Methodology

Shrinkflation monitor

We detect shrinkflation through two mechanisms: within-SKU (the same product shrinks) and new EAN (a retailer deactivates a large SKU and introduces a smaller SKU under the same brand). Both are documented below.

1. Within-SKU

The same product (same EAN, same retailer SKU) shrinks over time. We register this once we observe ≥ 2 different quantity values for the same retailer-product in our nightly retailer_product_size_history audit log. The earliest possible event therefore happens afterthe monitor system’s first registration date — we cannot retrospectively reconstruct shrinkflations from before that date.

2. New EAN (“EAN swap”)

A retailer removes a product from the shelf and within 60 days introduces a new product under the same brand with a closely-similar name and a smaller size. Detection steps:

• Find pairs (deactivated RP-X, introduced RP-Y) within a 60-day window at the same retailer, with identical brand (case-insensitive).
• Variant-token strip: names are stripped of size \d+, nappy \d+, and multipack markers \d+ x \d+ before similarity scoring. Without this step, Pampers size 3 vs size 8 would be incorrectly flagged as a swap.
• Threshold: trigram similarity (pg_trgm) ≥ 0.85 on the cleaned names.
• Requirement: normalised quantity of RP-Y must be smaller than RP-X, on the same base unit (g↔kg, ml↔l are normalised first).
• Price snapshot at detection time — no later lookup. This prevents an intervening price change from distorting the editorial baseline.
• Operator verification mandatorybefore any publication. Every published event has been manually checked against the retailer’s own product page.

Confidence tiers

EAN-swap candidates are classified into three tiers:

• Tier 1 — High confidence: both RPs already point to the same canonical product. Rare (≈ 2.5 % of NL volume).
• Tier 2 — Medium confidence: similarity ≥ 0.85 on cleaned names but different canonicals. The workhorse tier — the vast majority of published events come from this bucket.
• Tier 3 — Triage: similarity between 0.65 and 0.85. Never published publicly — operator-only inbox for manual review. The public API filters these rows out regardless of query parameters.

Not detected

• Recipe shrink: same packaging, different ingredient ratio (no historical data available).
• Cross-channel arbitrage: price differences between online and in-store.
• Within-EAN variant downgrade: quality or ingredient-grade reduction without a size or EAN change.
• Price hikes disguised as promo expiry — outside the shrinkflation definition (no size change).

See also the public register of verified events.