How to Monitor Competitor Grocery Prices in Canada

Start here

Paste this into the Canadian Grocery Price Comparison API and hit Start. It watches one category and one product line across five banners spanning two parent companies.

{
  "queries": ["oat milk", "greek yogurt", "ground coffee"],
  "categories": ["dairy-eggs"],
  "location": "Toronto, ON",
  "retailers": ["loblaws", "nofrills", "superstore", "saveonfoods", "pricesmart"],
  "outputMode": "intel"
}

curl -X POST \
  "https://api.apify.com/v2/acts/sunny_eternity~canada-grocery-price-comparison/run-sync-get-dataset-items?token=YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"queries":["oat milk"],"categories":["dairy-eggs"],"location":"Toronto, ON","outputMode":"intel"}'

intel takes queries, categories, or both. Queries answer what a named product costs at each banner. Categories answer what a whole aisle looks like, which is the half most monitoring setups skip.

What intel mode adds

A default run returns comparison rows. intel returns those plus four record types, each tagged with _type, so branch on that field before you do anything else with a dataset.

Read intel_run_summary first, every time. A category gap and a retailer failure look identical in the data and completely different in that record.

Price position by banner

The question a monitoring setup exists to answer: where does each banner sit against the others on the products you care about. query_summary gives you the cheapest per query; the comparison rows underneath give you the whole distribution.

import pandas as pd
from apify_client import ApifyClient

client = ApifyClient("YOUR_API_TOKEN")
run = client.actor("sunny_eternity/canada-grocery-price-comparison").call(run_input={
    "queries": ["oat milk", "greek yogurt", "ground coffee"],
    "location": "Toronto, ON",
    "retailers": ["loblaws", "nofrills", "superstore", "saveonfoods", "pricesmart"],
    "outputMode": "intel",
})

rows = list(client.dataset(run["defaultDatasetId"]).iterate_items())
df = pd.DataFrame([r for r in rows if r["_type"] == "comparison"])

# Index each banner against the cheapest offer for that query. 100 = the floor.
df["floor"] = df.groupby("query")["price"].transform("min")
df["index"] = (100 * df["price"] / df["floor"]).round(1)

print(df.pivot_table(index="banner", columns="query", values="index"))

That index is the number to chart over time. A banner drifting from 104 to 112 has repositioned, whether or not anyone announced it, and you’ll see it weeks before it shows up in a trade story.

Category benchmarks

Seventeen unified category slugs, resolved to each retailer’s own taxonomy at runtime: produce, dairy-eggs, meat, seafood, bakery, frozen, pantry, snacks, beverages, international, deli, prepared-meals, plant-based, health-beauty, baby-care, pet-care and household.

Coverage isn’t uniform, and the shape of it matters when you design a benchmark. Twelve resolve across all six default retailers. international skips T&T. The last five resolve only for Save-On-Foods and PriceSmart.

stats = pd.DataFrame([r for r in rows if r["_type"] == "intel_category_stats"])

stats["sale_share"] = (stats["on_sale_count"] / stats["product_count"]).round(3)
print(stats[["banner", "category", "product_count",
             "price_median", "sale_share"]].sort_values("sale_share", ascending=False))

sale_share is the one I watch. A banner running 8% of a category on promotion one week and 22% the next is buying traffic, and price_median tells you whether the everyday shelf moved with it or stayed put.

Running it weekly

Canadian grocery promotions turn over weekly, so schedule weekly on a fixed weekday. Tuesday one week and Saturday the next mixes promotional and regular pricing into what will look like a trend.

• Schedule it in the Apify console, on the Actor’s Schedules tab. Each run writes its own dataset.
• Keep every dataset. The Actor holds no history between runs, so your archive is the only history there is.
• Store run_id and schema_version with the rows. Both make a two-year-old table explainable.
• Reconcile against intel_run_summary before loading. A run where one retailer failed is a smaller sample, not a price movement.

The mechanics of turning scheduled runs into a series, including the two things that corrupt one, are in the fixed-basket guide. For pulling many runs into one table, the Python guide covers the client side.

What bites

• A skipped category looks like a zero. Ask for pet-care at Loblaws and that retailer is skipped, because no mapping exists. Nothing in the category rows says so; retailers_succeeded does.
• Sticker price across banners measures packaging. Benchmark on comparable_unit_price with comparable_unit_basis, and drop rows where is_comparable_unit_price is false.
• Matches are search results, not identities. Filter with minMatchConfidence at medium or better before you report a number to anyone, and spot-check source_url on the outliers.
• Naming no retailers runs six. The other ten are opt-in, so a monitoring brief covering Quebec needs maxi and provigo named explicitly.
• Region changes the answer. Eight metros are supported, and a banner’s price position in Toronto tells you little about Halifax. Run per region rather than averaging across them.