New in SAC: Prompt Insights

With the SAP Analytics Cloud update of August 15 and 16, 2026 (QRC3 2026), SAP introduced a way to bring Generative AI directly into your SAC dashboards!

With Prompt Insights, you add a text widget to your Story that runs a prompt you define on the data of a table. The result is a written analysis.
Good to know: it is generated at view time and based on the filters that are active at that moment.

In this blog, Kris will take you through how to set it up, how to get the most out of it, and where the current limitations are.
We tested it on a real-life purchase commitments report, so expect honest numbers, including the waiting times. Spoiler: the more you ask, the better the answer, but the longer you wait.

Note: As of October 2026, Prompt Insights only work on Tables.

What are Prompt Insights?

Prompt Insights use AI to answer predefined questions about the data in a table at runtime. The Story designer decides in Edit mode what the AI should analyze, for example:

• "Explain the main conclusion from this table."

• "Rank my suppliers by open purchase order value."

• "Which purchase orders have the highest open amount and pose a risk?"

In View mode, the end user can't change the question, but they can run it.
What does change is the outcome. The analysis is applied to the table with its current filters and hierarchy state: collapsed hierarchies will not be expanded for the analysis, and filtered-out values are not included.

That's the real difference with a chat-style AI: the designer controls the question, the user controls the context.

Prompt Insights vs. Smart Insights vs. Search to Insight

SAC already had AI-flavored features. So where does this one fit?

In short: Smart Insights explains one number, Search to Insight answers a spontaneous question with a visual, and Prompt Insights writes the commentary you'd otherwise type yourself.

How to set it up

Setting it up takes three steps. The first two are for your admins and only need to happen once.

Step 1 – Connect SAC to Joule (BDC admin)

Your SAC tenant must be connected to the Joule settings of your SAP Business Data Cloud (BDC) tenant. SAP describes this step by step here: Enable AI-Assisted Features for SAP Analytics Cloud | SAP Help Portal

Step 2 – Grant the Generative AI permission (SAC admin)

1. Go to Security → Roles.

2. Open the role of the users who will work with Prompt Insights.

3. Find the Generative AI permission and set it to Execute.

4. Save the role.

Step 3 – Add a Prompt Insight to your Story (Story designer)

1. Open your Story in Edit mode.

2. In the toolbar, under Insert, choose AI Prompt Insights.

3. In the Builder panel, under AI Prompt Insights based on, select the table the widget should analyze.

4. Choose a Scenario Type:

– Custom Prompt: the default, fully your own.

– Performance Review: predefined, SAC fills the Hidden Prompt with a Performance Review prompt.

– Risk Analysis: predefined, SAC fills the Hidden Prompt with a Risk Analysis prompt.

5. Fill in the Visible Prompt (max. 256 characters). This is the short description your end user sees.

6. Fill in the Hidden Prompt (max. 2,048 characters). These are the actual instructions for the AI: which data to use, which definitions apply, which exceptions matter and what the answer should look like. End users never see this field.

7. Under Settings, decide whether to Allow content from external sources (and Show Source URLs).

8. Save the Story, switch to View mode and click Generate.

How to work with Prompt Insights

Our test setup

We built a Prompt Insight on the detail table of a purchase commitments report: open purchase orders per supplier, with their open order value. Then we ran five variants, from a lazy one-liner to a strictly formatted prompt.

Note: These are indicative numbers from one tenant and one report. Your mileage will vary with data volume, model and tenant load.
Even though these numbers are indicative, they do show how much the format of your question matters: if you want a simple question and a simple answer, instructing the Prompt Insight on the output does have an impact!

What we learned

• A lazy prompt already goes a long way! A single line ("Explain the main conclusion from this table") produced a surprisingly good analysis.

• Three specific questions beat one vague one. The output got more useful and faster.

• External sources: performance dropped significantly.
With them switched on, the AI cited web articles that were plausible for the topic, but irrelevant for the report. Switching them off saved time without any visible loss of quality.

• The biggest speed gain comes from the output format, not the analysis. Telling the AI exactly what to return brought the response time down to ~15 seconds on a small set. The answer got less elaborate, but kept the essence.

• Bigger data, longer wait. Every optimization helped less as the dataset grew. That makes sense, but it's something to tell your end users upfront.

• Treat Prompt Insights like a colleague. Ask a vague question, get a vague answer. Ask for a deep-dive, and it takes time. Waiting 100 seconds for a text box feels long, but if the same analysis would take a colleague an hour, those 100 seconds suddenly look pretty reasonable. A big difference at the moment: a colleague tells you how they got there. Prompt Insights simply provide an answer.

The main trade-off: detail costs time

The most prominent trade-off in our tests: every bit of detail you add costs waiting time. If you know exactly what type of answer you’re looking for, it really cuts waiting time if you provide the structure within the hidden prompt.

• The question.
  Multiple questions, multiple measures and dimensions, more conditions, more history to consider: the AI needs longer.

• The answer you ask for.
  An explanation, a trend, recommendations: the AI needs time.
  A fixed format with only the numbers you require: a lot faster.

Logical, really: the AI has to read, calculate and write more. But it's good to know before you design a Prompt Insight, because it forces a choice:

• A quick check on a page users open frequently? Keep both the question and the answer small.

• A deeper analysis users run consciously, say once a month? The level of detail that can be provided is really impressive, but it takes time. This is fine, as long as users know they'll have to wait for it.

Note: Not all detail is worth its wait. In our tests, switching off external sources and strictly instructing the requested answer made it faster without losing anything we needed.

Our best-performing prompt

This is the format that gave us the fastest and cleanest results.

Known limitations

• Tables only. Charts are not supported (yet).

• Waiting time. Between ~15 and ~100 seconds in our tests. Most users won't wait a minute for a text box.

• Fixed questions. End users can't ask follow-up questions or rephrase the prompt.

• No reasoning shown. You get a conclusion, but not why the AI chose it. That makes validation harder.

• Hard to verify at scale. On our dataset, the answers seemed correct. With thousands of rows, checking the output is a job in itself.

• External sources can wander off. They may add content that has nothing to do with your report.

characters for the Visible Prompt, 2,048 for the Hidden Prompt. That's /ough, but not for an entire requirements document.

• No hidden tables. Linking a Prompt Insight to a hidden table doesn't work. The table has to be visible on the page.

Good to know: the analysis isn't limited to what your table shows on screen. With more than 500 rows, it still analyzes everything. You can also link multiple Prompt Insights to one table, and every filter that applies to the table (Story, Page and Widget) is picked up.

Final thoughts

The potential of Prompt Insights is huge. Having your tables explained, based on the data that’s visible for the user, with questions that matter most to your business, is exactly what AI is good at!

But it's not plug-and-play. Every extra bit of detail in the question or in the requested answer costs waiting time, so decide per Prompt Insight what's worth the wait.
The quality of the answer depends on the quality of the question.

Our advice: start small. Pick a few reports where users now manually read through large tables, and invest time in sharp, well-formatted Hidden Prompts.

Curious about what else was in SAC QRC3 2026? Check SAP Analytics Cloud QRC3 2026 – AI Integration & New Chart Types or read more about the three new chart types in New in SAC: Funnel, Gauge, and Sankey Charts

Are you interested in the Release Cycle of SAP Analytics Cloud? The Release Cycle of SAP Analytics Cloud in 2026: When does SAC Update?

Or would you like to know more about SAP Analytics Cloud or Reporting in general? Feel free to reach out to Kris van Brouwershaven.