How Job&Talent tested 150+ controls with Oxus in just three months
A conversation with Blai Sanchez, VP of Internal Audit and Risk at Job&Talent.
controls tested with Oxus in three months
efficiency gain in interim testing, including time spent developing controls and refining testing attributes
projected reduction in year-end testing hours
of interim testing work expected to carry forward at year-end
Led by Blai Sanchez, Job&Talent’s internal audit team brought AI-powered testing to more than 150 controls across a complex, multi-country environment. Oxus performed the initial testing and produced traceable working papers, allowing the team’s experienced auditors to focus on reviewing the results, assessing risk, and applying professional judgment.
This conversation has been condensed and edited for clarity.
Q: As head of the function, what made you decide it was time to bring AI into control testing?
Blai: Two things. First, AI is here to stay, and either we adopt and embrace it or we will be left behind — and if we fall behind, catching up will be a real struggle. Second, with AI on board we can be much more efficient in our testing and operational audits. That's why my focus over the last twelve months was on finding one tool that we can adopt to do what I would otherwise do with a larger team.
Q: Why Oxus rather than a general-purpose AI?
Blai: As much as you can build projects in general-purpose AI platforms, to test a control you have to go beyond that. It's otherwise challenging to replicate across different controls and different documentation support. That's where Oxus differs — how smart it is at dealing with different documentation and controls isn't something you can get from any general-purpose AI platform. Another factor that differentiates Oxus is the working papers: how you're able to document, trace and reference them. The working papers are very complete — they're the kind of working papers I would ask a team member to create. I have played with general-purpose AI quite a lot, but struggled to come close to producing working papers consistently and repeatedly at the same quality and standard that Oxus is giving to us.
“I have played with general-purpose AI quite a lot, but struggled to come close to producing working papers consistently and repeatedly at the same quality and standard that Oxus is giving to us.”
Q: How did you validate Oxus before expanding beyond the initial pilot?
Blai: We started small. The initial pilot covered around 15 controls that had already been tested manually, without AI. This gave us the opportunity to understand how Oxus handled different types of evidence, testing procedures and working papers before committing to a broader rollout, while also allowing us to compare the outputs directly against the work our team had already done: whether the evidence had been interpreted correctly, whether the testing conclusions were supported, and whether the working papers were clear and reviewable.
We also tested the practical side of it: how much auditor intervention was needed for the platform to produce the expected output, where the platform needed further refinement, and whether the process could realistically scale.
Once we were comfortable with the quality of the outputs and the level of intervention required, we gradually incorporated Oxus into our 2026 interim testing, while continuing to closely monitor how the platform was performing.
Q: How does Oxus fit into Job&Talent's control environment and testing process?
Blai: Our framework varies by country, with each region operating differently. What worked really well was the ability to adapt Oxus to our control framework and the particularities of each country. The team was also really surprised by how great the testing attributes suggested by Oxus were, and we adopted some of them. For a company that has a more standardized control framework, the opportunity to scale up could be really quick, and the benefits of using a platform like Oxus are exponential.
In practice, we treat Oxus as a junior team member: it runs the testing of the controls and prepares the working papers, and then a senior auditor reviews the results. The feedback we give is also incorporated into the platform — we are not correcting something and then starting from zero when we test the same control at year-end or in another jurisdiction.
“We treat Oxus as a junior team member: it runs the testing of the controls and prepares the working papers, and then a senior auditor reviews the results.”
Q: What results can you point to?
Blai: In interim testing we ran over 150 controls from selected processes, including ITGCs, through Oxus. While still adopting the platform, we've experienced a ~20% efficiency gain in interim testing and project a 30–40% reduction in year-end testing hours. By year-end testing, I expect we should be getting good results leveraging 75–80% of the controls and testing attributes that we built in interim testing.
Q: What has it been like working with the Oxus team?
Blai: We are really satisfied. We have had very productive, regular meetings with Oxus. They have been proactive about how they communicate with us: creating a Slack channel, responding to our emails before the next day even with the time-zone difference, weekly calls, and other calls whenever we needed them. When we share feedback or ask for something to be done differently, Oxus almost always addresses it in less than 24 hours, and that level of responsiveness has been exceptional.
Q: How should peers think about adopting AI?
Blai: Do not feel the obligation to transform your entire audit function from day one. Start with a real use case, maybe an operational audit or 10–15 mature controls from your ICFR or SOX Program, and see how AI performs: how it handles the evidence, how the testing works, what the working papers look like, and importantly, how your team interacts with it.
That gives you a relatively low-risk way to understand where the technology adds value, where it still needs human judgment, and what it would take to scale.
I also wouldn't wait for everything around AI to be perfectly defined. Of course you need the right guardrails around data, audit quality, traceability and professional judgment, but this isn't something I see as five years away. It's already changing how audit work can be done today.
And ultimately, the opportunity is bigger than efficiency. If AI can take away part of the repetitive evidence review, testing and documentation, auditors can spend more of their time understanding risk, challenging the business and applying professional judgment. That's where I think the real value is.