What growing companies can learn about scaling AI with ERP

There’s a growing debate right now about whether AI makes traditional ERP platforms unnecessary. To find out what’s actually happening on the ground, we sat down with Stefan Frühauf, COO and CFO of PwC Germany, about the firm’s own cloud ERP transformation — and how it shaped their approach to AI.

Is cloud ERP still relevant in the age of AI?

This is the question driving most of the “SaaS is dead” conversation right now, and it’s exactly what we asked Frühauf directly. His answer:

That view is backed by more than one data point. McKinsey’s recent analysis of AI’s impact on ERP addresses this scenario directly — sometimes called the “SaaSpocalypse,” the idea that AI will simply replicate ERP capabilities and make the underlying platform redundant. Their research points the other way: a clean, governed data and applications core is what prevents AI from becoming a sprawl of disconnected tools and inconsistent customizations, and early adopters of AI-integrated ERP are already seeing measurable returns, including EBIT improvements of 5 percent or more.

IDC’s latest AI Maturity research adds the market-wide context: 81 percent of organizations now have a documented AI strategy, but only 12 to 16 percent have reached meaningful AI-driven execution. The gap between those numbers is largely a data and governance readiness problem — precisely the groundwork PwC did before layering in AI.

Where is AI actually delivering value in ERP today?

Rather than speculating about AI’s future potential, it’s worth looking at where it’s already working. PwC shared one concrete example: AI-assisted VAT code classification on incoming invoices, reaching what Frühauf described as roughly 95 percent accuracy and an estimated 75 percent reduction in processing time compared to manual processes.

That result lines up with what IBM’s recent research on agentic data pipelines has found more broadly: automation can reliably interpret a request, locate the right data, and validate it against governance policy — but only once that data and governance already exist. IBM describes this as turning a process that used to take days into one that takes minutes, provided the underlying foundation is already in place. PwC’s VAT example is that principle showing up in a live finance operation, not an exception to it.

Should you standardize your ERP, or customize it?

This is one of the most debated questions among organizations evaluating ERP today, and it came up directly in our conversation with Frühauf. His answer reflects a framework increasingly seen across large-scale transformations: standardize by default, and customize only where there’s a genuine legal or regulatory requirement, or a clear, measurable case for business value. As he put it:

Frühauf shared a concrete example of where that business-value bar was met: time recording, something every PwC employee touches daily. Rather than leaving it as a standard, generic workflow, PwC invested in making that specific touchpoint as easy as possible—while holding the line everywhere else. The decision was governed by senior business owners, not IT alone, which is what kept the exception from becoming a pattern.

Why do most ERP transformations struggle?

The common assumption is that ERP projects stall on technology gaps — missing features, integration limits, migration complexity. Frühauf’s read on PwC’s transformation points elsewhere:

That’s the other half of the standardization story above: PwC could hold a strict default because it invested just as heavily in getting people to actually adopt it. Organizations that treat adoption and governance as first-class parts of the project, not an afterthought once the system is live, are the ones that see the transformation stick.

What does good AI governance actually look like in practice?

As more organizations start using more embedded and autonomous AI capabilities, governance stops being a compliance checkbox and becomes an operational requirement. PwC, as a network of audit and accounting firms, treats data security as core to the business. Frühauf shared the specifics of the model they built: territory data stays within its territory, access rights differ by management layer, and specific data is visible only to specific groups. PwC also encrypts its data using its own keys—meaning even SAP, as the platform administrator, cannot see PwC’s data.

What can other organizations take from this?

Few organizations operate at PwC’s scale—150 territories, 150 years of history, and what Frühauf described as “150 times 150 ways of doing the same business” before standardizing. But the underlying pattern is broadly applicable to any organization evaluating this path:

1. Standardize before you scale anything else—AI included. Fragmented processes produce fragmented outcomes, regardless of what technology sits on top.
2. Set clear rules for when customization is allowed—legal necessity or measurable business value, nothing else, governed by business owners.
3. Pilot in one place first—PwC proved the model in Germany before expanding globally, giving other territories a live reference rather than a theoretical rollout.
4. Build champions from real users, not just IT teams—PwC’s finance back-office staff became the advocates who helped other territories adopt the system.

Watch the session with PwC and hear ERP transformation lessons every growing company should know.

FAQ