A CSV import begins before the Import button
Design CSV column mapping, a preview of changes and update matching. A user should know the import's effect before writing to a catalogue.
Article language: English

In this article
Twelve rows in a file do not necessarily mean twelve new products. In our hypothetical import, three items are to be created, four updated and two left unchanged; three need clarification. A label saying “Import 12 products” blurs all those differences before the user makes a decision.
CSV is a format for transferring a table; importing is an action that changes data. Before saving, the user needs to understand column meanings, record identity and the expected effect. “Valid file” alone does not say whether width will be saved in the right unit or whether an empty cell will erase an old description.
We will walk through the design of an import into a fictional catalogue of workshop equipment. The example numbers and states explain decisions; they are not results from an implementation or a test of a provider's service. We are interested in the person's path from a received spreadsheet to knowingly accepted changes.
The selected file is material to interpret
The first screen should identify the destination: which catalogue or table will receive the data, and what actions the import can perform. “Add new items” has a different scope from “Update existing ones”. If the product permits both at once, the user needs to recognise them before pressing the final button. This is part of the task, not an advanced administrator setting.
A supplier's file may contain columns named “code”, “description”, “width” and the Polish word “stan”. In Polish, “stan” could describe a quantity in stock or the item's condition. A clear target-field name and sample values help reveal the difference. Automatic assignment based on a similar label should remain a suggestion for review.
Before mapping, show how the application parsed the file. Was the first row read as headers? Are Polish characters preserved, and have values stayed in place despite a separator inside a description? A user should recognise these effects even without knowing encoding, the way characters are stored. If an entire row has landed in one column, they should not have to map it to an arbitrary field and continue.
A header is a clue, not proof that the data are correct. The first column might be called “ID” yet contain consecutive spreadsheet row numbers. The catalogue needs its own identifier definition. In a product data model (in Polish), those concepts are established earlier; the importer should use them instead of creating a new definition for every file.
The reader should also understand the preview's scope. “Sample of 5 rows” says something different from “Entire file checked”. If the application shows only the beginning, say so beside the table. Do not give the whole file a green status merely because a fragment fitted neatly in the window. A full check needs a mechanism that actually covers the agreed scope.
When the file is replaced, identify the new material clearly and invalidate the previous check results. Earlier mapping may be useful, but must be compared with the current columns. A person should not see file B's preview under file A's name or accept old check results as new ones. Sketch this state before connecting the screens to background processing.
Mapping connects meanings, not just headings
For each source column, show the selected target field, several values and the option to deliberately skip it. A required field needs an explanation of what is missing. If the file lacks suitable information, the right outcome may be to return with corrected material. Do not force someone to map an arbitrary column merely to complete the form.
Atlassian describes automatic and manual column mapping in the new Jira Cloud import. It recommends checking automatic assignments and notes that unmapped columns are excluded. This is one concrete example of the difference between information being present in a file and being accepted into a system.
In our catalogue, “width” is to map to product width in centimetres. The supplied file records values in millimetres: an input of 120 mm should correspond to 12 cm. Merely selecting a field should not silently trigger a default conversion. If conversion is offered, the preview must show the value before and after it. If not, indicate the correction needed in the source rather than pretending the unit names match.
For numeric values, commas, precision and permitted ranges matter too. Designing a numeric input (in Polish) helps decide how an entry is interpreted, but an import must apply that decision to an entire column. An example such as “12,5” should reveal whether the application received twelve and a half or interpreted the notation differently. Showing the result beside the input removes guesswork.
Values from a predefined list need their own mapping. “Ready” in a spreadsheet may correspond to “Ready to publish” in the catalogue, but need not mean “Published”. Show which values will be mapped to an existing list and which need a decision. Automatically creating options also changes how the team works. It should not happen simply because accepting every word is easier.
Separate skipping a column from an empty value in its cell. When updating a description, an empty cell may mean “I have no new information” or “clear the existing text”. The user must choose a rule that fits the task. Better still, the preview shows its effect on a specific record whose old description is not empty.
Shopify distinguishes an empty value in a non-required column from omitting that column when overwriting products. The first can clear an existing value; the second can preserve it. Its documentation also describes dependencies between fields. Your own importer therefore needs to disclose its rule, rather than suggest that every blank is harmless.
An update needs an identity criterion
In our catalogue, a product with code PR-014 already exists. If the import concerns the same variant, a width change may be suggested. A similar title for a new product is not enough to update that record. The preview should show the basis for the match and a recognisable target object so the person can confirm what will be replaced.
Airtable's CSV import extension lets users choose a field for merging with existing records. The documented rule is case-sensitive, and a blank key can create a new record. This is that extension's behaviour, not a rule to attribute to every importer. It shows a real effect of choosing a key and of missing keys.
In the interface, show a new record, an update and an unresolved case separately. “Similar match found” is not yet the decision to overwrite. In assessing AI record linkage (in Polish), that boundary needs its own evidence. Here we design the moment when a person knowingly accepts a particular suggestion, regardless of the method that produced it.

Two file rows may point to the same product. The user needs to know the rule: refusal, one combined change or another agreed treatment. Do not assume the last row simply wins. If the source contains conflicting values, a duplicate label should lead to a decision rather than merely decorate the preview with a warning.
Similarly, several target records may match one input. A product name repeated across two variants is an ordinary case, not an exotic failure. Without the right identifier, the importer should not choose a record by a convenient order. Before saving, show the ambiguity and the scope held back. The person needs to recognise what still needs clarification.
When designing an import with Aeneia Design, such examples help connect the mapping screen, change preview and outcome report. Rather than start from a blank template, bring a few real column names and illustrative values showing difficult cases. The design must reflect the agreed rules for writing data.
A sample preview is not confirmation of the whole file
The supplier prepared the first rows carefully. Missing codes and an unrecognised unit appear only near the end. If the panel shows only the beginning, the person cannot guess those problems. The status should distinguish a preview of the representation from the result of a full check. They support different decisions and may be produced at different times.
The new Jira Cloud import documentation describes checking the first 30 rows during mapping. It warns that errors in later rows can still cause failure or a partial import. The sample's limit does not disappear because the screen allows the user to continue.
In our hypothetical file of twelve entries, the summary should show three new, four to update, two unchanged and three needing clarification. The numbers add up to twelve input rows in this example. We do not generally assume one row always equals one product: a file of variants or relationships may have a different structure, which the design should name.
Before approving updates, let people inspect the difference in values. “Width changed” does not show whether 120 was replaced with 12 or the notation was merely formatted differently. A change view should show the current and proposed value with the unit. A preview of the whole row without highlighting the difference shifts the comparison onto the operator's memory.
Changing a mapping requires the effects to be recalculated. If the “stan” column is assigned to another field, old update counts must not still look current. The design needs a waiting state or an indication that the previous summary needs refreshing. Otherwise the user will approve the scope shown before their last correction.
Approval applies to a chosen version and a defined effect
The final button should name the scope being accepted. For our file it might say “Add 3 and update 4 items” if the system permits those seven while leaving exceptions outside the operation. If one error stops the whole file, a different completion model is needed. The screen cannot promise a partial write that the application does not perform.
Between preview and saving, someone may change an existing product. This calls for a rule agreed with the implementer: compare again, hold the conflict or use another version check. A confirmation window alone does not freeze data. The interface should show the result of the actual check and the available next step, not merely repeat the question of consent.
After starting, show the state of the ongoing task and what cancellation can do. Closing a window may only hide the view, not stop the write. If “Cancel” becomes unavailable after a certain stage, explain that effect clearly. Do not suggest that all changes can be undone merely because a cross appears in the panel's corner.
A slower operation particularly needs a way to find the result after returning. File name, import destination and task identification help distinguish two similar runs. A person should know whether a new action creates another task or returns to one already accepted. Double-clicking must not become an accidental way to add two copies of a product.
Get a report that helps you resolve exceptions
The post-operation summary shows what was actually written, left unchanged and rejected. It may differ from the earlier forecast, but every material difference needs explanation. “Import complete” without counts and references to records does not establish whether the catalogue is ready to use. The overall status should also reveal a partial result.
Airtable also explains that a field value not accepted need not stop a record being created or updated. Its importer may leave a cell empty, while the summary does not list skipped values. A successful record therefore does not automatically confirm that every value was accepted.
For that reason, our report should separate saved records from fields needing correction. For a problem, identify the source row, field and a cause the user can act on. A row number needs a clear definition, especially when cell text contains a newline or the file has a header. The user should reach the right entry rather than count lines differently from the application. A recognisable product code helps confirm the right case is being corrected.
An exceptions file can help with further work if it preserves the necessary columns and reason for rejection. Its form also needs design. An export to another program (in Polish) should preserve the meaning of values, not just the row count. A report that cannot be read correctly creates the next problem in the same journey.
Reimporting corrected entries should not create copies of products already added merely because the user sent the whole file. The record-matching rule and scope selection must settle this. The design should show what will be retried and what retains its earlier result. Otherwise fixing three gaps may reopen work on all twelve.
If the material was prepared in a spreadsheet used in an AI process (in Polish), also check values actually saved in the file after editing. A displayed product number may have changed type, or a code lost a zero. Input checking should detect such a change before record matching. Trust in the spreadsheet preview does not replace reading the material handed over.
A properly received import leaves an understandable decision: which columns were accepted, how products were matched and what result each group received. Our twelve rows then stop pretending to be twelve new things. The user knows what affected the catalogue and what still needs clarification before the supplier sends the next, exceptionally final spreadsheet.







