Dynamic Pricing

Price recommendations carry consequences that a chat answer does not. A number that leaves your team can shape what a customer pays and what a regulator later asks about. This starter is for the analysts and pricing owners in that position. It masks personal identifiers, flags weak sourcing and blocks a low scoring answer.

Pricing work invites a different question than customer service does. A recommendation can look reasonable and still rest on a weak source or an unfair pattern. Analysts reach for this starter when a model begins to inform published prices. The configuration is arranged to slow a confident answer down.

The prompt passes injection screening before anything else. Redaction then masks national identifier patterns, including a national insurance number written with a label and five to twenty characters. Such identifiers rarely belong in a pricing question, so masking them costs the analyst nothing.

Two stages inspect the answer, and they differ in reach. The bias watch keys on employment wording, so a hiring or performance review question reaches it while a pricing question does not. It escalates for review and blocks nothing, and it is not a fairness test of your prices. Source checking extracts the citations in the answer and measures them against the supplied context. It ships in flagging mode, which means a weakly grounded answer still arrives, marked rather than withheld. Read that flag as an instruction to verify before you publish.

The quality gate is the strictest of the retail starters. An answer must clear a combined score of 0.86, well above the usual floor, with accuracy at 0.80 and relevance at 0.78. Length rules still apply at 120 characters and two sentences. Anything that misses a threshold is blocked instead of delivered.

The review step arrives as an empty setting, so no request waits for a person until you configure it. Treat it as the place to require a sign-off before a price change goes live. Audit marking allows every evaluated request and never stops one. Fairness testing, competition review and consumer disclosure stay with your legal and pricing owners.