Industry-specific operational data
Wealth management client-service and advisor notes data for advisor copilots
Quick answer
Wealth management data for AI training usually means four linked record sets: advisor CRM notes, meeting summaries, client service requests and the NIGO (not in good order) reasons that operations teams attach to rejected paperwork. Copilots learn most when each note or request is joined to its outcome, such as a completed ACATS transfer or a resubmitted distribution form. Buyers should specify those joins, the redaction method for client nonpublic personal information, and the licensing basis for reuse before negotiating.
By SourceX Editorial · Updated
What the record set looks like inside a wealth firm
The useful record set spans the advisor desk and the service-operations back office, and it rarely sits in one system. Advisor notes and meeting summaries typically live in a CRM such as Salesforce Financial Services Cloud, Wealthbox or Redtail, while service requests and NIGO dispositions live in a workflow or case tool tied to the custodian or clearing firm. A buyer who asks only for "CRM notes" gets text without consequences; a buyer who asks for the request lifecycle gets labels.
The main components are:
- Advisor notes: free-text call and meeting notes, task follow-ups, household and relationship fields, next-step dates.
- Meeting summaries: review-meeting agendas, discussion summaries and action items, sometimes generated by an existing note-taker and then edited by the advisor.
- Service requests: account opening, ACATS transfers in and out, distributions and required minimum distributions, beneficiary updates, address and contact changes, standing instructions and journal requests.
- NIGO events: the rejection reason, the document or field at fault, who caught it (home office, custodian, transfer agent) and how long the resubmission took.
- Outcomes: completed, rejected, withdrawn or escalated, with timestamps.
For adjacent record shapes, compare after-call work notes and disposition codes and fraud investigation case notes with analyst decisions. Research and portfolio commentary is a different asset with different limits, covered in investment research notes and analyst work product.
Which model tasks each record type supports
Each record type maps to a distinct model task, so the specification should start from the task and work back to fields. Summarization copilots need raw meeting transcripts or advisor-written notes paired with the final summary and action items. Classification and routing agents need the original request text plus the request type assigned by operations, not just the type the advisor picked.
| Task | Minimum records | Label or target | Common failure |
|---|---|---|---|
| Meeting summarization and follow-up drafting | Notes or transcripts plus edited summaries | Advisor-approved summary, action items | Summaries trained on unedited drafts copy their errors |
| Service-request classification | Request text, channel, form type | Operations-assigned request type | Advisor-chosen category differs from final category |
| NIGO prediction | Submitted form fields, account type, custodian | NIGO reason code, resubmission count | Reason codes vary by custodian and change over time |
| Next-best-action and task generation | Notes plus subsequent tasks and requests | Task actually created and completed | Survivorship: only completed tasks are logged |
| Evaluation sets | Held-out households and time periods | Human-reviewed gold answers | Leakage when the same household appears in train and test |
NIGO reasons are a natural label because operations staff assign them at the point of rejection, but they are only as consistent as the firm's taxonomy. Ask for the code list as it stood across the sample period, including retired codes. Request distributions for a sample month so you can see whether "missing signature" dominates and the long tail is too thin to learn.
Recommended versus executed: labeling advice-related notes
Notes that discuss recommendations need a field that separates what was discussed, what was recommended and what was executed. Under the SEC's Regulation Best Interest and adviser fiduciary standards, those three states carry different weight, and a copilot trained on blended text can learn to present a discussion as a recommendation. This is a labeling problem the supplier can only partly solve, so plan for buyer-side annotation.
Practical steps: ask whether the CRM has structured fields for recommendation type and rationale; join notes to trade or order records by account and date window where possible; and mark notes written after the fact (late entries) separately from contemporaneous ones. Exclude or flag supervisory review comments, which reflect compliance judgment rather than client service.
Privacy and reuse rules that travel with client notes
Client names, account numbers, balances and life events in advisor notes are nonpublic personal information under the Gramm-Leach-Bliley Act privacy rules, and those rules limit reuse by whoever receives the data. Regulation P says a recipient that got NPI under an exception may use and disclose it only in the ordinary course of business to carry out the purpose for which it was received [2]. Parallel reuse and redisclosure sections appear in the other agencies' privacy rules, including the CFTC's version at 17 CFR 160.11 [1]; for broker-dealers and SEC-registered advisers the corresponding rule is the SEC's Regulation S-P (17 CFR 248.11) [7], so counsel should confirm which version applies to the supplying firm. The FTC explains that a recipient outside an exception steps into the shoes of the originating institution and is bound whether or not it is itself a financial institution [3].
The SEC amended Regulation S-P in May 2024 to add incident response, customer notification and service-provider oversight requirements; as of October 2026 the compliance dates for larger entities (December 3, 2025) and smaller entities (June 3, 2026) have both passed [6]. For an AI buyer, that means the supplying firm will scrutinize where its data goes and how quickly you would report a breach. Expect security questionnaires and contractual notice terms in the data agreement.
Advisor notes also carry sensitive content that standard PII scrubbers miss: diagnoses behind a long-term-care discussion, divorce and custody details, a beneficiary's disability, or a client's cognitive decline flagged for a trusted contact. If any notes originate from a HIPAA covered entity context, de-identification must meet Safe Harbor or Expert Determination, and neither removes all re-identification risk [4]. Otherwise, ask for a documented method covering free-text entities (names of children, employers, small towns, dates of death), not just structured fields. For deeper checks on contract language, see customer contracts and DPAs for training use.
Recordkeeping duties stay with the supplying firm
Licensing a copy of advisor communications does not change the supplying firm's own recordkeeping duties, so buyers should never be asked to act as a record of retention. For broker-dealers, FINRA Rule 4511 and SEA Rule 17a-4 govern how books and records, including business communications, are made, preserved and stored [8], and firm policies often set longer retention than the rules require. Treat these as the supplier's obligations, confirmed by its compliance team, not terms you inherit.
Two practical consequences follow. First, the extract you receive is a derived copy, and the firm's archive (often a Smarsh, Global Relay or similar system for communications) remains the record. Second, extracts pulled from an archive may include messages that are on legal hold or subject to regulatory inquiry, so ask the supplier to confirm holds were screened out.
If the supplier is a service provider or BPO handling multiple advisory firms, the rights question changes; see client data held by service providers. Exports from a SaaS CRM also raise platform-terms questions covered in SaaS platform terms and exported data.
A request specification you can send
A strong request names the record types, the joins, the time window and the redaction standard without naming target firms. The schema below shows the level of detail that lets a supplier assess fit quickly.
Illustrative example: invented to show structure; it does not describe an available dataset.
{
"service_request": {
"request_id": "SR-2025-118204",
"household_token": "HH_7f3a",
"account_type": "Traditional IRA",
"custodian_type": "third-party custodian",
"request_type_submitted": "distribution",
"request_type_final": "required minimum distribution",
"channel": "advisor portal",
"submitted_at": "2025-11-04T15:12:00Z",
"nigo_events": [
{
"reason_code": "WITHHOLDING_ELECTION_MISSING",
"reason_text": "Federal withholding election blank",
"detected_by": "custodian",
"detected_at": "2025-11-05T13:40:00Z",
"resubmitted_at": "2025-11-06T16:05:00Z"
}
],
"final_status": "completed",
"completed_at": "2025-11-07T19:30:00Z"
},
"linked_note": {
"note_id": "N-55120",
"note_type": "client call",
"entered_at": "2025-11-03T21:10:00Z",
"late_entry": false,
"text_redacted": "[CLIENT] asked to take RMD early this year; wants cash moved to [BANK]. Confirm withholding.",
"recommendation_state": "discussed",
"redaction_method": "NER + rules, sampled QA"
}
}
Checklist to accompany the request:
- Record types and approximate volumes wanted per type, with the date range.
- Required joins: note to request, request to NIGO event, request to final status.
- The NIGO code list and any custodian-specific mappings.
- Redaction scope for free text, including family, health and estate details, and the QA sample size.
- Whether household and account tokens stay consistent across files for longitudinal tasks.
- Intended uses (fine-tuning, evaluation, retrieval) and where the model will be deployed.
Disclosure duties on the buyer side
Builders of generative AI products may have their own disclosure duties that depend on what the training data contains. California's AB 2013 requires developers of generative AI systems made available to Californians to post documentation about training data, including whether datasets include personal information [5]; those disclosures were due January 1, 2026. Keep the supplier's description of sources, redaction method and collection period so your documentation is accurate.
How SourceX handles wealth service data requests
SourceX sources operational datasets, including support histories, documents and finance workflows, from US companies on request; nothing is held in stock, and a request does not guarantee a match. Buyers describe the data, not the businesses; every release is approved by the supplying company, and nothing is contracted until a supplier agrees. Each dataset is rights-reviewed for ownership and consents, personal details such as names, account numbers and phone numbers are removed or replaced before delivery with the method recorded and a sample checked, and delivery runs through private, access-controlled workflows after an executed agreement. You can describe your advisor-copilot data needs to SourceX, and see the finance view at buyers by industry: finance.
Related reading: AI use cases for financial advisors data, sales CRM pipeline histories, policy-following service agent data, and the industry-specific operational data hub within the AI data guide.
Source wealth service and advisor notes data for your copilot
SourceX looks for US businesses that hold the advisor notes, service requests and NIGO histories you describe, assesses data and licensing permissions, and agrees pricing and allowed uses in a license before any transaction. Terms are agreed per deal, and SourceX serves AI teams wherever they are based. Start a buyer request at SourceX.
This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.
Frequently asked questions
Are advisor CRM notes useful without the linked service outcomes?
They are useful for summarization style and vocabulary, but weak for agents that act. Without the request outcome or NIGO event, a model cannot learn which notes led to completed work, so prioritize joined records.
Can a broker-dealer license communications it must retain under Rule 17a-4?
Retention duties and licensing are separate questions. The firm keeps its archived record regardless, under FINRA Rule 4511 and SEA Rule 17a-4 [8]; whether it may license a redacted copy depends on its privacy notices, client agreements and the GLBA reuse limits [2][3], which counsel should review.
How should NIGO reason codes be normalized across custodians?
Map each custodian's codes to a buyer-side taxonomy (signature, missing form, field error, eligibility, funding) and keep the original code in a separate field. Retired and renamed codes should be mapped, not dropped.
Sources
- Electronic Code of Federal Regulations (eCFR), "17 CFR 160.11 - Limits on redisclosure and reuse of information". https://www.ecfr.gov/current/title-17/chapter-I/part-160/subpart-B/section-160.11
- Consumer Financial Protection Bureau, "12 CFR 1016.11 - Limits on redisclosure and reuse of information (Regulation P)". https://www.consumerfinance.gov/rules-policy/regulations/1016/11/
- Federal Trade Commission, "How To Comply with the Privacy of Consumer Financial Information Rule of the Gramm-Leach-Bliley Act". https://www.ftc.gov/business-guidance/resources/how-comply-privacy-consumer-financial-information-rule-gramm-leach-bliley-act
- U.S. Department of Health and Human Services, Office for Civil Rights, "Guidance Regarding Methods for De-identification of Protected Health Information in Accordance with the HIPAA Privacy Rule" (2012). https://www.hhs.gov/hipaa/for-professionals/special-topics/de-identification
- California Legislature, "AB-2013 Generative artificial intelligence: training data transparency" (2024). https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202320240AB2013
- U.S. Securities and Exchange Commission, "Regulation S-P: Privacy of Consumer Financial Information and Safeguarding Customer Information" (Release No. 34-100155, 2024). https://www.sec.gov/files/rules/final/2024/34-100155.pdf
- Legal Information Institute, Cornell Law School, "17 CFR 248.11 - Limits on redisclosure and reuse of information". https://www.law.cornell.edu/cfr/text/17/248.11
- FINRA, "4511. General Requirements". https://www.finra.org/rules-guidance/rulebooks/finra-rules/4511
Tell us what your models need
Share scope, volume, language, format, timing and licensing requirements.