Finding Canadian-Made Products with AI

Supporting local manufacturing is one of those preferences that polls extremely well and converts poorly, and the reason is almost entirely mechanical.

People will tell you they want to buy Canadian. Then they open a marketplace, search, and buy from page one. Nothing on page one told them where anything came from, and finding out would have taken more effort than the preference was worth in that moment.

That is not a values problem. It is an information architecture problem, and it is fixable.

Why local sellers end up buried

Marketplace ranking rewards sales volume, review count, fulfilment speed and advertising spend. A small Canadian maker competes badly on every one of those, not because the product is worse but because the signals ranking uses are proxies for scale.

A domestic producer with four hundred sales and genuinely better manufacturing loses to an importer with forty thousand sales and an ad budget. The algorithm is working as designed. It was simply never designed to surface origin.

So the products exist, people say they want them, and the two never meet.

Verification is the actual hard problem

The instinct is to treat this as a search problem. It is not, because the underlying data is unreliable.

Origin claims on product listings range from strictly accurate to purely decorative. A maple leaf on packaging means nothing on its own. Designed in Canada means the design happened here and the manufacturing did not. Assembled in Canada means components arrived from elsewhere. All of these appear alongside genuine domestic manufacturing and look similar to a shopper scanning quickly.

There are also real legal definitions. The Competition Bureau distinguishes Product of Canada from Made in Canada, with different thresholds for Canadian content and where the last substantial transformation occurred. Most casual usage corresponds to neither.

Doing this properly means analysing product data, supplier records and manufacturer information, classifying against a consistent standard, and being explicit about the basis. Which is a data pipeline, not a search box.

Show the basis, not just the badge

The design decision that determines whether people trust a platform like this is whether it explains itself.

A verified badge with no explanation is just another claim, and shoppers have learned to discount claims. Showing what is known and how it was determined lets someone decide for themselves whether it meets their bar, which for some people is strict and for others is loose.

This also makes the platform honest about its own limits. Some products are partly domestic. Some have supply chains nobody can fully trace. Saying so is more useful than forcing a binary and being wrong at the edges.

The filter is the product

All of the verification work only matters if it reaches someone at the moment they are choosing.

Origin as an article people read is interesting. Origin as a filter on a results page, while someone is comparing three options, is behaviour changing, because acting on the preference now costs nothing. That is the entire difference between a nice idea and a working one.

It is also why this cannot be a directory. A list of Canadian makers requires the shopper to already be looking for one. A filter meets them inside the decision they were already making.

Why this is worth building

Supply chain disruption over the last few years made a lot of businesses and households notice how far their goods travel and how fragile that is. The interest is real and durable.

What has been missing is the practical means to act on it without extra effort. That is a solvable software problem, and it is a good example of AI being useful in an unglamorous way: classification and data cleaning at a scale nobody would do by hand, in service of a filter that takes one click.