Earlier this month I attended SAPHILA 2025 in Sun City – a conference where SAP professionals, customers, and partners gather to explore emerging SAP tech and how it can be used to solve real-world problems. Of course, AI was woven into just about all of the presentations, but one of the demonstrations I saw that made an impression showcased the newly released SAP Joule for Consulting — SAP’s AI-powered digital assistant.
SAP Joule is essentially a chatbot that uses natural language processing and large language models to pull answers from SAP’s internal knowledge base. But what really caught my attention — and what I feel is sorely missing from a lot of generative AI content today — was Joule’s transparent confidence scoring.
Lately, I’ve become more sceptical of AI-generated content — especially thanks to the recent addition of natural language summaries that we now see in response to a search. These summaries often use information from unreliable sources, but because the language is so polished, they come across as authoritative.
We can probably all agree that generative AI is very good at sounding like it knows what it’s talking about — and that can be useful when we’re trying to appear confident in a subject we don’t know well. But that same fluency becomes a problem when it masks uncertainty.
What I feel is really needed is a way for the system to signal how certain it is about what it’s saying. Wouldn’t it be great if, at a glance, you could know whether your search engine is pretty sure — or just taking a stab?
This is where Joule appears to be different — instead of just giving you an answer, it also gives you a sense of how confident it is. It’s a small feature, but it makes a big difference when one needs to quickly decide whether to trust the response and move ahead — or carry on looking.
As consultants, we should already be equipped to identify misinformation in our field of expertise, but when AI tools serve up unqualified information from unreliable sources, instead of helping us to be more efficient, it forces us to double-check everything before we can move forward. Confidence scoring helps shift AI back into the role of a genuine accelerator.
Another nice but subtle feature of Joule’s confidence scoring is that it avoids raw percentages (which can be confusing) and instead uses plain-language categories like Recommended, Best Option, or Low Confidence. Better still, if it’s really unsure, it doesn’t answer at all — it just presents you with potentially relevant links.
SAP Joule for Consulting incorporates transparent confidence scoring — a subtle but sensible feature that makes it a handy and supportive tool for consultants, saving us time rather than trying to mimic us with false certainty. In consulting, we don’t need AI to sound authoritative; we need it to provide reliable answers — or clearly say when it doesn’t know. This kind of transparency isn’t just helpful, it also ensures that responsible governance is applied when leveraging AI to accelerate consulting.
