Ecommerce order total reconciliation statistics
https://takeratehq.com/guides/ecommerce-order-total-reconciliation-statistics/
## Quick answer
Payment totals matched merchandise plus freight exactly for 98,089 of 98,665 joined orders in Olist’s public dataset, or 99.42%. Another 273 orders differed by at most one Brazilian centavo. The 249 differences above BRL1 included 246 delivered orders. Use a discrepancy flag to open a reconciliation check before asking a shopper to pay again. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
A high reconciliation rate can coexist with records that need investigation. The comparison below separates exact equality, rounding-sized differences and larger discrepancies, then checks final delivery status. It also keeps orders missing from one source file visible. The arithmetic describes Olist’s historical Brazilian marketplace records; a current payment request still depends on the live invoice and transaction history.
## Key Takeaways
Exact equality held for 98,089 of 98,665 matched Olist orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
273 matched orders had a nonzero absolute difference of at most BRL0.01 (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
54 matched orders differed by above BRL0.01 through BRL1 (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
246 of 249 matched orders with a difference above BRL1 were marked delivered (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
## Which order records were compared?
The analysis sums every payment row for an order, separately sums item price and freight, then joins the totals by order ID. The comparison uses the final records supplied in the historical public files.
Figure description: Ecommerce order total reconciliation statistics: of 249 matched orders differing by more than BRL 1, 17 have payments below item and freight totals and 232 have payments above them.
Figure 1: Direction reverses the first action for 232 of the 249 material discrepancies. Source: Olist public dataset, original signed reconstruction (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026. Original analysis of the cited source. Source periods and calculation limits appear in the text.
This gives a precise consistency check between two records a conversational agent might encounter. A payable balance requires more information, including the current invoice, captured payments, refunds, cancellations and adjustments.
## How large were the recorded differences?
Absolute difference between recorded payment totals and item prices plus freight, 98,665 matched Olist orders. BRL denotes Brazilian reais. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
Difference | Orders
Exact match | 98089
Nonzero, at most BRL0.01 | 273
Above BRL0.01 through BRL1 | 54
Above BRL1 | 249
The four bands account for every matched order. Exact matching identifies 576 exceptions. A one-cent tolerance leaves 303. A BRL1 threshold leaves 249. Those choices change the investigation queue, so record the tolerance alongside each reported reconciliation rate. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
Choose a tolerance from your payment system’s documented currency and rounding rules. Keep material differences visible even when the overall match rate is high. A percentage rounded to a whole number would conceal much of this tail.
This distribution covers 98,665 matched orders from Olist’s historical 2016 to 2018 files. Each value is an order count. Unmatched records are excluded from the rate denominator. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
## Does a large difference identify an unfinished order?
Of the 249 orders with an absolute difference above BRL1, 246 had a final delivered status. These records show why a fulfillment flag alone leaves the accounting question open. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
A store can investigate the underlying transaction while acknowledging that an order was delivered. Before a recovery message asks for money, the agent should obtain the current payable balance from the system responsible for invoices and transactions.
## What happens to orders found in only one file?
The payment file covered 99,440 distinct orders and the item file covered 98,666. Their intersection contained 98,665. There were 775 payment-side orders without an item-side match, plus one item-side order without a payment-side match. (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
An inner join excludes those records. The headline match rate therefore describes the intersection. For a merchant workflow, retain the unmatched group as a separate review queue and inspect why a source record is missing.
Export the coverage counts with the result. This keeps a high matched-order rate from obscuring incomplete coverage of the original files.
## What should be checked before another payment request?
First resolve the order ID and currency. Sum all rows belonging to that order on each side. Compare the payment total with the item-and-freight total using the documented tolerance.
Then inspect the current invoice balance and transaction history. If the order has a material difference, assign an investigation before sending another payment request. If the files fail to join, route the record for reconciliation instead of estimating its amount.
For a simple arithmetic example, item prices of BRL90 and freight of BRL10 produce an expected total of BRL100. Recorded payment rows of BRL60 and BRL40 reconcile to that total. Looking only at the first row would leave an apparent BRL40 gap.
That example also explains why a matching total cannot verify which products were paid for. Product identity and line-item allocation require their own checks.
## Which source figures support the findings?
The table retains the source measure and its population. Independent results in the next table use the stated formula; their precision follows the recorded inputs.
Source measure | Value and unit | Population or denominator
Exact payment and item-total matches | 98,089 orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98665
Nonzero differences up to BRL0.01 | 273 orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98665
Differences above BRL0.01 through BRL1 | 54 orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98665
Differences above BRL1 | 249 orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98665
Delivered orders with differences above BRL1 | 246 orders (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98665
### How were the additional comparisons calculated?
These are calculations from the source inputs. Each retains its original unit and comparison population.
Calculation | Result | Formula
Exact agreement | 99.42 % (Olist data, independent analysis (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026) | 98089 / 98665 * 100
## Which exceptions would actually resemble an underpayment?
Only 17 of the 249 differences above BRL 1 have recorded payments below merchandise plus freight. The other 232 run in the opposite direction, with recorded payments exceeding those charges. The 232 records account for BRL 3,064.76 in positive differences; the 17 account for BRL 197.57 in negative differences. These are signed export differences, not verified amounts refundable or collectible. (Olist data, independent signed reconstruction (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
That direction changes the first support action. Treating every material discrepancy as an unpaid balance would put 232 of 249 records, 93.17%, on the wrong side of the comparison before any invoice history had been read. Even the 17 negative differences need investigation, because missing adjustments or different accounting scope can produce a shortfall without an amount currently owed. The net positive total of BRL 2,867.19 also conceals both populations. Keep the count and gross amount for each direction instead of reporting net exposure alone.
### Does a more permissive tolerance solve the coverage problem?
Moving from exact equality to a BRL 0.01 tolerance reduces the matched-record queue from 576 to 303, a 47.40% reduction. Moving from that tolerance to BRL 1 removes another 54, a 17.82% reduction. Meanwhile, 776 orders appear on only one side of the join. That unmatched population is 3.12 times the 249 material matched differences. A smaller discrepancy queue can therefore coexist with a larger unresolved coverage queue. (Olist data, independent reconstruction (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
A merchant should reconcile the union of identifiers before celebrating the intersection's match rate. In these files the union is 99,441 orders. At a BRL 1 threshold, 1,025 records require one of two different investigations: 249 matched but discrepant, and 776 unmatched. That is 1.03% of the union, with no claim that these records represent losses. The queues need different owners: accounting can inspect a signed discrepancy, while an integration or data owner checks why a counterpart record is absent.
### Are multiple payment rows enough to explain the tail?
The material-difference rate is 11 of 2,936 matched orders with multiple payment rows, 0.375%, compared with 238 of 95,729 single-row orders, 0.249%. The former is approximately 1.51 times the latter, but the groups are observational and the smaller group contains only 11 material cases. Multiple rows are a useful aggregation requirement, not an adequate diagnosis. Excluding single-row orders would miss 238 of the 249 material discrepancies. (Olist data, independent subgroup reconstruction (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce), Nov 2026).
The downloadable signed discrepancy calculations (https://takeratehq.com/guides/ecommerce-order-total-reconciliation-statistics/data/discrepancy-exposure.csv) and executable reconstruction (https://takeratehq.com/guides/ecommerce-order-total-reconciliation-statistics/analysis.py) preserve these populations. Run the reconstruction against the original Olist ZIP. A live implementation should additionally test refunds, partial captures, currency mismatches and missing source records against its own documented accounting rules.
## Methodology and Sources
Inner-join payment sums and item-price-plus-freight sums by order_id using exact Decimal values. Positive difference means recorded payments exceed item and freight totals.
The source purchase file contains 99,441 orders spanning September 2016 through October 2018. Counts in this study use all available matched payment and item records. Historical Brazilian marketplace observations require validation against a current merchant’s payment systems and market.
The replay uses decimal values rather than binary floating-point arithmetic. It keeps positive and negative differences available for investigation, while the headline bands use absolute differences. The supplied source tables retain order counts and denominators.
The data support a record-consistency analysis. Establishing why a difference arose, whether a balance remains due or whether a recovery message would increase revenue requires additional evidence.
The source period is Olist historical public dataset; purchase source file spans 2016-09-04 to 2018-10-17. Independent calculations were completed in November 2026 and use the formulas shown above.
## Glossary
Order ID: The key used to join records belonging to an order.
Payment record: One row describing a recorded payment amount or method.
Freight: The shipping amount recorded with an item.
Matched order: An order found in both files used for a comparison.
Denominator: The population used to calculate a share or rate.
Recorded final state: The status present in the historical export.
Payable balance: The amount currently owed after relevant payments and adjustments.
Reconciliation: A check that related financial records agree under stated rules.
## Sources
Olist data: Brazilian E-Commerce Public Dataset by Olist (https://www.kaggle.com/datasets/olistbr/brazilian-ecommerce)
Check the source figures
The data file retains source labels, values, populations and limitations for this analysis.
Quick answer
Payment totals matched merchandise plus freight exactly for 98,089 of 98,665 joined orders in Olist’s public dataset, or 99.42%. Another 273 orders differed by at most one Brazilian centavo. The 249 differences above BRL1 included 246 delivered orders. Use a discrepancy flag to open a reconciliation check before asking a shopper to pay again. (Olist data, independent analysis, Nov 2026).
A high reconciliation rate can coexist with records that need investigation. The comparison below separates exact equality, rounding-sized differences and larger discrepancies, then checks final delivery status. It also keeps orders missing from one source file visible. The arithmetic describes Olist’s historical Brazilian marketplace records; a current payment request still depends on the live invoice and transaction history.
The analysis sums every payment row for an order, separately sums item price and freight, then joins the totals by order ID. The comparison uses the final records supplied in the historical public files.
Figure 1: Direction reverses the first action for 232 of the 249 material discrepancies. Source: Olist public dataset, original signed reconstruction, Nov 2026. Original analysis of the cited source. Source periods and calculation limits appear in the text.
This gives a precise consistency check between two records a conversational agent might encounter. A payable balance requires more information, including the current invoice, captured payments, refunds, cancellations and adjustments.
Absolute difference between recorded payment totals and item prices plus freight, 98,665 matched Olist orders. BRL denotes Brazilian reais. (Olist data, independent analysis, Nov 2026).
Difference
Orders
Exact match
98089
Nonzero, at most BRL0.01
273
Above BRL0.01 through BRL1
54
Above BRL1
249
The four bands account for every matched order. Exact matching identifies 576 exceptions. A one-cent tolerance leaves 303. A BRL1 threshold leaves 249. Those choices change the investigation queue, so record the tolerance alongside each reported reconciliation rate. (Olist data, independent analysis, Nov 2026).
Choose a tolerance from your payment system’s documented currency and rounding rules. Keep material differences visible even when the overall match rate is high. A percentage rounded to a whole number would conceal much of this tail.
This distribution covers 98,665 matched orders from Olist’s historical 2016 to 2018 files. Each value is an order count. Unmatched records are excluded from the rate denominator. (Olist data, independent analysis, Nov 2026).
Of the 249 orders with an absolute difference above BRL1, 246 had a final delivered status. These records show why a fulfillment flag alone leaves the accounting question open. (Olist data, independent analysis, Nov 2026).
A store can investigate the underlying transaction while acknowledging that an order was delivered. Before a recovery message asks for money, the agent should obtain the current payable balance from the system responsible for invoices and transactions.
The payment file covered 99,440 distinct orders and the item file covered 98,666. Their intersection contained 98,665. There were 775 payment-side orders without an item-side match, plus one item-side order without a payment-side match. (Olist data, independent analysis, Nov 2026).
An inner join excludes those records. The headline match rate therefore describes the intersection. For a merchant workflow, retain the unmatched group as a separate review queue and inspect why a source record is missing.
Export the coverage counts with the result. This keeps a high matched-order rate from obscuring incomplete coverage of the original files.
First resolve the order ID and currency. Sum all rows belonging to that order on each side. Compare the payment total with the item-and-freight total using the documented tolerance.
Then inspect the current invoice balance and transaction history. If the order has a material difference, assign an investigation before sending another payment request. If the files fail to join, route the record for reconciliation instead of estimating its amount.
For a simple arithmetic example, item prices of BRL90 and freight of BRL10 produce an expected total of BRL100. Recorded payment rows of BRL60 and BRL40 reconcile to that total. Looking only at the first row would leave an apparent BRL40 gap.
That example also explains why a matching total cannot verify which products were paid for. Product identity and line-item allocation require their own checks.
The table retains the source measure and its population. Independent results in the next table use the stated formula; their precision follows the recorded inputs.
Only 17 of the 249 differences above BRL 1 have recorded payments below merchandise plus freight. The other 232 run in the opposite direction, with recorded payments exceeding those charges. The 232 records account for BRL 3,064.76 in positive differences; the 17 account for BRL 197.57 in negative differences. These are signed export differences, not verified amounts refundable or collectible. (Olist data, independent signed reconstruction, Nov 2026).
That direction changes the first support action. Treating every material discrepancy as an unpaid balance would put 232 of 249 records, 93.17%, on the wrong side of the comparison before any invoice history had been read. Even the 17 negative differences need investigation, because missing adjustments or different accounting scope can produce a shortfall without an amount currently owed. The net positive total of BRL 2,867.19 also conceals both populations. Keep the count and gross amount for each direction instead of reporting net exposure alone.
Moving from exact equality to a BRL 0.01 tolerance reduces the matched-record queue from 576 to 303, a 47.40% reduction. Moving from that tolerance to BRL 1 removes another 54, a 17.82% reduction. Meanwhile, 776 orders appear on only one side of the join. That unmatched population is 3.12 times the 249 material matched differences. A smaller discrepancy queue can therefore coexist with a larger unresolved coverage queue. (Olist data, independent reconstruction, Nov 2026).
A merchant should reconcile the union of identifiers before celebrating the intersection's match rate. In these files the union is 99,441 orders. At a BRL 1 threshold, 1,025 records require one of two different investigations: 249 matched but discrepant, and 776 unmatched. That is 1.03% of the union, with no claim that these records represent losses. The queues need different owners: accounting can inspect a signed discrepancy, while an integration or data owner checks why a counterpart record is absent.
The material-difference rate is 11 of 2,936 matched orders with multiple payment rows, 0.375%, compared with 238 of 95,729 single-row orders, 0.249%. The former is approximately 1.51 times the latter, but the groups are observational and the smaller group contains only 11 material cases. Multiple rows are a useful aggregation requirement, not an adequate diagnosis. Excluding single-row orders would miss 238 of the 249 material discrepancies. (Olist data, independent subgroup reconstruction, Nov 2026).
The downloadable signed discrepancy calculations and executable reconstruction preserve these populations. Run the reconstruction against the original Olist ZIP. A live implementation should additionally test refunds, partial captures, currency mismatches and missing source records against its own documented accounting rules.
Why matching totals can show a $40 gap
A $40 order has two item lines and three payments. Join those exports on order ID before either is totaled and the result has six rows. Each item appears three times and each payment appears twice. The joined table shows $120 in item charges and $80 in payments, even though each original export adds up to $40 for this order.
Total each export by order ID first, then compare the two totals. Keep identifiers that appear on only one side so a missing record stays visible.
Six illustrative orders after totaling each export by order ID
Order
Order lines
Payment entries
What to check
DEMO-101
$40.00
$40.00
Both totals match. The earlier $40 difference came from the join.
DEMO-102
$32.00
$30.00
Check the $2.00 shortfall.
DEMO-103
$18.00
No record
Check why the payment export omits this order.
DEMO-104
No record
$7.25
Check why the order export omits this identifier.
DEMO-105
$13.00
$13.00
Both totals match.
DEMO-106
$21.00
$24.00
Check the $3.00 excess on the payment side.
Confirm the currency, date window and amount definition on both sides before treating any difference as real. This check compares two exports. A bank settlement, refund or fee can require a separate record.
Inner-join payment sums and item-price-plus-freight sums by order_id using exact Decimal values. Positive difference means recorded payments exceed item and freight totals.
The source purchase file contains 99,441 orders spanning September 2016 through October 2018. Counts in this study use all available matched payment and item records. Historical Brazilian marketplace observations require validation against a current merchant’s payment systems and market.
The replay uses decimal values rather than binary floating-point arithmetic. It keeps positive and negative differences available for investigation, while the headline bands use absolute differences. The supplied source tables retain order counts and denominators.
The data support a record-consistency analysis. Establishing why a difference arose, whether a balance remains due or whether a recovery message would increase revenue requires additional evidence.
The source period is Olist historical public dataset; purchase source file spans 2016-09-04 to 2018-10-17. Independent calculations were completed in November 2026 and use the formulas shown above.
Use the citation below when referring to this analysis. The original source and calculation scope accompany each finding.
Take Rate. "Ecommerce order total reconciliation statistics." Take Rate, November 2026. https://takeratehq.com/guides/ecommerce-order-total-reconciliation-statistics/
<a href="https://takeratehq.com/guides/ecommerce-order-total-reconciliation-statistics/">Ecommerce order total reconciliation statistics</a> (Take Rate, November 2026)