Written to be used against us as much as by us. The paper separates the category into contact-centre software being repositioned for stores, camera-and-footfall analytics being described as conversation data, and systems that actually score in-store speech, and explains why the distinction changes both the legal review and the achievable coverage. It provides the questions that expose a weak product: whether any voice embedding is created at any stage including model training, whether capture can run without a visible indicator, whether the associate sees what the manager sees, what the dispute path is when a score is wrong, how coverage is reported, and what the vendor requires in order to prove a revenue result. It covers the two budget lines these programmes are typically funded from, the co-funding routes available in watches, beauty and automotive, and the failure modes we see most often — no point-of-sale access, a taxonomy imported from another market, and a manager dashboard shipped before an associate view. Includes a scoring sheet for vendor evaluation.