed separately before launch and is not included by default.
What you can undo
Material changes are reviewable, traceable, and reversible. If a recommendation, score change, or proposed content update is wrong, you can roll it back — the trail of what changed and why stays available.
Who can see what
Administrative access to production systems is restricted, authenticated, and logged. Customer content is handled with contracts and technical controls appropriate to a subscription software provider, with retention configured in line with your policies.