How Smaller Firms Can Achieve Data Lineage with Simple Agents

Advisor Perspectives welcomes guest contributions. The views presented here do not necessarily represent those of Advisor Perspectives.

The data lineage conversation in wealth management has been dominated by enterprise software vendors promising sophisticated tracking systems that cost six figures and require dedicated IT teams. But here's what the sales presentations miss: Smaller firms can achieve meaningful, defensible data lineage using tools they already own — if they understand what lineage actually means in practice.

Data lineage isn't about expensive technology. It's about answering three straightforward questions when a regulator, client, or internal auditor asks about AI-generated content:

What data went into this output?
Who used it, and when?
Where is the record stored?

If you can answer those questions reliably, you have functional lineage. Everything else is infrastructure theater.

The ChatGPT Session Tracking Foundation

The starting point for practical lineage is treating every business-related AI conversation as a documented process. This requires a structured prompt and response log stored in a secure, shared location such as SharePoint, Box, or Google Drive.

Each entry should capture four elements: the date and timestamp of the interaction, the user's name, the purpose or client context (using non-identifying descriptors), and the nature of the output generated. When that conversation influences client work, the final output gets saved directly to the relevant CRM record or document management system.

This creates what engineers call a manual lineage trail. In plain language, it means you can show exactly how advice or content was generated, who generated it, and when. The system isn't automated, but it's defensible — and that's what matters when compliance questions arise.

Version Control as Schema Tagging

When advisors upload firm templates, policy documents, or research materials to ChatGPT, those source files need to live in a central repository with clear version numbers. Label them v1.1, v1.2, and so on. Then note the version in your prompt log: "Generated client summary using Estate Planning Template v1.2."

This simple discipline transforms generic documentation into traceable lineage. You're creating what data engineers call schema tagging, which is the ability to connect each output to its specific source version. If a template changes, you know exactly which client communications used the old version versus the new one. That matters enormously when you are sending new communications or regulations change.

The beauty of this approach is its scalability. A solo practitioner can manage it with a numbered folder system. A 50-person firm can implement it with basic document management discipline. No specialized software required.

Automated Logging Through Platform Features

Firms using ChatGPT Teams or Enterprise already have automated lineage tools built into the platform, they just need to turn them on. These versions capture prompts, timestamps, and user IDs automatically through activity logging and workspace exports.

The implementation step is straightforward: enable these features, export the logs monthly, and store them in your compliance archive alongside other business records. This automated logging eliminates the manual documentation burden while creating a comprehensive audit trail.

For firms without access to these enterprise features, the same principle applies using other tools. Microsoft 365 and Google Workspace both offer activity logs that track document creation, modification, and sharing. Configure those logs to capture AI-related work, and you've built automated lineage tracking without additional software purchases.

Timestamps and Controlled Storage

Every AI-generated output needs to land in a repository that automatically records upload time and author identity. SharePoint, Google Drive, and Box all provide this functionality as a standard feature. The file metadata itself, including the creation date, modification date, and author name, becomes your timestamped lineage record.

This approach delivers the same evidentiary value as blockchain hashing or other exotic verification methods, but without the complexity or cost. When an auditor asks about a client communication generated nine months ago, you can produce the original file with embedded timestamps showing exactly when it was created and by whom.

The critical discipline here is consistency. Ad hoc storage on local drives defeats the entire purpose. Everything AI-generated goes into controlled storage immediately, no exceptions. That simple rule transforms ordinary document management into a lineage system.

The AI Interaction Register

The final component is a lightweight tracking mechanism that brings all the pieces together. Build a simple spreadsheet or CRM module with these columns: Date, User, Tool Used, Data Sources Referenced, Output Type, Storage Location, and Reviewed By.

Each row documents a single AI interaction. The entry for October 25, 2025 might show: "J. O'Connell used Perplexity with public training data plus internal marketing outline to generate blog draft, stored in SharePoint/Marketing/AIOutputs, reviewed by H. Potter."

That single table replaces expensive lineage systems for firms under 100 users. It makes governance visible, verifiable, and accessible to anyone who needs to understand the firm's AI usage patterns. For regulators, it demonstrates due diligence. For management, it provides oversight. For users, it clarifies expectations.

Making Lineage Work in Practice

Resistance to these approaches usually centers on the manual effort required. It’s a fair concern as documentation takes time. But consider what you're buying with that time: regulatory defensibility, client confidence, and operational transparency. Those benefits far outweigh the 90 seconds it takes to log an AI interaction.

The firms succeeding with practical lineage share a common characteristic: They treat AI documentation as a standard business process, not a special project. Just as advisors document client meetings and investment decisions, they document AI usage. The discipline becomes routine within weeks.

The Path Forward

Data lineage at smaller firms doesn't require data engineering infrastructure. It requires consistent documentation and version control applied systematically. The tools already exist in your technology stack. The challenge is implementation discipline, not technical capability.

Start with one area, such as client communications or marketing content, and build the documentation habit there. Once the process feels natural, it expands to other use cases. Within six months, you'll have created a comprehensive lineage system using nothing but existing tools and structured thinking.

The future of wealth management AI isn't about who has the most sophisticated technology. It's about who can demonstrate responsible, traceable, defensible usage. Smaller firms can compete effectively in that future if they stop waiting for perfect solutions and start building practical ones.

Take Action: Review your current AI usage patterns this week. Identify where documentation gaps exist. Implement a simple logging system for your highest risk use cases. That's how lineage becomes real instead of theoretical.

John O’Connell is founder and CEO of The Oasis Group, a leading consultancy for the wealth management industry that specializes in helping wealth management and technology firms solve their most complex challenges. The Oasis Group offers award-winning consulting services, industry-leading research, and compelling on-demand training for wealth management firms and the service providers who serve the wealth management industry.

John O’Connell will be speaking at the Exchange conference on March 15-18 in Las Vegas. You can register for the conference here.

A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.