Recover a Partially Failed CRM Import Without Duplicating Successful Rows
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A CRM import stops partway through. Some contacts are visible, others are missing, and the results file does not explain every row. Rerunning the original file is not a recovery plan: it could create a second record for a successful row or replace a correction someone made after the first attempt.
Instead, preserve the original evidence and classify every source row as confirmed success, confirmed failure, or uncertain. Retry only rows whose prior outcome and intended next write are understood. Keep uncertain rows out of the retry file until you can establish whether they already changed the destination.
The working tool is a batch-recovery ledger: one entry per stable source reference, with evidence, destination identity, error category, retry eligibility, and a reconciliation result. The ledger below shows how to use it without treating an importer’s summary count as proof of what happened.
1. Contain the batch before changing anything
Pause further submissions of the affected batch and assign one recovery owner. Ask colleagues not to rerun their own copies. Where authorised and available, contain downstream workflows that could act on newly imported records. Coordinate any pause so it does not unnecessarily interrupt unrelated work.
Preserve the original source file unchanged, the submitted mapping, import settings, available result files, job identifiers, and the time range of the attempt. Record whether the operation was intended to create records, update records, or do both. A recovery decision depends on that distinction.
Also record any changes made since the attempt: corrected phone numbers, reassigned owners, merged duplicates, or edits to customer preferences. A successfully imported record can now be more accurate than the original source file.
- Do not edit the original file in place. Make a separate correction file with a clear version label.
- Do not delete apparent duplicates immediately. First establish which record owns related activity and which identity the migration intended to preserve.
- Do not assume rollback is harmless. Deleting imported records may also remove or detach later work, depending on the destination’s behaviour.
Containment should produce an observable result: one owner, one preserved evidence set, and no competing replay of the same batch.
2. Give each source row a durable identity
A spreadsheet row number is useful for finding a line in a preserved file, but it is not a durable record identity. Sorting, filtering, or exporting a corrected subset changes line positions.
Use a stable source reference that remains the same across attempts. Prefer the source system’s record identifier, qualified by source and entity type where necessary. If none exists, assign a migration reference before recovery and maintain its relationship to the immutable original file and row.
Keep the source reference separate from the destination record ID. The first identifies what you meant to migrate; the second identifies what the destination actually contains. An email address or phone number can help investigate a match, but neither should be treated as uniquely identifying a record without checking the migration’s matching rules and the data itself.
For update operations, record the exact intended destination ID and the fields authorised to change. A known destination ID does not mean an old source value is still appropriate. For creates, establish whether a record was already created before considering another create.
If a retry mechanism supports a deduplication key or an update-or-create mode, verify its documented scope and matching behaviour. Do not infer duplicate protection from the feature’s name or assume the original attempt used it correctly.
3. Classify outcomes by evidence, not by error wording
The three outcome classes are decision states, not guesses about how likely a write was.
- Confirmed success: evidence ties the source reference to the intended destination write. This might combine a row-level result with inspection of the destination ID and relevant fields.
- Confirmed failure: evidence establishes that the intended write did not commit. A clearly documented pre-write rejection can support this; an unexplained error message cannot.
- Uncertain: available evidence does not establish whether the intended write committed, or whether it committed completely.
A timeout is uncertain until investigated. A missing result line is uncertain. A record not appearing in a search is not necessarily proof of failure: search scope, permissions, filters, and visibility delays can affect what the investigator sees.
For rows that can produce several objects or relationships, track the expected components separately. A contact may exist while its company association is missing. Calling the entire row failed and replaying all of it can duplicate the completed component.
Keep the error category separate from the outcome. Validation, missing dependency, authorisation, transport interruption, and identity conflict describe different problems. None, on its own, proves the final write state.
4. Use a filled batch-recovery ledger
Fictional worked example: All identifiers, times, counts, and outcomes in this example are hypothetical. A contact-create batch, REC-A, contains eight source rows. The original file is preserved as contacts-original-v1.csv. A destination enquiry and authorised record-history review are available in this hypothetical environment.
The ledger keeps the original outcome even after investigation changes the recovery decision. This preserves why a row was retried or withheld.
Batch: REC-A Original file: contacts-original-v1.csv Initial outcome totals: success 2 | failure 3 | uncertain 3 Recovery owner: Migration lead Source: LEGACY:CONTACT:101 | Original row: 2 Initial outcome: CONFIRMED SUCCESS Evidence: row receipt and destination C-701 agree Category: none Retry eligible: NO Action: retain C-701; owner was corrected after import Post-recovery check: C-701 remains unique for this source; later owner correction preserved Source: LEGACY:CONTACT:102 | Original row: 3 Initial outcome: CONFIRMED SUCCESS Evidence: row receipt and destination C-702 agree Category: none Retry eligible: NO Action: retain C-702; exclude from recovery file Post-recovery check: source mapping and intended fields agree; no additional destination record found Source: LEGACY:CONTACT:103 | Original row: 4 Initial outcome: CONFIRMED FAILURE Evidence: explicit pre-write rejection; no create performed Category: validation / invalid date value Retry eligible: YES, after approved correction Action: verify source date; correct only that value in v2 Post-retry check: receipt maps to C-703; corrected date agrees; no second mapping for this source Source: LEGACY:CONTACT:104 | Original row: 5 Initial outcome: CONFIRMED FAILURE Evidence: explicit pre-write rejection of missing company link Category: dependency / unresolved parent Retry eligible: YES, after parent mapping verified Action: verify company P-44 and submit its destination reference Post-retry check: C-704 exists and links to P-44; no placeholder company created Source: LEGACY:CONTACT:105 | Original row: 6 Initial outcome: CONFIRMED FAILURE Evidence: request rejected before write; authorised review agrees Category: authorisation / intended owner unavailable to operator Retry eligible: YES, after authorised access decision Action: approved operator submits with intended owner unchanged Post-retry check: C-705 exists with approved owner; no broader access granted merely to bypass the error Source: LEGACY:CONTACT:106 | Original row: 7 Initial outcome: UNCERTAIN Evidence: connection ended before row result was received Category: transport / missing acknowledgement Retry eligible: NO Investigation: destination history confirms create at C-706 Action: resolve as success; exclude from recovery file Post-recovery check: C-706 maps to source; later phone edit retained Source: LEGACY:CONTACT:107 | Original row: 8 Initial outcome: UNCERTAIN Evidence: no row result in original response Category: interrupted processing Retry eligible: YES, only after investigation Investigation: authoritative job evidence confirms row not executed; authorised destination check finds no matching source mapping Action: include unchanged source payload in recovery file Post-retry check: receipt maps to C-707; intended fields agree; no additional source mapping found Source: LEGACY:CONTACT:108 | Original row: 9 Initial outcome: UNCERTAIN Evidence: two plausible destination matches; no reliable mapping Category: identity conflict Retry eligible: NO / HOLD Action: data owner must resolve identity before any write Post-recovery check: no retry submitted; hold remains open
The recovery file contains only references 103, 104, 105, and 107. It does not contain all rows labelled with an error, nor every row absent from the first success report.
Notice the difference between references 106 and 107. Both initially lacked an acknowledgement. One had committed; the other had not executed. Their original symptoms did not justify the same action.
5. Build a correction set, not a replay of the original
Generate the retry file from the ledger’s approved references. Preserve the stable source reference in the recovery evidence even if the importer cannot store it directly. Record the recovery file version, mapping, intended operation, and approval before submission.
Make the smallest justified correction. A validation failure should not become an opportunity to refresh every field from an old export. Check the destination’s handling of omitted columns, empty cells, defaults, and update mappings: leaving a field blank is not necessarily the same as leaving it untouched.
For an update that failed earlier, re-read the destination before retrying. Compare the current value with the value observed before the original attempt. If someone has changed the field since then, hold it for a decision rather than overwriting it silently. Where supported and verified, conditional writes can help enforce this check; otherwise use a controlled review process.
Fix dependencies before dependent rows. Confirm that a parent record is the correct entity, not merely that some record with a similar name exists. Resolve permission errors through an authorised decision rather than giving the migration operator unrestricted access.
A small recovery set is easier to inspect, but size alone does not make it safe. The essential control is that each included row has evidence of non-commit, a resolved cause, and a defined intended write.
6. Treat the retry as another attempt that can fail
Give the retry its own attempt identifier and append its evidence to the ledger. Keep the original status and investigation history rather than replacing them with a single final label.
If the retry is interrupted, apply the same three-way classification again. Do not assume that membership in a carefully prepared recovery file makes the second attempt all-or-nothing. Some rows may now have committed while others remain unresolved.
If the service returns HTTP 429, it is signalling rate limiting, and a Retry-After header may indicate when to retry, as described in MDN’s HTTP 429 guidance. Timing guidance does not establish the write outcome of other rows in the batch. Preserve row-level evidence before scheduling another submission.
Stop recovery when a new uncertainty appears that cannot be isolated safely. Examples include unexpected matching behaviour, successful writes to the wrong entity type, or a retry that changes fields outside the approved set. Continuing to submit more rows can expand the investigation.
7. Reconcile identities, values, and relationships
A successful retry report is evidence, not the entire reconciliation. Check the intended destination identity, the relevant field values, required relationships, and any later corrections that must survive.
Counts are useful as a cross-check but cannot prove correctness. One duplicate and one missing record can leave a total unchanged. Reconcile stable source references to destination IDs and inspect unexplained many-to-one or one-to-many mappings against the migration’s intended model.
For the hypothetical ledger, the closing position is:
- Two original confirmed successes remain untouched.
- One originally uncertain row is confirmed as an existing success.
- Four approved retry rows are confirmed successful after reconciliation.
- One identity conflict remains on hold and has not been resubmitted.
That accounts for all eight source references: seven reconciled successes and one unresolved hold. It does not justify marking the whole batch complete. The held row needs a named decision owner and a next review point.
For any downstream actions, check them separately from record creation. A correct CRM record does not establish that an associated task, notification, or other workflow completed correctly. Avoid replaying those actions merely because the data import required recovery.
8. Set a measurable release gate
Before the next recovery submission, require every proposed retry row to have a stable source reference, evidence that the intended write has not committed, a resolved error cause, and an approved payload. Require every excluded success to have a destination mapping, and every uncertain row to have an investigation owner.
Start with the failed batch already in front of you: preserve it, classify every source reference, and count the entries in each state. The next action is not to rerun the file. It is to reduce the number of unexplained outcomes without creating new writes.
Close the recovery only when every source reference is either reconciled to its intended result or recorded as an explicit exception with an owner. A visible hold is safer than a duplicate created to make the completion count look right.
Put the next conversation to work.
Explore Alactic Communications for WhatsApp conversations, campaigns and AI-assisted workflows. Choose the product and plan that fit your team.
Explore Communications