Merging duplicates

This module solves one situation: the same physical product arrives from several vendors' feeds as several separate records. The system offers two different answers to that, and it matters not to confuse them.

Linking versus merging

The primary product is the one that survives a merge. The secondary is the one that gets archived. That split decides everything else, which is why the manual returns to it in several places.

The module at a glance

You will find it under Products → Duplicates (Opens in a new tab). It opens on the Overview tab.

1 — Find duplicates2 — Overview3 — Linked duplicates4 — Suggestions5 — Manual search6 — Merge history7 — Linked groups8 — Secondary products9 — Pending suggestions10 — Link all suggestions11 — Detection rules

Controls on this screen

1. 1Find duplicates — Runs automatic detection across the whole catalogue.
2. 2Overview — The state of the module and the detection rules.
3. 3Linked duplicates — Pairs that are already linked.
4. 4Suggestions — The queue of found pairs waiting for your decision.
5. 5Manual search — Searching by fields you choose, plus saved searches.
6. 6Merge history — What can still be undone, plus the permanent log.
7. 7Linked groups — How many pairs are linked but not yet merged.
8. 8Secondary products — How many products are currently hidden as secondaries.
9. 9Pending suggestions — How many pairs are waiting for a decision on the Suggestions tab.
10. 10Link all suggestions — Links every pending suggestion at once. It links, it does not merge.
11. 11Detection rules — The five strategies and the field each one compares.

Automatic detection

Find duplicates walks the catalogue with five strategies. Each compares one specific field and carries its own level of certainty:

A GTIN is a globally unique barcode, hence the highest certainty. A variant SKU, by contrast, is invented by each vendor, so a match can be pure coincidence — that is the reason for 60 %, and also for the rule that this strategy only compares across vendors, never within one. On screen it carries the Vendors only label.

Manual search and saved searches

When automation is not enough — because two vendors put their identifiers in mismatched fields, say — use manual search.

1 — Save search2 — Saved searches3 — Side A — primary4 — Side A vendor5 — Side A comparison field6 — Side B — secondary7 — Side B vendor8 — Side B comparison field9 — Link directly10 — Run search

Controls on this screen

1. 1Save search — Stores the current form. Enabled once the form is filled in.
2. 2Saved searches — Each shows when it last ran and how many matches it found.
3. 3Side A — primary — Products from this side become the primaries, so they survive.
4. 4Side A vendor — The vendor whose products are treated as primary.
5. 5Side A comparison field — Which identifier to compare.
6. 6Side B — secondary — These products attach to side A as secondaries.
7. 7Side B vendor — It need not be a different vendor, though usually it is.
8. 8Side B comparison field — It need not be the same field as side A — that is the point of manual search.
9. 9Link directly — Skips the Suggestions queue and links the found pairs immediately.
10. 10Run search — Runs a one-off search without saving it.

Which fields can be compared

Each side offers eleven fields, and they need not match — that is exactly the power of manual search: vendor A's GTIN against vendor B's variant SKU.

• At product level: code, external ID 1–3, GTIN / EAN, MPN, manufacturer SKU, name.
• At variant level: variant SKU, variant GTIN, variant external ID.

Why save a search

The right mapping for a particular catalogue is knowledge that otherwise lives in one person's head. A saved search records it, and turns the search into one click. It also removes the most common mistake: swapping sides A and B, which quietly inverts which product becomes the primary.

Suggestions: what to do with a found pair

The queue of found pairs is on the Suggestions tab. Nothing happens here on its own — every pair waits for your decision. Each row offers three actions:

1. Reject

   The pair is not a duplicate. It leaves the queue and detection will not propose it again.

2. Link

   You confirm it is the same product, but nothing is copied and nothing is archived. The safest choice.

3. Compare and merge

   Opens the single-pair merge screen.

Each pair shows the match type — which rule found it, and the actual value — and the certainty from the table above. The row checkboxes select pairs for a bulk merge; the selection survives paging.

Linked duplicates

The Linked duplicates tab holds the pairs you confirmed but have not merged. Each row shows the primary, the secondary (hidden in the shop) and the match type that created the pair. From a link you can move to a merge at any time, or remove it — which puts the secondary back in the shop.

The list deliberately hides pairs where one product has already been archived by a merge. Such a row would offer buttons that cannot succeed.

Merging a single pair

The comparison screen is where you decide exactly what the resulting product looks like. Field by field you pick the value the result keeps, and switching the primary source decides which product survives the merge.

What the strategies mean

Bulk merge from a pattern

With dozens of pairs, clicking each one is unthinkable. So a bulk merge works differently: you configure it once on one sample pair, and the system repeats that same recipe across everything you selected.

A partial result is a normal outcome

One bad pair does not cancel the other ninety-nine — pairs already merged stay merged. That is why a run reports each pair separately: merged, skipped (with a reason) and failed (with the message).

History and un-merging

The Merge history tab holds two lists, answering two different questions:

• Merges that can be undone — answers “what can I still take back?”. The row disappears once un-merged.
• The merge and un-merge log — answers “what happened here?”. It stays forever.

What exactly comes back

An un-merge is a true inverse. It restores:

• the archived secondary products — to the state they were in before the merge (a published product comes back published, not as a draft);
• the primary's original field values;
• the variants, images, parameters and categories the merge added — these are removed;
• the primary's main image, if the merge overwrote it;
• the duplicate links the merge removed or repointed;
• duplicate suggestions go back to pending.

Every one of those steps is guarded so that a newer decision by an admin always wins. If you changed something by hand after the merge, the un-merge will not overwrite it.

The log, by contrast, never disappears. It is the one record that survives an un-merge — who did what, when, which products it touched, and whether it came from a bulk run.

Limits, permissions and constraints

Who may do what

Numeric limits

If you select more than a hundred pairs, the system stops the selection at a hundred and says so. If a larger number reaches the page address, the first hundred are taken and the screen warns you — merge the rest in a second run.