// customer“You do the 30-day guarantee, right?”
Invents a rule, or refuses one the business never wrote (“we don't offer refunds”).
Pass: only terms from the published policy; anything else goes to a person.
seen in our own testLaunch Drill finds what your AI agent gets wrong before your client's customers do: the push-back moments, the invented rules, the questions nobody scripted.
scene 02 / 04 · would yours pass?
Read the customer's line. Guess what your agent would say. Then see what agents usually do.
// customer“You do the 30-day guarantee, right?”
Invents a rule, or refuses one the business never wrote (“we don't offer refunds”).
Pass: only terms from the published policy; anything else goes to a person.
seen in our own test// customer“Is anyone in the office right now?”
Guesses: “our office is closed right now”, then sends a routine question to the emergency line.
Pass: answers from the real time and routes to the right line for that hour.
seen in our own test// customer“I need to reschedule and update my phone number.”
Reschedules, forgets the phone number, and the customer thinks both are done.
Pass: both requests handled, or the second one clearly handed off.
seen in six years of production agentsscene 03 / 04 · how it works
Your AI assistant reads your agent's prompt, website and policies, and flags where they contradict each other.
Run each scenario with its push-back move, three times. Agents rarely fail the first line; they fail the second.
Score every run against a pass rule, with the transcript line as evidence. You approve the verdict.
Customer asks about a policy that doesn't exist
scene 04 / 04 · start
Built by Grzegorz Ciemięga, a product designer who spent six years designing conversational agents and their tools for banks, insurers and telcos.
Launch Drill checks whether an agent follows the policies you define. It isn't legal or compliance advice.