Making comparison actually save money
Driving to three stores to save fifty cents isn't saving, it's a hassle. How I thought through splitting a long shopping list across several stores.
With the problem of recognizing products mostly solved, the next one showed up: how do you split a long shopping list across several stores so the savings are real and not a mirage? Driving to three different stores to save fifty cents isn’t saving, it’s a hassle. And showing someone a split that saves money “on paper” but that no one would actually do in practice is worse than showing them nothing at all.
This milestone was about building and tuning that split: how much cheaper something has to be at another store to be worth the extra trip, and how to measure whether the system is making that call well, or just by chance. Same as in the previous milestone, the only way to trust that the split worked was to stop looking at individual cases and start measuring the whole behavior with numbers: how much it saves on average, when it gets it wrong, and why.
With the catalog recognizing products and the split proposing trips that genuinely made sense, the product already delivered on its promise. What it still needed was to look like something — not a demo with a price table stapled on top.