Easy Isn't Reliable: What AI Doesn't See in a SAP Analytics Cloud Model

This isn't an argument that AI is bad at building dashboards. It's a demonstration of one specific thing a prompt cannot do — told through a model that fooled the person best placed to catch it: me.

The model that was wrong — and where it came from

A SAP Analytics Cloud model I was relying on produced regulatory ratios that didn't add up. Tracking the cause back was uncomfortable: the flawed structure came from a template I had published myself. It looked complete. It imported cleanly. It ran without a single error. It was still wrong.

That's the whole point. The failure mode that matters in SAC isn't the model that breaks — it's the model that works perfectly and reports the wrong number.

The part AI does well — genuinely

Ask a capable model for a planning structure and it will give you something reasonable: a Version dimension with Actual, Budget and Forecast, a few business dimensions, a measure or two. For a straightforward internal report with no regulatory logic, that is often good enough, and pretending otherwise would be dishonest. It's one reason plain reporting templates aren't where the real value sits.

Where it breaks in silence

Here is the concrete failure. Solvency and liquidity ratios — SCR coverage, LCR, NSFR — routinely exceed 100%. A well-capitalised insurer can sit at 180% SCR coverage; that's healthy, not an error. But a model can be configured to treat those measures as percentages capped at 100%, silently clipping every value above the ceiling.

Nothing about that setting looks wrong. The model imports, aggregates, and renders a clean dashboard. The numbers are just quietly amputated — a 180% ratio shows as 100%, and the story reads as if the insurer is at the regulatory floor. An AI generating this model optimises for a result that looks finished. It has no mechanism to know this measure must be allowed to exceed 100%, because that isn't a coding fact — it's a regulatory one.

The fix is one property on the measure. Knowing that it needs fixing is the entire job.

What a prompt never sees: the tenant

There's a second layer a prompt can't reach. A CSV that's perfectly valid on screen still has to survive a real SAC tenant: mapping each column, coexisting with the system Version dimension that SAC adds automatically alongside your own, and an upload job whose behaviour doesn't always match what its buttons imply. That friction is invisible until you're inside a live tenant — which is precisely where a tested accelerator has already been.

What about SAP's own Business Content?

The same honesty applies here, and it points the other way. SAP's Business Content packages are genuinely good — and if you're already connected to a live SAP source system, they're often the better choice. This isn't a competitor to that.

The difference is where the buyer starts. Business Content assumes the source integration already exists. A standalone accelerator serves the moment before it: modelling, testing, or getting a first planning model live without waiting on the plumbing — imported by file, self-contained, ready to populate. For a company already live on S/4HANA wanting standard reporting, SAP's content wins. For everyone standing something up before that point, it doesn't help yet.

So what is the accelerator actually for?

Not the CSV. A prompt generates that in seconds, and no one should pay for it. The value is the layer underneath: values checked against real regulatory logic, a model proven inside a live tenant rather than theoretically valid, and a knowledge document explaining why each threshold and aggregation is shaped the way it is.

That's what the validation passport on each regulatory accelerator records — not marketing claims, but what was actually verified. If a template I built myself could cap a solvency ratio at the wrong ceiling and still look complete, the difference between "looks right" and "is right" isn't a slogan. It's the product.

Note: independent project, not affiliated with or endorsed by SAP SE. Accelerators save setup time; you supply and validate your own figures. "SAP" and "SAP Analytics Cloud" are trademarks of SAP SE.