Multipack product statistics for quantity matches and substitutions By Checkout Reply, Checkout Reply. Sources checked 2026-11-11. A public search benchmark shows when product titles disagree with the pack size shoppers asked for. ## Quick answer In Amazon ESCI’s US test data, 25 of 152 exact candidates with one explicit pack count in the title named a different count from the query. The same check flagged 98 of 147 substitutes. Confirm the pack size from the sellable variant before suggesting a replacement or sending a checkout link. Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) ## Key findings and sources ### Exact candidates with a single stated title count disagreed 16.45% of the time Checkout Reply’s reanalysis of Amazon ESCI found a different explicit count in 25 of 152 exact candidate titles when the US test query and title each stated one pack quantity. Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) US large test candidates for Amazon ESCI queries containing a single explicit pack count Extract one distinct count from the query and title, then divide 25 disagreements by 152 comparable exact pairs. ### Two-thirds of comparable substitute titles stated a different count In the US Amazon ESCI pack-query slice, 98 of 147 substitute titles with one explicit count stated a different pack quantity from the query, or 66.67%. Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) US large test candidates for Amazon ESCI queries containing a single explicit pack count Count 98 disagreement pairs and 49 agreement pairs within the substitute relevance label. ### Most substitute titles had no single usable count under the rule In the same lexical analysis, 231 of 378 substitute candidate titles, or 61.11%, had no single explicit pack count. That includes 227 silent titles and four with multiple distinct counts. Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) US large test candidates for Amazon ESCI queries containing a single explicit pack count Add silent and multiple-count titles, then divide by all substitute pairs in the filtered sample. ### The substitute disagreement rate was 50.22 points above exact candidates Among Amazon ESCI candidate pairs where the title stated one count, the substitute disagreement rate was 50.22 percentage points higher than the exact disagreement rate in this US test slice. Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) US large test candidates for Amazon ESCI queries containing a single explicit pack count Subtract the unrounded exact disagreement rate from the unrounded substitute rate. ## Compare the results Figure description: Title count differed from the query. Exact values are also in the study data. Title count differed from the query. ## A relevant item can be the wrong quantity A shopper asks for a 12-pack and the search result looks like the right brand and flavor. The count is still a separate buying requirement. In the exact group, 25 of 152 candidate titles that stated one pack count named a different number from the query, or 16.45%. In the substitutes group, 98 of 147 named a different count, or 66.67%. That 50.22-point gap is a reason to inspect the count on suggested alternatives. The benchmark’s exact label comes from its relevance assessment, while our check uses just the wording in the title. A title can contain a count that belongs to an accessory or a different option. Follow a discrepancy through to the listing or variant before treating it as a catalog mistake. ## Handle an absent count differently from a conflict Among all 311 exact pairs, 157 titles lacked an explicit count that matched the search patterns. Another two titles contained multiple distinct counts. The missing count can be a normal consequence of the title format. A size such as 12 bottles or a case of 12 is outside the exact phrase rule used here. A broader catalog audit can include those patterns once they have been checked against real examples. Use the structured units-per-pack field or verified variant options if the buyer asks how many items are in one order unit. When the record is silent, ask a clear question or send the shopper to the right option selector. When the sources conflict, pause the replacement and verify the combination with the catalog owner. ## Keep quantity intact from the reply to the order Check three fields before showing an alternative: the number of units in the package, the number of packages the cart will add, and the total the shopper sees after discounts. Two six-packs can satisfy a 12-unit need if the shopper accepts two packages and the total is clear. A single six-pack is a different offer. Measure the share of attempted suggestions where a verified option and the final cart agree on all three fields. Review silent records separately from contradictions. High substitution traffic can hide frequent quantity issues if you count clicks alone. ## Study data Table: Amazon Science Shopping Queries Dataset, public examples and product files Measure | Value | Context Exact candidate pairs | 311 query-product pairs | Original aggregate from public Parquet join Exact distinct queries | 40 queries | Original aggregate from public Parquet join Exact pairs with no explicit title count | 157 query-product pairs | Original aggregate from public Parquet join Exact pairs with multiple title counts | 2 query-product pairs | Original aggregate from public Parquet join Exact single title count agreed | 127 query-product pairs | Original aggregate from public Parquet join Exact single title count disagreed | 25 query-product pairs | Original aggregate from public Parquet join Exact pairs with a different stated count | 16.45% | Original conditional share Substitute candidate pairs | 378 query-product pairs | Original aggregate from public Parquet join Substitute distinct queries | 36 queries | Original aggregate from public Parquet join Substitute pairs with no explicit title count | 227 query-product pairs | Original aggregate from public Parquet join Substitute pairs with multiple title counts | 4 query-product pairs | Original aggregate from public Parquet join Substitute single title count agreed | 49 query-product pairs | Original aggregate from public Parquet join Substitute single title count disagreed | 98 query-product pairs | Original aggregate from public Parquet join Substitute pairs with a different stated count | 66.67% | Original conditional share ## What changes when you compare the figures Table: Calculations from the published values above. Comparison | Result | Calculation Exact pairs with a different stated count | 16.45% | = 25 / 152 * 100 Exact pairs with no title count | 50.48% | = 157 / 311 * 100 Substitute pairs with a different stated count | 66.67% | = 98 / 147 * 100 Substitute pairs with no title count | 60.05% | = 227 / 378 * 100 Substitute minus exact title disagreement | 50.22 percentage points | = (98 / 147 - 25 / 152) * 100 Exact pairs with no single explicit count | 51.13% | = 159 / 311 * 100 Substitute pairs with no single explicit count | 61.11% | = 231 / 378 * 100 ## Methodology and sources Filter Amazon ESCI to large US test candidates. Find whole phrases such as 12-pack or pack of 12 in each query, retaining queries with exactly one distinct number. Join on product ID and locale, apply the same patterns to titles and count agreement only when the title names one distinct number. The analysis counts pairs, so products can appear more than once. A disagreement flags a title to check. It does not establish that the item delivered contains the wrong quantity. The replay script records all four relevance labels. [Download the source data (CSV)](decision-table.csv) [Download the calculations (CSV)](calculations.csv) [Download the citation notes (CSV)](citation-notes.csv) [Download the evidence behind the new findings (CSV)](finding-evidence.csv) [Download the reproducible analysis script](esci_pack_titles.py) ## Related reading AI shopping recommendation accuracy statistics (https://checkoutreply.com/guides/ai-shopping-recommendation-accuracy-statistics/) Ecommerce product review statistics and what shoppers use to decide (https://checkoutreply.com/guides/ecommerce-product-review-statistics/) AI shopping price and availability error statistics (https://checkoutreply.com/guides/ai-shopping-price-availability-statistics/) Product search statistics on exact matches and substitutes (https://checkoutreply.com/guides/ecommerce-product-substitute-statistics/) Product color data statistics for exact matches and substitutes (https://checkoutreply.com/guides/product-color-field-coverage-statistics/) What to text when a shopper leaves checkout (https://checkoutreply.com/guides/first-imessage-abandoned-checkout/) How to answer “Will it get here by Friday?” before they buy (https://checkoutreply.com/guides/delivery-question-before-checkout/) Browse checkout recovery research (https://checkoutreply.com/research/) ## Sources and definitions Amazon Science Shopping Queries Dataset, public examples and product files (https://github.com/amazon-science/esci-data/tree/main/shopping_queries_dataset) The source data includes the population, period, denominator and limitations for every figure. Checkout Reply · checkoutreply.com