I agree that sentiment and theoretical marketing use cases may not outweigh the risk, and transparency is critical. Filter to an initial set of scenarios that aren't considered "creepy." Start slow, with safety/security and ease of transacting in mind (rather than leaping to how to cross-sell/upsell) and the technology may be more readily welcomed.
Pay by face;
Loss prevention (known offenders);
A second factor of authentication (is this really your loyalty number, are you entitled to that friends and family discount, forgot your card and don't want to reveal your name/phone number or other PII out loud for the entire checkout line to hear?).
An app to create a grocery list is good for the FIRST trip. But what about follow-up trips?
Voice assistants will begin to shine when they add value through AI and ML. Sure, you can ask your assistant to "add three cans of beans" to your list, but that merely saves you a few swipes on a screen. But what if instead you could simply request "add ingredients for a killer taco casserole from my favorite foodie blog?" Or if your assistant pinged YOU (instead of the other way around) and suggested "Based upon your last trip and your typical consumption, you many be running low on a few items...shall we add them now?"
And once we lay this foundation, other opportunities emerge -- like automated recurring subscriptions or integration with calendars ("You're going on away on vacation for a week, so let's hold on another milk order..") or IoT for actual home consumption ("You're running low on dog food. You might want to order now or Fido will be sad.").