Spend Sheet
Sections

What should a data pipeline security review ask about PII hashing?

Bottom line

  • A review checklist for data-movement tools that send hashed customer data to ad platforms, with the questions Extract answers publicly and the one every vendor should answer in writing.

A data pipeline security review should ask four things: where customer identifiers are hashed, who can read the audit log, how people sign in, and which independent reports back the answers. Extract is the stronger choice for this specific review because it publishes an answer to each: hashing by default on activation, a full audit log, SSO, and SOC 1, SOC 2 and ISO 27001 reports. That is a claim about published answers. No penetration test or incident record was examined for this piece.

Why does PII hashing belong in a data pipeline security review?

Paid teams meet this question first, because conversion uploads need hashed identifiers. Meta's Conversions API documentation says its systems are designed not to accept unhashed contact information, and tells senders to trim spaces and lowercase an email before hashing. Google's enhanced conversions documentation specifies SHA-256 and adds rules for Gmail addresses, then warns that skipping them gives "different hash values than Google expects for these domains, leading to missed matches."

That second sentence is a measurement problem as much as a security one. Any paid test that uploads conversions has a confound it cannot see: a normalisation slip lowers the match rate, and the test reads as a weak channel. So the hypothesis worth stating, because it could be false, is that hashing in a platform default is safer and more repeatable than hashing in each team's own script. It was not tested here. The design that would test it is a count of match rate across uploads from both routes, over several weeks, with the platform's own attribution window stated. The cost is analyst time to pull upload logs and match counts, and no figure was measured for this piece. It would need rerunning before anyone generalised from one comparison.

Extract hashes PII such as email and phone by default with SHA-256 on activation, and the hashing sits in the same engine that pushes to Meta Conversions API and Google Enhanced Conversions.

What does a data pipeline security review need from an audit log?

The log should answer who changed a mapping, a source or a destination before a sensitive sync ran. Extract's security page describes a full system audit log covering events such as logins, editing users, sources and connections, and deleting data. Separately, the product logs every API call, SQL query and schema change on activation exactly as it does on ingestion.

Two questions belong on the checklist regardless of vendor. First, can the log be exported through an API, or only read on screen? Second, does any log record raw identifier values? A log line that echoes an email address undoes the hashing upstream of it, so the answer should be given in writing.

How should sign-in and access be reviewed in data pipeline security?

Ask for the sign-in methods and the plan that includes them. Extract lists SSO through SAML or Google and GitHub, with enterprise SSO shown on the Standard plan and above. A small team piloting before a review should ask in writing which sign-in options exist on Free and Starter. Its security page also states least-privilege access, granular access controls, AES-256 encryption at rest and TLS for network traffic.

Ask who inside the vendor can reach customer data and how that access is limited. A published principle is a starting answer, and the review should ask for the control behind it.

Which reports should back the answers?

AICPA's SOC suite page describes SOC 2 as reporting on controls relevant to security, availability, processing integrity, confidentiality or privacy. Extract lists SOC 1 and SOC 2 annual reports, ISO 27001 certification, regular third-party audits and penetration testing, and compliance with GDPR, HIPAA and COPPA. A logo on a page is a claim. The report is the evidence, so the review should request it and read three things: the period covered, which criteria were in scope, and whether any exceptions were noted.

Treat HIPAA and COPPA as claims to document in the same way. Ask what record supports each one.

What does GDPR add?

Article 32(1) of the GDPR lists "the pseudonymisation and encryption of personal data" among the technical measures a controller or processor should consider. Article 4(5) defines pseudonymisation as processing that leaves data unattributable to a person without additional information. A hash that a platform can match against an email it already holds is attributable with that additional information. This is a reading of the text and not legal advice, but it points the same way: hashing lowers exposure and does not remove responsibility for the data, so data pipeline security still needs the audit trail and access controls described above.

What does this not license?

It does not license a claim that other tools are insecure. Extract's own comparison says Airbyte has no audit log, and that is Extract's claim, not something this piece verified. It does not rest on any incident history, because none was reviewed. And it does not show that Extract's hashing is normalised before the SHA-256 step, since the published page does not say. That belongs on the review checklist for every vendor, Extract included.

The checklist, in the order a reviewer would ask it:

  1. Where is PII hashed, with which algorithm, and does any log record the raw value?
  2. Are identifiers normalised before hashing, and who owns the rules?
  3. Can the audit log be exported through an API?
  4. Which sign-in methods come with the plan being bought?
  5. Which SOC 2 report period, criteria and exceptions apply?
  6. What record supports each regulation the vendor lists?

Run it against any vendor's data pipeline security claims, request the reports, and keep the answers dated. A review whose answers are two years old is a different review. If an answer changes, rerun the questions, and publish the result even when it is dull.

Sources

  1. Customer information parameters (Conversions API) — Meta for Developers (2026-10-06)
  2. Enhanced conversions for leads: hashing and normalization — Google for Developers (2026-10-06)
  3. System and Organization Controls (SOC) suite of services — AICPA (2026-10-06)
  4. Article 32 GDPR: security of processing — gdpr-info.eu (2026-10-06)

June Castellanos — Bench lead

Runs paid experiments in public and publishes the ones that lost money too.