Mirakl product import failed: Analyse error reports with AI

If products have not been imported into a Mirakl marketplace, first check the status of the specific product import ID, then the available error reports. Distinguish two stages: transforming your source data and integrating it into the operator’s catalogue. The message associated with the affected product identifier tells you whether to examine a field mapping, an attribute value or an identifier, for example. A completed import alone does not establish an error-free catalogue entry. The Mirakl Seller API for product import status separates these details.

Mirakl is the technical platform behind various marketplaces. Each operator’s configuration determines the applicable categories, required fields and values. With the Backlane Mirakl integration, you can examine available import data from your connected seller accounts in an AI chat. This guide describes a read-only diagnostic workflow; the example is entirely fictional.

Which import is affected: products or offers?

Start with the type of import. Product data describes the catalogue item, such as its title, category and attributes. Offer data concerns your sales offer, including price and stock. Mirakl uses different import processes and reports for the two.

What do you want to check?	Appropriate request	What does it tell you?
Find a product import	P51: List product imports	Which import IDs and status details are available
Investigate a known product import ID	P42: Product import status	Processing status, counters and available reports
Understand errors when transforming the source file	P47: Source file error report	Which transformation problems were reported
Investigate products that were not integrated	P44: Non-integrated products report	Which products were not integrated and what information the report contains
Investigate an offer import	OF02 and OF03	The offer import’s status and error report, respectively

These names follow the Products API overview. For offers, Mirakl documents OF02 and OF03 separately. Do not use an offer import ID as a product import ID without verifying it. A successful product import also does not confirm a successful offer import.

What information do you need for an unambiguous diagnosis?

You need the marketplace, the correct seller connection, the shop ID where relevant, and the product import ID. Add the uploaded file, import time and affected product identifiers where available. Your own SKU and the operator’s product identifier can be different values; their relationship must remain traceable.

If you do not have the import ID, use P51 to check the list of product imports. Include all required results pages. The last_request_date filter means “changed since this time”, not “created since this time”. This is described in the P51 documentation.

The Backlane documentation explains how to set up the connection. Replace the placeholders in the first prompt with the details you already know:

Run a read-only check of the product import with the known import ID
[import ID] on [marketplace, connection and shop ID if relevant].

Show the import ID, creation time, import_status, reason_status,
transformation counters, available integration details and
information about the availability of error reports.
State the source and retrieval time; flag missing fields.

Explain whether processing is still running and which reports
can usefully be checked now. Do not claim success merely because
a report is missing. Do not start a new import.

What is the difference between transformation and integration errors?

A transformation error concerns processing the source file. An integration error concerns incorporating the products into the marketplace operator’s system. Check both available reports rather than judging the entire import from one stage without apparent issues.

For P47, Mirakl returns CSV, XLSX or XML depending on the submitted file. For P44, the report is a CSV file when Mirakl Catalog Manager, or MCM, is enabled; without MCM, the operator determines the format. You should therefore assume neither fixed column names nor a standard error structure across all marketplaces. Sources: P47: Transformation report, P44: Non-integrated products report.

Timing matters too. According to P42, has_error_report is populated at COMPLETE, and has_transformation_error_report at SENT or COMPLETE. A missing report while processing is ongoing therefore does not prove that there are no errors. Read the status, counters and report together; missing values are not zero.

Backlane provides read access to statuses and report files. Whether your AI tool can directly analyse a returned file also depends on its format and the tool’s file capabilities. If the file cannot be read, you can supply the original report from the seller portal in a supported format. The AI should state this limitation and not invent messages.

How do you turn many messages into a concrete worklist?

Group messages by processing stage, affected field and actual error text. This helps you identify corrections that affect several products. Keep the product identifiers and original messages alongside each group so it remains verifiable.

For a compact template covering status and error reports, use the Mirakl import diagnosis prompt.

Analyse the available error reports for product import [import ID]
in [marketplace, connection and shop ID if relevant], using read-only
access. Include both transformation and integration.

Group by stage, affected field and original message.
For each group, show the number of messages, the number of distinct
affected products, product identifiers, any error codes, the location
of the evidence in the report and a concrete next check.

Separate reported causes from your assumptions.
Do not count a product with several messages as several different
products. Do not invent codes or required fields.
Flag unreadable files and unresolved identifier mappings.
Do not change data or upload any file.

Counters from different processing stages cannot automatically be added together to produce a total number of failed products. One product can have several messages. A distinct count requires a verified mapping across the reports.

What does the workflow look like in a seller example?

Suppose you upload twelve home accessories. In this fictional example, the product import has ID 64021: two records fail during transformation. Of the ten products processed further, three are not integrated and seven are successfully synchronised. The affected products explicitly do not overlap in this example.

The following table contains paraphrased, fictional findings, not official error messages or error codes. The assumed required attribute applies only to this example.

Finding in the example	Affected products	What should you check next?
The source category has not been mapped to a valid target value	VASE-01, VASE-02	Compare the source file’s category and its mapping to the operator’s category
A material attribute required for this category is missing	BASKET-03, BASKET-04	Check the current attribute configuration and add the verified product value in the source system
The submitted product identifier is not accepted	LAMP-05	Compare the reported identifier type and submitted value with the operator’s requirements

The first two vases require a mapping check. For the baskets, an actual product value is missing in the example. These are different tasks: a material value cannot reliably be copied from a similar product, and a new category should not be selected simply to make the error message disappear.

Mirakl provides separate interfaces for the category’s attribute configuration and the operator’s value lists. Your marketplace’s current requirements are what matter. This example does not establish universal required fields for home24, Decathlon or other operators.

You then review the five records, correct them in the appropriate source system and start a new import yourself. Record the new import ID and which products it actually contains.

How do you check whether your correction worked?

Compare the new import with the old one using the same product identifiers. A lower error count is not enough: products that previously had errors might not have been included in the second import at all.

I have corrected the data and imported it again myself.
Using read-only access, compare the old product import [old import ID]
with [new import ID] in [marketplace, connection, shop ID].
Products expected to have been resubmitted: [product identifiers].

For each product, show the old message, evidence of resubmission,
the new processing status and any available new messages.
Flag missing evidence and processing that is still running.
Use the original file or my confirmed product list if the reports
do not allow a complete mapping.

Separate confirmed integration, errors still being reported and
outstanding checks. Do not infer offer approval from these results.
Do not make changes or start another import.

If a rejection remains unexplained, collect the import ID, product identifier, original message and the data already checked for the appropriate support team. This narrows down the unresolved question without creating more import runs by repeatedly uploading unchanged data.

The practical guide to analysing marketplace data with AI explains how to investigate other questions about your connected marketplaces.