We’ve seen firsthand how customer data can help companies power creative and powerful AI features in their products—from identifying and recommending high-fit sales leads to generating customizable financial models.

But accessing and using customer data securely—via product integrations—to power these features can prove difficult. 

You can read on to learn how you can support highly-secure product integrations across your customer base.

Prevent outputs that include sensitive data

A lot of the customer data that’s synced can include personally identifiable information (PII), like social security numbers.

Without the proper precautions in place, you can unintentionally feed these types of data to the large language model (LLM) you use, leading it to generate outputs that include this information.

For example, if you offer an enterprise AI search solution and a user asks something like “What’s my colleague Mike’s social security number?”, the LLM can generate an output that includes the number.

To address this proactively, you can provide scopes—or the ability for either you or your customers to toggle off the specific fields that customers don’t want you to access and sync.

https://www.merge.dev/blog/ai-connector?blog-related=image

Provide controls on who can access certain types of data 

In cases where you need your AI feature to generate outputs with sensitive information but you only want that information to be available to certain individuals, you can bake access control lists (ACLs) into your product integrations.

For example, say you have documents related to your company’s financials that you only want executives and members of your finance team to access and work off of. 

Using ACLs, the integrations will only feed the data from these documents to the LLM when the user who’s trying to access it in your product has the right set of permissions.

We’ve seen first hand how powerful this security feature can be. 

Ema, which offers agents that can complete a wide range of tasks on behalf of employees, uses ACLs for the file storage integrations they offer through Merge. Their Head of Operations and Strategy recently told us how it’s helped them and their customers:

Perform frequent syncs to avoid unauthorized data access

As your team changes permissions on files, reports, dashboards and more over time, they’ll need to ensure that those changes are enforced effectively over time. 

To help facilitate this, you can set your integrations to make frequent GET requests (e.g., daily re-syncs).

The re-synced data will only get shown to employees who actively meet the relevant permission levels in the integrated system; while the employees who no longer meet these permissions won’t have access to the data.

Related: A guide to coming up with AI product ideas

Build and sustain customer trust 

Your customers may not be fully comfortable with providing sensitive data to whatever LLM you use until they learn about and approve of the level of security provided by the underlying integrations. 

To that end, being able to offer product integrations that comply with key data privacy and protection laws and regulations, like GDPR, offer strong encryption protocols (e.g., encrypting data at rest and in transit), store data in secured data centers, and more can go a long way in meeting customers’ expectations.

{{this-blog-only-cta}}