We're building Snap because audit teams spend a lot of hours on the mechanical side of the work: vouching transactions to invoices, analyzing journal entries, documenting control tests. We think much of that time can be returned to the auditor.
That matters because the mechanical work is where junior staff currently spend the bulk of their early years. It's necessary work, and it has to be done well, but it doesn't develop the skills that make a great auditor. The skills that do: professional skepticism, risk assessment, evaluating evidence, navigating client conversations. Those get developed later, when there's time left over after the mechanics are done.
PwC's AI Assurance Leader Jenn Kosar made this point well in an interview with Business Insider last year: "People are going to walk in the door almost instantaneously becoming reviewers and supervisors," she said, describing how AI is changing what PwC expects from first-year hires. "Three years from now, we will feel like the first years are functioning more like fourth years." The premise is that AI handles the routine, repetitive work, and junior staff move directly into the judgment-heavy work that used to come years into the career.
We think this is broadly good for the profession, not just for the firms running the AI. When the mechanical work is offloaded, several things follow. The auditor's time shifts toward higher-value work: judgment, advisory, client relationships. The work itself becomes more interesting, which makes the profession more attractive to people deciding whether to enter it. As output per auditor rises, so does the value of the work, and over time that flows through to compensation. And critically, audit quality has room to improve, not just because AI can be more consistent on the mechanical side, but because the human auditor has more time to spend on what actually requires a human.
We started Snap to build the tools that make this shift practical for firms that aren't PwC. The work has to be reviewable to be trusted, which is why every testing line we produce shows the reasoning behind the conclusion and links directly to the source document it came from. The reviewer, whoever that is on a given engagement, should be able to verify any result quickly.
This blog is where we'll share what we're learning as we build it. Some posts will be about regulatory developments. Some will be about specific procedure types and how they're evolving. Some will be about the practical questions of getting AI-generated workpapers to a standard that holds up to review.
If you're working through any of these questions at your own firm, we'd be glad to talk.
Want to talk about this?