Security & Privacy

🏣 Intro

The Census synchronization and publishing pipeline is designed around the core principle of storing and handling as little of your sensitive data as possible. By default, Census is able to determine what records need to be synced within your own data warehouse.

When your customer data is handled by Census’ application servers, we use ephemeral workers that do not store your data, and practice data defense-in-depth with multiple layers of data scrubbers (managed by our platform and our cloud computing partners) to ensure that none of your sensitive data is mistakenly left behind on our servers or in our cloud storage.

This page describes how Census syncs work, in which cases Census will handle or store your sensitive data, and the various techniques the Census platform employs to minimize the risk of a data breach.

Our architecture is always evolving to support higher performance publishing to more destinations, but our commitment to handling your data securely remains our north star, and we believe the best way to ask for your trust is to store as little of it as possible.

📛 Compliance

We are compliant with SOC 2 Type 2, HIPAA, GDPR, CCPA and, Privacy Shield.

In addition to these certifications:

• We have regular third-party penetration testing (contact us for the latest report)

• Have automated vulnerability scanning in our platform

SOC2 Type 2 Audit Report

The latest copy of our SOC2 audit report can be requested by existing Census customers here. If you are currently trialing Census, your point of contact can provide you the SOC 2 audit report under NDA.

ℹ️ How it works

Step 1: Identify Changes in Warehouse

As part of onboarding, you’ll create a Census schema within your data warehouse. This is a bookkeeping area used by the platform to keep track of what changes have occurred to your data and what still needs to be synced.

Instead of storing this data in our platform, we keep it in your data warehouse for two reasons.

1. Improved security: in the event of a breach of the Census platform, your customer data is not at risk because we do not store it.

2. Efficiency: we can save load on your warehouse and provide a more cost-effective platform by avoiding unnecessary copying back-and-forth between your systems and ours.

In addition, Census will access your data warehouse through a dedicated gateway, which always uses an encrypted connection and originates from a set of well-known static IP addresses that you can add to the allowlist for your warehouse for additional control.

It’s impossible for Census to read data from your warehouse that you don’t explicitly opt-in, and if you choose you can even set up fine-grained table-level access for the Census user. When a new sync starts, Census looks for rows of data that have been added or changed since the last sync, and records a temporary snapshot of those rows in the Census schema.

Your warehouse reports back to the Census platform only the number of rows that have changed or been added - the actual data is not copied to Census servers.

Step 2: Unload “Diffs” to Cloud Storage

Once the data differences, or “diffs”, have been calculated, Census instructs your warehouse to copy just those rows to our cloud provider’s object storage bucket (either AWS S3 or Google Cloud Storage).

The temporary credentials we provide to your warehouse that are used to copy data are only capable of writing data, not reading it back out, so this is a one-way data flow. Diffs are assigned cryptographically unique key paths in the cloud storage bucket, making it impossible for an attacker to guess paths to customer data.

The cloud storage buckets used by Census are configured with two additional security measures, managed by AWS and Google Cloud.

• Items in these buckets are automatically removed after 7 days. This is a fallback if our platform doesn't remove this temporary data as soon as it is processed by a sync.

• Data in these buckets are encrypted using the cloud providers’ server-side encryption, protecting your data from some classes of attacks on the cloud providers themselves.

Step 3: Prepare Data for Loading into SaaS Applications

In most cases, the data from the warehouse cannot be loaded directly to the SaaS application without changes.

Census needs to adjust data types and formats, apply your mapping rules to translate columns in the warehouse to fields and attributes in the SaaS app, check to see if the data already exists in the SaaS in order to decide whether to create or update data, and handle any errors or data validation issues.

Instead of bringing the data into the core Census platform, Census starts up a fleet of stateless, ephemeral mappers (similar to “Lambdas” or “Cloud Functions”) that each pull in one batch of diffs and apply it to the service.

Mappers use separate temporary credentials that are only capable of reading data from object storage for least-privilege isolation. Once a mapper finishes its work, it deletes the diff batch from cloud storage and reports its results (see step 4).

Step 4: Report Skipped Records and Feedback to Warehouse

When a mapper finishes its work, it has two lists of records - those that were successfully loaded to the SaaS application, and those that failed, either because of validation issues in the data or transient errors (SaaS outages or errors, networking issues, etc).

The list of records that failed is written directly back to the Census schema in your data warehouse, and a small sample of those failures (no more than 100) are captured and sent to the Census platform. These diagnostic samples are the only customer data that will ever be stored in Census, and their storage is limited to 7 days.

Exceptions to our temporary data storage policy

As described above, Census's architecture is built from the ground up to handle your data with care and, by default, not to store any of your data on our infrastructure.

However, when setting up a new source, we do provide the option to use our Basic Sync Engine, which relies on Census-managed infrastructure for storing information about past syncs and computing incremental changes.

Also, the following data sources only support Basic Sync Engine (as of March 2024):

• Amazon Athena

• Amazon S3

• Azure Synapse

• ClickHouse

• Elasticsearch

• HTTP Request

• Google Sheets

• Materialize

• Microsoft Fabric

• MotherDuck

• MySQL

• SingleStore

Finally, enabling the Profile API requires us to store the relevant data on our own infrastructure. Census will remove any data stored on your behalf after 14 days of disabling this feature.

🤔 Have more questions?

We are happy to do an architecture deep dive or answer your security questionnaire to help answer any questions you might have about your data security.