By Siddharth GangalA call-coaching migration must preserve more than recordings. The evidence graph includes transcript and speaker turns, rubric definitions and versions, manual and AI evaluations, evidence links, comments, assignments, completion and dispute history, users, managers, teams, CRM relationships, permissions, consent, retention, and audit events. A media folder without those edges cannot reproduce a coaching record.
This vendor-neutral guide uses representative official documentation reviewed August 8, 2026. Gangly did not execute an export. Plans, permissions, formats, histories, media paths and employee/privacy obligations vary. Verify the actual tenant, contract, DPA, policies, holds and target.
Keep coaching custody distinct from adjacent canonicals
This page owns vendor-neutral custody and transition. Use the call-coaching buyer guide to select a destination, the live-coaching guide to validate in-call prompts, and the training-platform checklist for courses, certifications and broader learner records. Named vendor pages such as the Gong exit checklist own product-specific paths.
Freeze source/target, plans, contracts, capture systems, CRM, archive, holds, snapshot, cutoff, parallel-access, rollback, cancellation, retention and deletion dates. Name enablement, source, target, manager, data, CRM, security, privacy/HR, legal, validation, dispute and rollback owners.
Hard gates: no missing held recording; accepted critical transcript/evaluation/rubric edges; no unauthorized visibility or evaluation; preserved disputes and corrections; one evaluation authority; calibrated scoring; tested rollback; and closure evidence.
Inventory coaching media, definitions, evidence, people, and rights
Inventory calls/meetings, audio/video, recordings, channels, timestamps, participants, consent/notice, transcript versions, speakers, utterances, summaries, highlights, snippets, topics, trackers and source evidence. Add scorecards, rubrics, sections, questions, response scales, weights, guidance, examples, required evidence, pass rules, versions, effective dates and owners.
Inventory evaluations, manual/AI identity, scores, answers, evidence timestamps, confidence, edits, overrides, comments, replies, disputes, appeals, resolution, assignments, queues, due dates, completion, coaching sessions, action plans and acknowledgments. Add users, learners, managers, teams, hierarchy versions, employment status, territories and access groups.
Include CRM account/contact/opportunity/activity IDs and associations; roles, permission profiles, shares, links, downloads, audit events, integration users, OAuth/webhooks/API keys; retention, legal holds, deletion state, exports and support deliveries. Every row needs stable ID, parent/edges, authority, owner, version, format, portability label, target, test and evidence.
Map rubric versions, evaluations, people, CRM, and disputes
Build edge tables for call→recording→transcript version→speaker turn; rubric→version→question/weight; evaluation→rubric version→call→evidence timestamps; evaluator/model→evaluation; evaluation→edit/override/comment/dispute/resolution; assignment→learner→manager→team hierarchy version; and call→CRM activity→account/contact/opportunity.
Preserve source IDs and immutable source timestamps. A score of 4 is meaningless without the rubric version, question, scale, evaluator and evidence. A manager relationship must use the hierarchy effective at evaluation time, not only today’s org chart. Link deleted/deprovisioned users to historical pseudonymous identities rather than orphaning evaluations.
Use the CRM migration checklist for CRM object and relationship maps. CRM association is supporting context; it must not silently rewrite the historical coaching record.
Classify portability without assuming universal export
Use documented exportable, buyer verified, support/contract dependent, reconstruct, and unavailable. Record URL/ticket, plan, permission, path/API, filters, format/schema, media/history/relationship coverage, limits, exclusions, sample and test date.
Gong’s official scorecard documentation describes scorecards, scoring, visibility and export-related behavior. Its call-data download documents call metadata, participants, CRM associations and privacy state. Those sources demonstrate useful paths but do not establish a universal complete export of media, transcripts, rubric-version history, evaluations, comments, assignments or hierarchy.
Microsoft’s portability documentation separates customer data from some protected operational/configuration data. Apply the underlying principle to every coaching artifact: ownership or visibility does not itself prove a machine-readable, relationship-complete export.
Freeze configuration and preserve snapshot plus final delta
- Freeze rubrics, versions, weights, models, prompts, assignments, teams, hierarchy, permissions, CRM maps, capture, retention and deletion.
- Record active calls, processing media/transcripts, queued AI scores, open manual evaluations, comments, disputes, assignments and exports.
- Take immutable snapshot manifests and export configuration before derived records.
- Stop new capture/evaluation in a rehearsed order while preserving required late media, comments, dispute decisions and corrections as a final delta.
- Keep source read/dispute access until acceptance and rollback expiry.
For each file/batch store source, filters, UTC range, schema/version, rows, bytes, media count/duration, errors, exclusions and SHA-256. Hashes prove file identity, not coaching completeness. Transform versioned copies only and log re-encoding, redaction, mappings, rejects and manual repair.
Reconcile media, fields, rubric edges, and history
- Artifact coverage = accepted artifacts ÷ eligible source artifacts.
- Media coverage = accepted playable required recordings ÷ eligible recordings.
- Field fidelity = accepted values ÷ eligible tested values.
- Rubric/evidence edge accuracy = correct evaluation-to-version-to-evidence edges ÷ eligible edges.
- People/hierarchy edge accuracy = correct learner/evaluator/manager/team edges ÷ eligible edges.
- History coverage = accepted versions, edits, comments, disputes, assignments and audit events ÷ eligible history.
Full-reconcile held media, current rubrics, open assignments/disputes, critical evaluations, permissions and deletion restrictions. Stratify samples by team, manager, learner, rubric version, evaluator type, channel, date, language, score band, edit/dispute and CRM object. A playable call attached to the wrong rubric version fails.
Test duplicates, replay, out-of-order media/transcript/evaluation, late comments, missing evidence, corrupt media, deleted user, team move, CRM merge, permission loss, token expiry, rate limit, schema change and malicious transcript content. Expected outcomes specify quarantine, retry, abstention, alert, owner, recovery and no duplicate evaluation.
Stage the target and calibrate one evaluation authority
Build target identities, teams, permissions, rubrics/versions, assignments, CRM maps and retention with capture/evaluation disabled. Load a consented golden set and verify media, transcript, evidence timestamps, manual scores, AI labels where applicable, comments, disputes and hierarchy edges.
Run parallel access, not parallel production judgment. Users may inspect both, but only one system creates official evaluations and assignments. Calibrate multiple reviewers against observable rubric anchors; measure agreement before resolving differences. Preserve the former system’s score as historical evidence rather than recalculating it under a new rubric.
Apply the NIST Privacy Framework as general privacy-risk structure, not an export procedure. Define employee notice, access, correction/dispute, permitted employment use, retention, sharing and incident ownership with qualified stakeholders.
Cut over, prove rollback, and preserve disputes
Approve a canary only when snapshot/delta pass; held media and critical edges reconcile; permissions and privacy gates pass; rubrics are calibrated; one evaluation authority is enforced; monitoring works; and rollback is rehearsed. Start with a bounded team and rubric.
Rollback disables target capture/evaluation/assignments, preserves target-only evidence/comments/disputes, restores approved source configuration and identity, imports new corrections, reconciles, and prevents replay. Stop for missing held media, wrong learner/evaluator, rubric-version drift, unauthorized visibility, untraceable score, lost dispute, duplicate assignment or untested rollback.
Decommission, revoke, archive, and evidence deletion
After acceptance and hold/dispute review, revoke conferencing/dialer/CRM integrations, API keys, webhooks, OAuth, SSO, service identities, public links, exports, support access and user/admin accounts. Test access fails. Retain approved archives with limited custodians and documented expiry.
Request deletion scope for media, transcripts, derived insights, scorecards/versions, evaluations, comments, assignments, people/hierarchy, CRM links, permissions, audit logs, exports, support artifacts, subprocessors, backups, holds and exceptions. The conversation-intelligence privacy guide helps frame governance. Cancellation is not deletion; preserve evidence of what remains, why, protection and expiry.
Apply the worked example and calculate TCO
This fictional arithmetic is not a vendor result. Of 5,000 required recordings, 4,900 are playable; 2,000 evaluations yield 1,920 correct rubric/evidence graphs; 3,000 people/team edges yield 2,940 correct edges; 800 required history events yield 720; and all 100 open disputes remain accessible.
| Control | Calculation | Result |
|---|---|---|
| Media | 4,900 ÷ 5,000 | 98% |
| Rubric/evidence edges | 1,920 ÷ 2,000 | 96% |
| People edges | 2,940 ÷ 3,000 | 98% |
| History | 720 ÷ 800 | 90% |
| Open disputes | 100 ÷ 100 | 100% |
Do not average gates. Disputes pass; missing held media or critical evidence edges remain blockers. TCO = source overlap/exit + target setup/license + export/support + media storage/transfer + rubric reconstruction + mapping/engineering + enablement calibration + privacy/HR/legal/security + validation + monitoring + remediation + rollback + archive + deletion.
Print the coaching-platform migration checklist
| Control | Evidence | Status |
|---|---|---|
| Scope, holds, owners, cutoff frozen | Charter | □ |
| Media, rubrics, evaluations, people, CRM, rights inventoried | Graph register | □ |
| Portability labels verified | Docs/tickets/samples | □ |
| Snapshot/final delta preserved | Manifests/hashes | □ |
| Media, fields, edges, history reconcile | Signed report | □ |
| Permissions, disputes, privacy pass | Test logs | □ |
| One-authority canary and rollback pass | Rehearsal | □ |
| Access revoked and deletion evidenced | Closure pack | □ |
Gangly is not a universal coaching-record migration destination. Any target should pass the same custody, rubric-version, evidence, people/CRM edge, privacy, rollback and TCO controls.