As access to some external data sources becomes more restricted, companies must find ways to gain valuable insights without exposing sensitive information or giving up control of their proprietary assets. Combining outside information with internal records can provide a fuller picture, but it also raises questions about who can access the data, what can be shared, and how the information will be used.
These concerns make privacy and governance as important as the technology used to connect datasets. Here, members of Forbes Technology Council share strategies for combining external and proprietary data while maintaining privacy, security and organizational control.
1. Set Clear Limits On External Data Use
Treat external data as a rented input with a return path. Pull it in only as narrow, purpose-built signals, and build in a way to send it back: a documented use, an expiry and a kill switch. Privacy and control break less frequently on the first join than on the third reuse, when nobody remembers what the data was cleared for. – Stefan Gogoll, andagon Holding GmbH
2. Move Analysis To The Data
Every platform starts with “move your data here.” Moving data is the most painful job in the business. Do not move the data to the intelligence. Move the intelligence to the data and extract the meaning. Send the question in, compute there and bring back the answer. The value was never in the rows. – Prafull Khare, Radiant Digital
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
3. Use Clean Rooms To Match Data Securely
Data clean rooms and federated queries let teams match external signals against proprietary records without either side handing over raw data. Pair that with clear governance on who can query what and log every access. The result is richer insight, with sensitive information staying inside the company’s own walls. – Yardley Pohl, interos.ai
4. Map Sensitive Fields Before Joining Data
Clean rooms and federated queries protect data in transit, but the real privacy risk sits in the join. Before combining outside data with yours, know which of your fields are personal, which keys link records, and how reliable those links are. A clean room can’t fix a bad join, so map your own data first, then decide what may be matched and by whom. – Frederik Riskær Pedersen, Maplith
5. Use Public Threat Intelligence To Spot Risks
Companies should focus on open-source intelligence instead of acquiring massive, intrusive datasets. Companies can often go much further by making better use of publicly available information. Cross-referencing internal anomalies with publicly available threat signals can identify credible risks earlier so teams can act on them without resorting to deep surveillance or giving up control of their data. – Jonathan Graff, Liferaft
7. Unify Data Through Governed Platforms
A customer data platform is a data orchestration layer that aggregates and unifies external and proprietary data in a single view. Data is pulled together only when needed and is written back to the original source so, regardless of location, all key information remains under your organizational governance, with privacy and control intact. – Martin Taylor, Content Guru
8. Restrict Secure Data Analysis To Approved Queries
In our banking work, customer data often cannot leave the client’s environment. So, we send the analysis to where the data sits and only the aggregated results come back. Each party keeps control of its own data. What makes it work is agreeing in advance on a short list of approved questions and who may ask them, so privacy is settled before anyone runs a query. – Dimitar Dimitrov, Accedia
9. Train AI Models With Federated Learning
One effective approach is to use federated learning, which enables companies to train AI models across decentralized data sources, both external and proprietary, without sharing or exposing raw data. This method lets organizations gain broader insights while keeping sensitive information secure on their own systems, preserving privacy and regulatory compliance. – Will Conaway, Tuxedo Cat Consulting
10. Combine Data Only During Execution
Combine the data at the moment of execution, not before it. The instinct is to pool external and proprietary data in a governed space, then work on it. Change the timing instead. Fuse the two inside one governed execution step, scoped to that action, with every access logged and removed when the step ends. Nothing persists. The riskiest thing in most architectures isn’t the query; it’s the combined dataset left sitting afterward. Don’t create one. – Corey Ercanbrack, Vasion
11. Choose Anonymization Or Tokenization Based On Sharing Needs
There’s no single control; it depends on the business need. For one-way sharing, anonymize data and restrict access to what’s strictly needed. For two-way collaboration, tokenize identifiers so you can map results back to internal records while partners never see who or what is behind them. In both cases, clean rooms or shared workspaces add need-to-know access to anonymization and tokenization. – Kim Bozzella, Protiviti
12. Create Synthetic Versions Of Proprietary Data
Build a synthetic twin of your proprietary data: an artificial version that preserves statistical patterns without exposing real customer records. Join it with external data, test it with partners, and train models on it. Before sharing, test that no synthetic record can be traced to a real person. Poor generation can still leak information, and rare edge cases may be missed, so validate findings on the real data internally. You keep control of the originals. – Gayathri Kolandaisami, Quantumzet Technologies
13. Combine Data Through Governed Digital Identity Profiles
Rather than moving raw data between systems, combine it inside a governed corporate digital identity profile. For example, real-time data from authoritative public sources is retrieved on demand and combined with a bank’s private customer information. Every data point is sourced and timestamped, proprietary data stays in the bank’s control, and AI gets accurate, auditable data via permissioned access, supporting both innovation and regulatory review. – Alex Ford, Encompass Corporation
14. Find Matching Records With Private Set Intersection
Use private set intersection when the useful part of the data is the overlap. If a company wants to compare its customer list with an external dataset, for example, it can identify or calculate something about the records they have in common without handing over the rest of either dataset. That means you can get the signal you need without creating a new copy of everyone’s data somewhere else. – Elton Chan, Second Talent
15. Define The Business Decision Before Combining Data
I’d start with the decision the data is meant to improve. Once that is clear, companies can bring the external signals they actually need into a controlled environment and limit how much proprietary information is exposed. I’ve found that starting with the use case usually leads to a simpler, more responsible data system. – Deniz Gezgin, Eightpoint
16. Create One-Way Paths Into Air-Gapped Systems
Build a one-way door into an air-gapped system. Outside data lands first in a separate holding area, where its source, license and quality are checked. Only then is it read into the air-gapped environment that holds the proprietary data, with no route back out. Shared clean rooms ask companies to trust a space they do not fully control. A one-way door asks them to trust nothing, and every insight stays inside. – Vivek Thomas, AISensum
17. Protect Original Records From External Changes
Companies should prevent external information from automatically replacing trusted proprietary records. Outside sources can enrich business decisions, but they may contain conflicting or inaccurate values. Companies should keep original internal records intact and require approval before accepting material changes. This approach allows companies to use external insights while retaining control over the accuracy and integrity of their own information. – Salice Thomas, Wipro Limited

