Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
cancel
Showing results for 
Search instead for 
Did you mean: 
francois_imberton
Product and Topic Expert
Product and Topic Expert

Just ask is the new natural language query feature powered by AI. With just ask any user can query SAC data models by asking questions in everyday English. It supports SAP Analytics Cloud acquired data models and SAP Datasphere models

The initial release of just ask is version 2024 Q1 QRC or 2024.02.

Just ask is available in tenants running in Google Cloud Platform, Microsoft Azure, and Amazon Web Services data centers. Just ask is enabled for end-users through a tenant administration toggle. It is turned off by default. When turned on end-user will access just ask instead of search-to-insight.

Access to Just Ask

just ask is accessed directly from the SAP Analytics Cloud home page using the search bar and from light bulb icon in the main toolbar.

The search bar on the home screen is optional and can be enabled from the home screen settings.

SAC Home Screen.pngLaunching just ask from the home screen and the main toolbar

francois_imberton_1-1706453980559.pngJust ask main screen

Before Asking a Question

Selecting Models

Before asking a question, best is to narrow down the query by selecting one or more models. The available models are acquired data models that have been added to just ask and indexed by administrators.

francois_imberton_2-1706454980835.png

To query an SAP Datasphere model, end-users need to select a data model by selecting  Add Model to Search. Browse SAP Datasphere connections and spaces and pick an SAP Datasphere Analytic Model or Analytical Datasets. Once selected, the SAP Datasphere model will remain available for the duration of the session.

francois_imberton_3-1706455002679.png

Using Suggestions

Type ahead suggestions appear in grey while typing and can be used to autocomplete a word by hitting the right arrow key. Other suggestions can also be selected from the drop-down using mouse or keyboard: down & up arrow key and hit enter for selection.  Type ahead suggestions appear in grey while typing and can be used to autocomplete a word by hitting the right arrow key. Other suggestions can also be selected from the drop-down using mouse or keyboard: down & up arrow key and hit enter for selection.

francois_imberton_0-1706457838154.png

On the right-hand side of the screen suggested questions are based on the previous question.

francois_imberton_1-1706457838159.png

Measure & Dimension Picker
By pressing the + icon at the left of the search bar, the user can directly select the measures and the dimensions from the data models.

francois_imberton_2-1706457838162.png
Sample Questions

In addition to generated suggestions, administrators can define sample questions. These are available from just ask main screen.

francois_imberton_3-1706457838166.png

 

Exploring the Answer

From the natural language question just ask generates a technical analytical query and returns an answer as a visualization. At the bottom of the visualization the filters inferred from the query are displayed. In this example “How did client visits impact travel expenses last year in Denmark?” Last year generates a filter on year 2023 and in Denmark generates a filter on Country Denmark.

francois_imberton_4-1706457838172.pngAt the top of screen, buttons enable to switch to table, to open the visualization in Data Analyzer for further analysis and to copy to clipboard that can be pasted in an optimized story, (classic stories are not supported).

The ellipsis menu provides access to filters, sort, rank, csv export and additional options for tables and charts.

francois_imberton_5-1706457838176.png

 

 

Administration

Just ask administrators are accounts that are assigned a role with the just ask Managed permission. Built-in roles such as Admin, BI Admin, BI Content Creator or Modeler include this permission.

francois_imberton_6-1706457838181.png

 

Administrator Tasks

The main task of just ask administrator is to add models so that it can be queries by end users.

francois_imberton_0-1706463273847.png
Access to just ask model management via the “Managed models” button

francois_imberton_1-1706463273855.pngJust Ask model management area 

In addition, for indexed models. Once a model is indexed administrators can define synonyms and rules see details in the rollout section.

In this initial release only acquired models can be indexed. 

Indexed versus non-indexed models

Just Ask enables to query indexed models and non-indexed models  The table below summarizes the differences

 Indexed modelNon Indexed
Administrator not required to query model✔️
Administrator can define synonyms & rules✔️
Initial NLQ Scope✔️
Acquired models✔️✔️
SAP Dataspheres models✔️
Model selection by users required✔️
End user can narrow down query to selected models✔️✔️

Turning Just Ask On

Initially just ask is not active, end-users will keep using search-to-insights. This lets administrators prepare just ask before exposing it to end-users. To turn it on just ask for end-users, administrators will turn on the tenant wide Conversational Analytics toggle located in the Default Appearance tab in System Administration page.

francois_imberton_7-1706457838188.png

 

SAP Datasphere

In order for an end-user to select SAP Datasphere analytic models and analytical datasets, an SAP Datasphere connection must be created. End-user must be assigned an SAP Analytics Cloud role with the Connection:Read permission. End-user must have an SAP Datasphere user account, and to be a member of at least one space and permission to read models

More detail please refer to the documentation:
https://help.sap.com/docs/SAP_ANALYTICS_CLOUD/00f68c2e08b941f081002fd3691d86a7/ad4281e2875949f0b4d45...

Preparing for Rollout

Data Modeling

Before turning on just ask, administrators need to work with end-users, to identify relevant data and add to just ask the corresponding data models. In some scenarios, existing data models - that have been designed for professional story creators - may not be suited for end-users. Dedicated modeling work for end-user consumptions may be required. For instance, models may expose too many entities, entities with a meaning that require expert knowledge or conversely model may not include measure or calculations that are expected by end-users.

Synonyms

To complement modelling work, to tailor just ask to the terms end-users use and are familiar with, administrator can define synonyms to model entities and to dimension values,

Defining synonyms to a dimension name

francois_imberton_9-1706457838193.png

Defining synonyms to dimension values.

Rules

Rules may be used to provide relevant answers when end-user questions carry an implicit context not included in the question. For instance, a user may ask about the relationship between client visits and travel expense:” show me client visits and travel expenses” but he or she is looking for:” show me client visits and travel expenses by sales agents

Rule conditions are based on the terms used in the question, when condition is met and action can be to sort results, add a filter, and or remove an entity.

Define Rules.png

Auhorization

A data model together with its synonyms and rules target a specific audience. Such data model may be irrelevant and may confuse other end-users. Great care should be taken to ensure that only the target users have access to a given model. Firstly, by adding to just ask only models that are relevant for Natural Language Query and secondly by ensuring that model access rights are defined appropriately in SAP Analytics Cloud and in SAP Datasphere.

Rollout

Rollout may occur using a  phased  approach per audience and data models. In addition to the steps before rollout described above, a good practice to ensure successful adoption is to organize training sessions and material to help users grow their understanding of the data models.

 

17 Comments
xjhacking
Contributor

Hi Francois, 

Thanks a lot for detailing these steps, this helps a lot!

I'm still struggling to understand though what the "artificial intelligence" is actually doing in JustAsk. Especially because of the synonyms and rules you mention in the blog: Is the AI also doing this by itself, and are these manual synonyms/rules there to steer/finetune the AI? Or is the AI doing nothing in the area of synonyms at all? I would for example expect JustAsk to understand that the Dutch city "Den Bosch" is similar to "'s Hertogenbosch" in a JustAsk query. Will that be the case or is a synonym required for this?

In the table it shows that the live SAP Datasphere models can't/needn't to be indexed, but they then also won't have the option to define these synonyms and rules? How would such a scenario work then? 

Cheers,
Xavier 

francois_imberton
Product and Topic Expert
Product and Topic Expert

Hi Xavier,
Your question and  just ask underlying technology deserve a specific blog post !
Indeed admin defined synonyms and rules complement the AI models but are not AI.
Just ask is based on  an Named Entity Recognition AI service and a  LLM AI service to generate analytic queries from text. Just Ask also leverages word embeddings to assert similarity between the terms used in the question and the model metadata.
Thanks,
François

xjhacking
Contributor

Hi François, 

About these "word embeddings": What would that mean for my example with a "city name" dimension ("Den Bosch" vs "'s Hertogenbosch"). Would JustAsk be able to figure this out by itself, or would we have to provide a synonym? 

Regards,
Xavier

fran22
Newcomer

Hello!!
Thank you very much for the publication, it is very interesting. Do you think Just ASK could have potential to generate some integration with SAC applications? That is, generate a dashboard for the client in Story and have it recognized in the JUST ASK search engine and brought to view.

What do you tell me about that?

francois_imberton
Product and Topic Expert
Product and Topic Expert

Hi fran22,
Yes, with this initial release the user can open the result in Data Analyzer or copy the result and past is in an optimized story

francois_imberton_1-1706543700071.png

The story team is working on a story assistant that will provide an direct access to just ask within stories. Timeline is not finalized.
Thanks,
François

Richa_T
Discoverer
0 Kudos

Hello,
Thank you for sharing JustAsk details. Could you please share some information about Just Ask vs Search Insights features differences?

Thanks!

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Richard,
Just Ask provide more relevant answers through the following capabilities

  • Just Ask’s LLM enhanced learning with each release
  • Automatically infer similar words based on meaning
  • Custom synonyms for metadata and master data
  • Centrally set searchable models reduces ambiguities when different models include the same names
  • Administrators can capture implicit terms from the questions by defining rules that will customize the query

Just ask supports SAP Datasphere models, where as search-to-insight supports live Hana and live BW models

vsrini
Newcomer
0 Kudos

Some models are not being displayed in managed models when I tried to add models. I am BI Admin in the tenant

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Vsrini,
Could it be that these models are live models ? Only acquired data models can be indexed. 

MKreitlein
Active Contributor
0 Kudos

Hello @francois_imberton 

Nice Blog... do you know a kind of "feature overview" in terms of technical commands?

For example... I tried to sort a bar chart via the just ask feature: "Show Sales by <dimension> sort ascending by Sales".

Unfortunately it does not sort the chart by my key figure, but it seems I have to do this on my own, on the three dots button 😞

Would be nice to understand, what "Just ask" understands 🙂

Thanks,

Martin

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Vsrini,

Correct only acquired models can be indexed for the time being, it will remain that way in Q2 release.
To query SAP Datasphere models user needs to select "Add model to search" from the model selection drop down.

francois_imberton_0-1710343031665.png

 

 

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Martin,
Have you tried "sort sales by region" or "ascending sales by region"

Some of the supported syntax is documented as examples in SAP Analytics Cloud help:
https://help.sap.com/docs/SAP_ANALYTICS_CLOUD/00f68c2e08b941f081002fd3691d86a7/3a1c18ee1f034aa38dbfb...
We are working on improving this to have it more comprehensive.
François

MKreitlein
Active Contributor
0 Kudos

Hello Francois,

thanks for your feedback... I will try more with the help page, you provided.

What I tried is, to sort by Key figure, not by Dimension... this is already done alphabetically:

SORT.png

Interesting thing is, if you ask for Top 5, then the result is sorted automatically - even if I did not ask for it ... wondering if you could tell it to keep the alphabetical order then - I doubt it 😉

BR, Martin

SayedZubair
Active Participant
0 Kudos

Hi Francois, 

Thanks for the good explanation on Just Ask.

Couple of questions:

1. Which one give better results? Indexed Model or Non Indexed Model?

2. Administrator needs to define the Synonyms. In a larger platform where there are multiple such models, don't you think it is not Realistic for the Administrator to maintain the synonyms. Administrator will not have a complete idea on what the Model is aimed at doing. The Developer/User of the model would be the right person to maintain this. What is your thought?

3. It is required to maintain the model everytime the user launches Just Ask?

 

Regards,

Sayed Zubair

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Sayed,
Q1. Indexed models provide better  support for match question to models metadata and better suggestions.
Q2. We don't foresee a scenario were you would massively index all your models and assume end-users will seamlessly find their answer out of these.  Instead we would recommend is cherry pick a target audience, use cases and supporting data, define synonyms based on end-user language.  We'll  go deeper is best practices in a separate blog post.
In the mean time, just ask does have in its backlog the ability to automate the creation/import of synonyms for instance from SAP Datasphere catalog.
Q3. In the initial support of SAP Datasphere, yes, user needs to select the model(s) in each SAP Analytics Cloud session.


 
 

axel_
Participant
0 Kudos

Hi @francois_imberton 

couple of questions:

1. By when are other languages fully supported?
E.g. by when will the titels / description use the search language instead of English? By when will german umlauts ä,ö,ü correctly displayed and not displayed with ?? as currently

2. In rules I can select for my planing model as version value only actual version. I want to direct all searches to the plan version but it is not showing up. Bug?

3. Is somewhere a sample question list - what to ask to get what kind of result

4. Would have expected faster response. Sales_Sample model. Question:"sales per year" takes 5 sec to come back with the chart. Is this the normal performance?

Thanks and best regards
Axel

francois_imberton
Product and Topic Expert
Product and Topic Expert
0 Kudos

Hi Axel,
With regards to your questions.
1. Yes other languages are planned, please check roadmap explorer as product plan are subject to change:
https://roadmaps.sap.com/board?PRODUCT=67838200100800006884&range=CURRENT-LAST#Q2%202024
2. Please check your plan versions is public, if so please log an incident with SAP support (component LOD-ANA-ASK), 
3. Some sample questions are documented: 
https://help.sap.com/docs/SAP_ANALYTICS_CLOUD/00f68c2e08b941f081002fd3691d86a7/3a1c18ee1f034aa38dbfb... I'm adding below a table with other examples of supported questions.
4. For best performance select the model before asking questions, otherwise limit the number of the models that are indexed. 


#

Description

Examples of questions

1

Total value of measure

  • Gross margin 

  • All Taxes

2

Total of one measure for different time periods

  • Let me see the gross margin by month

  • Gross margin by quarter

3

Measure break down by one or more dimensions

  • Gross margin by store and location as radar

  • Net revenue for each product

4

Apply one or more dimension filters.

  • Show me gross margin in France

  • What is the discount for Red Wine

5

Apply a value-based filter on a measure

  • Gross margin by location > 10000

  • Gross margin by location < 10000

  • Gross margin by location greater than 10000

  • Gross margin by location less than 10000

6

Value based filter on Range

(In the case of a dimension with a hierarchy, values are aggregated on terminal leaf nodes)

  • Sales by products between 10m and 20m

  • Sales by products between 10000 and 20000000

7

Rank

  • Top 3 sales managers this month

  • Lowest/highest/bottom sales by product

8

Sort by value or by alphabetical order in a query
  • Sort sales by country

  • Sort ascending sales by country 

  • Sort descending sales by country 

9

Display two planning versions

Profit by product actual compared to forecast  

10

Excluding values
  • Sales by country excluding France in a Tree Map 

  • Top 3 brands excluding Dressbarn

  • Sales by store except Supermarket

11

Chart selection as outputExpense by product as pie chart

12

Measure break down by specific time period

Profit in 2023 and 2024 monthly in Stacked Area chart 

13

Filter a measure for specific time period or range

  • Gross margin until previous month

  • Gross margin from 2022 to 2024

  • Sales by products since January 2023

14

Relative date filters
  • Sales of last year 

  • Sales of previous year 

  • Sales last quarter

15

Time series chart

Trend gross margin and discount by year