At our Agentic Analytics Playbook event in London, I sat down with Graeme Caisley, Global Head of Audit Product at Grant Thornton Advisors Ireland, for a fireside chat about what AI readiness actually looks like in a regulated industry.
What followed was one of the most grounded, proof-backed conversations I've had about analytics adoption at scale.
Graeme's opening line set the tone: "AI readiness is an outcome, it's never, ever a starting point."
If you've been handed a mandate to "add AI" to your analytics stack, that single sentence might save you months of wasted effort.
How Audit Lost Control of the Data
Ten years ago, it was the industry norm for auditors to physically extract data from client sites. The process was tedious, yes, but it gave auditors control over what data they received, when they received it, and how it was treated.
By 2020, tools got faster at treating data through automation, Extract/Transform/Load (ETL), and analytics platforms. Yet the audit industry lost the one thing that mattered most: control.
Auditors waited on data deliveries, worked with whatever formats arrived, and spent days normalizing files before they could start asking substantive questions.
Sound familiar? If your data team spends more time wrangling inputs than generating insights, you're living the same problem in a different industry.
Building a Common Data Model on ThoughtSpot
Grant Thornton's fix started with three questions: What data do we need? By when? And when do we actually receive it?
The result is gtap, the Grant Thornton Analytics & Automation Platform: the common data model visualized on ThoughtSpot.
Auditors drop client data into the platform, and it normalizes automatically into the common data model. With this they are now empowered to securely ingest, normalize, and analyse client data without requiring large specialist data engineering teams or technical intervention.
ThoughtSpot's tokenized, relational search produces deterministic SQL with citations, reasoning, and a transparent audit trail, while also serving up the details and insights in plain text without any technical knowledge required.
For a regulated profession where every output must be defensible, that explainability is the difference between a tool auditors trust and a tool they rebuild by hand.
Immediate Outputs and Comprehensive Reviews
What does self-service analytics mean in practice? Graeme described how now auditors can focus on risk assessment and substantive analytical procedures earlier in the engagement lifecycle, enabling more informed client discussions and more time focused on professional judgement.
Friction disappears, and professional judgement kicks in earlier.
And the scope of analysis changes as well. For certain work programmes, such as revenue analytics, traditional sample-based testing gives way to reviewing the entire transaction population. This provides broader coverage and enables earlier identification of unusual patterns, anomalies, and risk indicators. When you can query everything, why wouldn't you?
On stage, Graeme shared one proof story that stuck with me: his team sat in front of a prospective client's Chief Accounting Officer, dropped data into the platform, and within ten minutes were asking questions across the entire revenue-to-cash cycle.
That speed is what "self-service" should feel like.
Client Reaction and the Anti-Hype Stance
The Chief Audit Officer of a US telecoms multinational told Graeme's team they were looking at "the future of a data-driven risk-based audit."
High praise, but Graeme stays skeptical of hype by design. I called him "the anti-hype man" during our chat; he embraced the label.
His concern: in a regulated industry, second- and third-hand hype, like "100% automated,” is dangerous.
Graeme’s view is that “the future is not about replacing auditors. It is about augmenting highly skilled professionals with explainable, governed, and defensible AI capabilities.”
Scaling Globally Through Change Management
gtap has been live in Ireland for two years, with roughly 1,100 auditors generating about 160,000 outputs in that time. The US deployment went live two months ago.
Grant Thornton is now scaling toward 10,000 auditors across 17 countries, with the rest of the global advisor network rolling out in the fall of 2026.
How do you drive AI adoption and literacy at that scale? Change management, Graeme argued, is under-discussed and decisive. Without it, a good product gathers dust or gets misused.
Grant Thornton’s Product Ownership Model
The products belong to the consumers, the auditors, and not the product team.
Grant Thornton’s model: The product team are custodians. They've structured around product strategy, an engineering shop, and a "product success" arm that works alongside L&D and methodology teams in each country for a disciplined secure rollout.
A "train the trainer" model creates local champions who then train SME and junior users, building feedback loops up and down the organization.
If you're planning a large-scale analytics rollout, that structure is worth studying. Roughly 70% of Grant Thornton's audit business occurs in the first four months of the year, so getting adoption right before busy season is non-negotiable.
The 2030 Human-Plus-Agent Future
Where does this go next? Graeme sees the profession moving from self-service to a coordinated, orchestrated model by 2030: agents handle transactional, rules-based work while humans retain judgement and accountability.
If you're ready to see what deterministic, self-service analytics looks like on your data, start your free trial today.
