Only 28% of inspected audits met the relevant audit standards in 2025. So finds IRBA, the Independent Regulatory Board for Auditors, in its latest public inspections report.
It's a jaw-dropping statistic, and it's tempting to immediately vilify: auditors must be missing risks, misunderstanding controls and making mistakes in their reports! But that's not where the failures are. The main contributor isn't planning, disclosures, or the conclusion phase of an audit. It's the hard grind in the middle: building workbooks, performing testing and compiling evidence. With two of our founders being ex-auditors, we remember it all too well. Audits aren't failing because auditors can't think. They're failing because there's too much grind and too few people left to do it.
And there are fewer of them every year. IRBA reports the number of auditors fell 14.9% from 2019 to 2025. Regulatory pressure has climbed sharply (thanks, Steinhoff and VBS), and the data underpinning financial statements has exploded in volume and complexity. Auditors have more work to do than ever before, yet fewer auditors to do it. The trends do not paint a pretty picture.
It's not only a local story. The nature of auditing has grown more complex everywhere — reporting frameworks and internal control environments more intricate, regulatory expectations higher, and the bar for expertise raised with them. And the profession is thinning globally: in the US alone, over 300,000 professionals have left the field, with CPA exam participation falling alongside. A demanding, time-intensive path has struggled to compete with more dynamic industries for the people entering it.
The world's leading firms have recognised this and are placing bold bets on artificial intelligence — a clear signal that change is no longer optional.
- PwC (US) has committed $1 billion to generative AI, including the launch of its internal platform ChatPwC.
- EY has announced a $1.4 billion investment in technology and AI.
- Deloitte has introduced PairD, supporting professionals with everyday tasks like email drafting, coding, and research.
- KPMG, through its alliance with Anthropic, is embedding Claude into its Digital Gateway, scaling AI capabilities across 276,000 employees.
These investments mark a shift from experimentation to real adoption. As the models mature, becoming more accurate, more reliable and better equipped to handle complex inputs, automating the execution layer of audit is no longer a question of whether. It's a question of who gets to be part of it. The firms with a billion dollars to spend have their answer. Everyone else needs another path.
So how do we shape this for the better?
According to a 2020 Wolters Kluwer survey of accounting firms, the top five most important areas when completing audit engagements look like this:
- Obtaining all necessary data / documents from clients43%
- Data entry and financial statement preparation39%
- ‘Plain English’ wording in checklists and forms36%
- Roll forward key data and update to newer content30%
- Efficient financial statement disclosure checklist with examples28%
Source: 2020 Wolters Kluwer survey of accounting firms
That's why we started Relay. We didn't begin with the technology, we began with the engagement: what makes one flow, what makes one stall, and what makes an auditor tick.
Picture the audit we're building. The client's trial balance and supporting workbooks land in your data room. Instead of three days of keying in figures and chasing them across a spreadsheet, you hand your folder of support over to Relay. A first draft workbook is compiled in minutes — every number tied to a source and referenced, with exceptions surfaced and awaiting your professional judgment. The risks and materiality set during planning stay top of mind and are carried through testing, keeping the work consistent and complete. The client changed their figures? No problem, hand it back to Relay and update your work in seconds. One-click review a workbook, checking for any formulae errors, hardcoded figures, inconsistencies or material misstatement, with review notes being raised on the go. And you can guess what will happen with those review notes...
This audit doesn't exist yet. It's what we're building.
But can you trust it?
It's the first question any auditor asks, and the right one. You sign the opinion. When IRBA inspects the file, it's your name on it — not a vendor's, not an agent's. We've sat on that side of the inspection, so we're building Relay around three things we won't compromise on.
It will never conclude. Relay does the work and shows you everything behind it — but the judgement, the sign-off, and the opinion stay with you. The one thing you can't delegate is the thing you're liable for, so we're not even trying to.
Everything it produces will be checkable, and checkable fast. Every figure traceable to the source it came from; every test recorded — what it looked at, against which standard, and what it found. If you can't follow how a conclusion was reached, you can't put it in a file. So we're building Relay so that you always can.
When it isn't sure, it will say so. Where the evidence is thin or the data won't reconcile, Relay flags it for a human rather than papering over it. A junior who guesses to look productive is a risk; we're building the opposite — surface the doubt, hand it up.
And the methodology underneath stays current, because a procedure is only ever as good as the standard it's tested against.
The next generation won't be replaced — they'll be amplified
We believe the next generation of auditors won't be replaced by AI — they'll be amplified by it. Audit is evolving into a relay of agents and human expertise, with each passing the baton at the right moment. The agent does the grind; the auditor does the thinking. With that hand-off working seamlessly, auditors can focus their attention on what they do best: understand risk, exercise professional judgement, and continue to deliver trusted insights.
We're heads-down building the audit tool we always wished we had. Demo dropping soon. Come watch.