Would love to see more natural language interfaces being built for business, cant agree more with your post! It is far more intuitive to ask questions of your data than to deal with select boxes and filters. For many cases, canned dashboards are perfect though. There are some questions so obvious that one doesn't want to ask them - for example, what was the number of visitors to my website yesterday. You would rather someone told you that even before you could ask!
Exactly! I believe natural language interfaces for BI will be one of those "in retrospect it was inevitable" situations. But for that it needs to work really well. Otherwise people try and just go back to how they usually do it. I have still lots to talk about regarding UX and tech challenges, but I guess this will need a separate post. Haha
Yes, agree it needs to work really well. If it answers a few questions and people have to pull out their original tools to go deep, it will never serve the purpose. Also, it needs to be really easy to set up. Understanding tabular data and entities and relationships will be key. Easier said than done, looking forward to what you have in store in future posts regarding tech and UX challenges! Exciting space to be in.
You wrote "Data Democratization is about bridging the gap between our language and the language of data"
Nicely put!
In this context, I believe you may be interested in test diving the Executable English system. It's live online at www.executable-english.com, with many examples that you can view, run and change using a browser. You are cordially invited to write and run your own examples too. If you are reading this, you already know most of the Executable English language !
"Next-gen pivot table" is very close to the vision of SeekTable, and several years ago I also was very inspired by the idea of NLQ and "google-like-search-for-data" (ThoughtSpot invested a lot into this idea). Simple variant of "Q&A" was in SeekTable from the beginning, so I have collected some feedback from users + usage logs of this capability. Now I'm not so optimistic with the assumption that users really want NLQ as a primary interface for in a "self-service" BI, this is a nice-to-have capability but it seems vast majority of users still prefer to use good-old interfaces with dropdowns / checkboxes (textboxes for local searches are also here).
After working on this paradigm for over 5 years, I tend to disagree. NLQ is a hard problem and it starts well before the question is asked. It's easy to get a first version going, we did it in less than 24h during a hackathon. But regarding user experience it's like Baseball: three strikes and you're out. When it does work however, you get even certified Tableau analysts to switch. It's just more convenient, no context switching and every information is just one step away. It's like comparing the internet before search engines with all the directories, drop downs, lists with after search engines. I even use Google as a calculator nowadays. 😂
I respect the power of pivot tables, but most of the users we have today at Veezoo are not capable of using one.
Don't get me wrong, I'm a big fan of search-like interfaces. And I use google as a calculator too :) nevertheless, I don't think that NLQ-first BI will be a silver bullet to solve 'self-service BI'.
To support this point, here is a bit of critique for your 'ideal NLQ BI' that should handle correctly this:
-----
most people usually just say "subscription" instead of "active subscription", so we need to account for that as well.
-----
1) when you google for a single-word you get a definition of 'subscription', because google doesn't know the context.
2) people can have different goal when they search for "subscription". This could be "active subscription", but also this could be a search for the word "subscription" that is the value of the some column.
Ambiguity is a big problem in NLQ processing. Doesn't matter how strong your AI that performs 'search-to-formal-query-and-visualize' translation, recognition context is a key (and usually it is very personal) -- and it is theoretically impossible to have AI that will not have 'misses'. If users have NLQ-first UI, they will have to type a lot of keywords to get what they need -- because desired output may require a lot of details. This doesn't seem as a much, much better UI than existing report designers -- instead of exploring dropdowns/checkboxes/etc users will have to explore what they need to type (= use a keyboard a lot instead of mouse/touch) to get desired output (in form of the table and/or chart).
To conclude, I think search interfaces is a good capability in addition to classic UIs. It is easier (from tech point of view) and more usable (from user's point of view) to have many 'context-dependent' searches across UI than simply one google-like textbox with "I'm Feeling Lucky" button :)
in addition: I've just checked Veezoo website and it looks like you guys implemented the vision that I had originally about 5 years ago -- and the product seems well-polished and useful in production!
SeekTable first customers were more interested in classic reporting and embedding (so it evolves in another direction and NLQ capabilities remain very basic and not polished -- because of no interest from existing users), so maybe my observations on NLQ is simply a result of inappropriate target audience.
Would love to see more natural language interfaces being built for business, cant agree more with your post! It is far more intuitive to ask questions of your data than to deal with select boxes and filters. For many cases, canned dashboards are perfect though. There are some questions so obvious that one doesn't want to ask them - for example, what was the number of visitors to my website yesterday. You would rather someone told you that even before you could ask!
Exactly! I believe natural language interfaces for BI will be one of those "in retrospect it was inevitable" situations. But for that it needs to work really well. Otherwise people try and just go back to how they usually do it. I have still lots to talk about regarding UX and tech challenges, but I guess this will need a separate post. Haha
Yes, agree it needs to work really well. If it answers a few questions and people have to pull out their original tools to go deep, it will never serve the purpose. Also, it needs to be really easy to set up. Understanding tabular data and entities and relationships will be key. Easier said than done, looking forward to what you have in store in future posts regarding tech and UX challenges! Exciting space to be in.
You wrote "Data Democratization is about bridging the gap between our language and the language of data"
Nicely put!
In this context, I believe you may be interested in test diving the Executable English system. It's live online at www.executable-english.com, with many examples that you can view, run and change using a browser. You are cordially invited to write and run your own examples too. If you are reading this, you already know most of the Executable English language !
Thanks for comments.
"Next-gen pivot table" is very close to the vision of SeekTable, and several years ago I also was very inspired by the idea of NLQ and "google-like-search-for-data" (ThoughtSpot invested a lot into this idea). Simple variant of "Q&A" was in SeekTable from the beginning, so I have collected some feedback from users + usage logs of this capability. Now I'm not so optimistic with the assumption that users really want NLQ as a primary interface for in a "self-service" BI, this is a nice-to-have capability but it seems vast majority of users still prefer to use good-old interfaces with dropdowns / checkboxes (textboxes for local searches are also here).
After working on this paradigm for over 5 years, I tend to disagree. NLQ is a hard problem and it starts well before the question is asked. It's easy to get a first version going, we did it in less than 24h during a hackathon. But regarding user experience it's like Baseball: three strikes and you're out. When it does work however, you get even certified Tableau analysts to switch. It's just more convenient, no context switching and every information is just one step away. It's like comparing the internet before search engines with all the directories, drop downs, lists with after search engines. I even use Google as a calculator nowadays. 😂
I respect the power of pivot tables, but most of the users we have today at Veezoo are not capable of using one.
Don't get me wrong, I'm a big fan of search-like interfaces. And I use google as a calculator too :) nevertheless, I don't think that NLQ-first BI will be a silver bullet to solve 'self-service BI'.
To support this point, here is a bit of critique for your 'ideal NLQ BI' that should handle correctly this:
-----
most people usually just say "subscription" instead of "active subscription", so we need to account for that as well.
-----
1) when you google for a single-word you get a definition of 'subscription', because google doesn't know the context.
2) people can have different goal when they search for "subscription". This could be "active subscription", but also this could be a search for the word "subscription" that is the value of the some column.
Ambiguity is a big problem in NLQ processing. Doesn't matter how strong your AI that performs 'search-to-formal-query-and-visualize' translation, recognition context is a key (and usually it is very personal) -- and it is theoretically impossible to have AI that will not have 'misses'. If users have NLQ-first UI, they will have to type a lot of keywords to get what they need -- because desired output may require a lot of details. This doesn't seem as a much, much better UI than existing report designers -- instead of exploring dropdowns/checkboxes/etc users will have to explore what they need to type (= use a keyboard a lot instead of mouse/touch) to get desired output (in form of the table and/or chart).
To conclude, I think search interfaces is a good capability in addition to classic UIs. It is easier (from tech point of view) and more usable (from user's point of view) to have many 'context-dependent' searches across UI than simply one google-like textbox with "I'm Feeling Lucky" button :)
in addition: I've just checked Veezoo website and it looks like you guys implemented the vision that I had originally about 5 years ago -- and the product seems well-polished and useful in production!
SeekTable first customers were more interested in classic reporting and embedding (so it evolves in another direction and NLQ capabilities remain very basic and not polished -- because of no interest from existing users), so maybe my observations on NLQ is simply a result of inappropriate target audience.