How AI supports our key SOX and financial controls

Revenue stream scoping

A “skill” is a reusable workflow that AI recognizes and runs on request. We have a skill in CoCo that we run to scope our revenue streams across their key dimensions in near real time and email a snapshot on demand, saving 500+ annual hours. When consumption spikes, we see it early and engage stakeholders on key controls impact, so we get ahead of SOX compliance work.

Automated key order-to-cash controls testing

Selecting and testing samples by hand across high volumes of orders was slow and prone to error. Now, Cortex AI Functions, such as AI_EXTRACT, pull key terms from order documents, including revenue recognition terms, start and end dates and billing terms. A CoCo-built app validates them against our billing and revenue recognition system, so we test every order and catch anomalies as they occur.

Automated key reports testing

We needed insights on “true” audit-relevant changes to test in audit logs where it’s hard to parse through that information. Process complexity and a high reliance on manual report extraction, analysis and verification made decision-making complicated and slowed audit velocity.

We built an automated “Report-as-a-Service” (RAAS) architecture with CoCo and Streamlit that ingests audit trails, enabling continuous, data-driven monitoring and digital process rationalization.

This achieved 450+ hours of annual testing savings and enabled 100% population assurance with near real-time audit visibility.

Key access controls (IT general controls) monitoring

To keep privileged access and permissions aligned with our business policies, CoCo-built dashboards scan continuously for provisioning and deprovisioning exceptions, segregation of duties (SoD) violations and unauthorized privilege escalation. The dashboards flag anomalies as they happen across the full population, so we no longer pull samples for periodic reviews.

Automated reconciliation testing

Reconciling payroll used to mean matching three to four HR and payroll reports by hand across more than eight Excel workbooks. Now a data app built with CoCo does the heavy lifting and flags discrepancies automatically. Control owners benefit most, getting back 500+ hours a year from this app.

This same pattern can work for any reconciliation that pulls data from multiple reports.

SOX controls testing and documentation

Drafting SOX documentation and updating testing templates used to take weeks. Now we submit our templates and evidence to Petual.ai, which runs as a Snowflake Native App and drafts the updated documentation. Auditors review the final output, and turnaround is down to hours.

This also frees capacity for more impactful emerging and strategic high-risk advisory work and for building CoCo skills for continuous monitoring over key risks that drive valuable impact.

SOC 1 reviews

Reviewing third-party SOC 1 reports was one of our most tedious jobs: long documents combed by hand for control exceptions. The Petual.ai Snowflake native app has analyzed 20+ of these reports and identified relevant controls and potential exceptions automatically. Preparation time is down by days, and auditors spend their time on high-risk findings.

Operational audits now pull evidence from data in Snowflake

Our operational audits used to be hard to run at speed because evidence sat in fragmented sources and gathering it took walkthroughs and meetings.

With our data already in Snowflake, CoCo pulls control-specific information from across those sources quickly, and we spend our time validating it. We used to wait weeks to get and analyze that evidence. This approach gives us near real-time visibility into areas such as vulnerability management and the access and security health of key systems.

We have also built repeatable skills that track progress on observations from completed audits:

• Key security metrics for our Salesforce platform
• Vulnerability management across our production assets
• Security and governance metrics for key databases in our Snowflake account

More skills are planned as we start new audits. With CoCo, we have increased the number and complexity of the audits we run, and we deliver findings and reports faster.

Continuous assurance gives auditors more time for strategic risk

The role of the internal auditor is evolving from historical record-keeper to strategic risk advisor.

For me, the biggest change is in our conversations with business leaders. With AI handling the baseline compliance work, my team brings product, engineering and finance teams forward-looking, data-driven insights alongside our findings.

For audit leaders considering the same shift, continuous assurance makes four things possible:

• Broader assurance through full-population testing.
• Less friction with the business, because manual evidence requests shrink.
• More auditor time on strategic risk, culture and emerging threats.
• Governance framework that helps the business move faster, because risks are caught before they turn into issues.

I’m proud of what my team is building at Snowflake. With the right data foundation, even one of the most traditional functions in a company can change how it works.

Ready to rewrite your own audit playbook? Here’s how to get started:

• Start with one high-impact control: Pick a high-volume process, such as access reviews or order-to-cash testing, where full-population testing pays off fastest.
• Bring your data together in Snowflake: Land key cross-system data sets such as billing, revenue, HR, and access and audit logs in one governed place so you can test against full populations.
• Build with Snowflake CoCo and Cortex AI Functions: Turn manual evidence collection and reporting into repeatable skills and apps for continuous monitoring.
• Keep auditors in charge: Let AI prepare the work while your team reviews it and owns every conclusion.

Try Snowflake CoCo today or talk to your Snowflake team about continuous assurance for your audit function.

Want to see how finance leaders from Snowflake, DoorDash, Stripe and Northern Trust are putting AI to work?

Watch the Expedition 2026 finance sessions on demand.

Internal audit has long been seen as a backward-looking function that sifts through historical samples to confirm that controls worked. On Snowflake’s Internal Audit team, AI has changed that.

We have cut the time we spend on evidence collection, testing and audit report production by 50%. Overall, we have saved more than 3,500 hours in audit in 2026 and over 1,500 hours for control owners across the business.

Whether it’s reviewing system access logs or analyzing key financial transaction patterns, AI allows us to shift from a historical “point-in-time” check to continuous, automated monitoring. We no longer just identify what went wrong six months ago; we can surface key risks for specific controls and use cases and ask targeted questions.

Here’s a look at how AI is helping the Snowflake Internal Audit team rewrite the audit playbook, backed by real-world data and measurable impact.

Key takeaways

• Fieldwork time for evidence collection and report production fell by 50%, from months to weeks (internal results as of September 2026).
• Testing moved from samples to 100% population coverage and proactive insights for testing several key controls and reports, plus scoping use cases (order management, Workday key reports, revenue scoping, key access controls).
• Snowflake’s Internal Audit team saved 3,500+ hours in 2026 (1,000+ on risk scoping and 2,500+ on testing and documentation), and control owners saved another 1,500+ hours.
• AI handles data processing and initial analysis, with auditors accountable for every conclusion.

None of this would work if our data was scattered across disconnected systems. With billing, revenue, HR, access and audit log data already live in Snowflake, we can test against full populations without stitching sources together. We use Snowflake CoCo, the AI coding agent in Snowflake, to build the scripts, dashboards and apps that do this work.

Auditors make every final evaluation

As we push further into AI, trust and accountability matter more to me than any efficiency gain. AI is our accelerator and human oversight is our anchor. My team operates under a human-in-the-loop model:

• AI prepares, auditors conclude: AI handles data processing, pattern recognition and initial reconciliation. Auditors make every final evaluation and risk assessment and own every conversation with business leaders.
• Auditors review AI output before anyone relies on it: When AI drafts testing documentation or surfaces exceptions, an auditor checks it. Our time has moved from gathering evidence to validating it.
• Third-party tools run inside Snowflake: Tools such as Petual.ai run as Snowflake Native Apps, so their processing stays inside our Snowflake account, secure and isolated.

The hard metrics: driving efficiency and expanded assurance