Case Study: Drupal gets smarter with AI driven Bot

>> KAMALPREET KAUR:  Good afternoon, everybody.  Thank you for your time to join the session.  I hope you all had a nice lunch and you're keeping yourself safe at home.  And of course, thank you, MidCamp organizers for turning this digital.  I'm just sharing my screen.  Yes, I can.  

So Drupal gets smarter with AI driven bot.  Mostly, where did we use Drupal, and how and why the solutions to work with our customers.  And I'll talk about the technical implementations.  But please feel free to reach me on Twitter and LinkedIn.  I'm active on both.  Shoot me an email if you want to.  I'll be happy to talk to you and get connected with you.  

All right.  So, just to give you an idea as in, you know, what's this all about.  This is for one of our customers.  They wanted to analyze to optimize the performance.  Amazon Lex, Drupal to build a dashboard to serve diverse cases.  So, we believe that going the extra mile always adds the extra value, so we're able to accomplish this    this new solution was actually able to help client, help increase the user retention rate drastically, was able to help upsell of the products, and was very much accessible.  What we did was we basically stepped down during the implementation phase.  We listened to the client's agony.  We listened    that, you know, the solution which was made was actually right, but then that was not accessible by the user.  

So, giving you an overview of the customer.  This is a fortune 500 company and is leading global provider of cleaning solutions and cleaning equipment.  

They don't manufacture any cleaning machines, but they do equip them with IoT sensors.  Used to analyze and monitor the performance data.  These are the kind of floor machines and other machines which they equip IoT with.  

And this data from the IoT, which is sent, is used to make the dashboards like this.  

Up until now, you might have got an idea that this is a fleet tracking solution, you can track and monitor the data online.  You can track and monitor the data on your hand held devices.  You can    you know, by going to the office or by going to any other    maybe at your home in the situation like this, you can ask as in, you know, how is your machine doing.  Just by simply looking at the dashboards, you could look at the portfolio, or you could ask the Chatbot.  

So this was like really helpful when it was turned from dashboards to the chat boards.  

So this is talking about this kind of trends that could be to the dashboard.  How many machines are available in the pool.  How much savings do I have.  So all of these could be asked from the network wherever you want.  

So this is just a snapshot of the dashboard which we made during the project implementation.  Just able to share this.  In this snapshot, the country France    and it clearly mentions that country France has this much savings.  

So when we logged into the dashboard and we could actually see the realtime data that savings incurred with the country of France is this much, and utilized this much machines.  

Similarly, if you could change the drop down boxes, you could see under utilization Dropbox, you would have over utilization machines, underutilized machining and the machine models.  Based on that, you could see the portfolio as in how much you are saving, which machine is not being used, how many machines are actually in the pool.  

So all of this data is realtime data.  These interactive dashboards prove operationally efficient for the customers.  Please let me know if I'm going too slow or too fast.  I'll be able to, you know, [indiscernible].maintain that.  

These dashboards, they were able to fetch the data, they were able to show the realtime data.  But these were not used as much as they would have.  

Now, when people are working at different zones, some people are working on manufacturing zones, some people are working at the field, some people are working in strategy on the offices.  All those people wouldn't have access to this dashboards.  They do have the data in one centralized place, but the dashboards were not accessible to all because you have to figure out the log ins, you have to set parameters, set data engines, et cetera.  

In short, it was too much time and effort, hence we came up with the solution of the chat boards that replace chat boards right in the middle of dashboards.  

This is not the only solution, but then this helps the customer increase the accessibility.  By placing Chatbot in the middle, could simply ask query and get the answers super fast.  The bots worked on ask and answer approach.  It would execute a query by performing an analysis of relevant data, saving huge time and effort.  Executives did not need to set any parameters.  

These automated tasks were done in Excel.  So all those fleet tracking management systems which were on dashboards tried to have an added value, added solution, and able to answer the questions every time and everywhere.  

We use Amazon Lex service to integrate this.  The Lex bot can be integrated with services like Slack, Messenger, and Skype.  We've used native iOS app for integration of this.  The bots can be integrated to provide the personalized experience with the deintegration of the bots.  We wanted to generate a lead as in if there are savings happening, more than 60% in my portfolio.  Might want to have integration of the machines in a particular zone.  

We could also make the bots continuously intelligent by using AI and ML services.  Again, as I was talking, we could have the trend analysis and the predictions done as we move on, saying that, you know, looking at the trends, every year in quarter four, there is a dip in savings, so we might have to increase or decrease our fleets onto that.  

So we could use AI and ML onto the systems to understand the trends, and then, you know, the bots could proactively ask the regional leaders to have the action in place.  

The bots which began, they are defined by information and personalized and transactional.  It could be as simple as which region I belong to, where is my machine.  

Now, as simple as give me information about [indiscernible].  The bots could be personalized, if given the information for your account.  If you say savings in my account, in my portfolio, it would become a personalized option.  Or probably you could ask questions like how are my machines doing.  

So all of those are examples of the personalized bots.  The bots understand the natural language to provide the answers to the questions.  The bots could be transactional and go as fast as I was taking the example of the savings across a particular quarter or increase across a particular quarter, they could be an integration and email which could be send, integration which could happen.  Predictive measures, which could be taking place.  

This is the architecture that we used.  In the user request, the user requests the intent.  What happens?  If are the iOS application sent to the Drupal along with the headers.  And the Drupal recognizes the request, and it places the right roles authorizations to the system in the place.  

The role of the user is used to decide the    beside the access of the user based on global level, country level, region level data.  The Auth header is added in the app response with the authorization and the role of the user.  

Lex recognizes the intent and the Lamda function.  Crunch the number of    to crunch the numbers of the savings in my account, and it returns a request.  The information which is retrieved is set as a response back to the user on the same phone.  This is how the user journey happens when user requests on the Chatbot.  

But how are we updating the data on a regular basis?  All the data which is on the bot is realtime data.  The data from the IoT devices is continuously sent to the S3 buckets, with the help of Lamda.  Whenever the data is updated, the Lamda is required to update the data basis.  

So quickly, to the demo.  So the bot would give me the information.  I'm your bot.  It welcomes you.  It is giving you a personalized patch, and it tells you what could it do.  It could actually tell you the number of overutilized machines.  It could tell you underutilized machines it has.  It gives you an option to select proactivity.  You're free to do    in this particular case, when I select overutilized machines, it tells me that I'm a user whose region is decided, as in France, the number of machines are 21.  The savings done in the region France is this much.  Number of sites, which are operating in France are 21.  

Similarly, to give me the top five sites, in France at this match, I can actually go into the site and check which machines are doing what.  In this particular site, it will tell you, these are the number of machines and this is much is the savings.  You can actually drill it down to a machine level.  This particular machine has just one machine in store.  Actual informations about the machine, all the parameters of the machine.  

As I was mentioning, right now, I'm asking the bot what are the overutilized machines in the country Turkey?  However, I'm a user who is assigned only the France data.  So the bot shouldn't answer this.  It would say that the data is not permissible.  But it will give me the rate of France because I'm a French user here.  And again, the flow would continue.  And as you saw here, then you would have here on the top right corner.  I also have an option to change the language of the bot.  

Right now, there's just five languages.  The top five languages, which were used during the customer    which are used most during the    by the customers.  

Once we change the language, we can interact with this particular system.  The natural language behind the scenes, which is Lex in this particular case would identify the language detected and change    get you the answer back into that particular language, changing the intent.  

These are the kind of    this is just one kind of user journey.  All other user journeys are done in a particular way where    all of the user journeys are done in a similar way, where we actively decline, understood the complete workflow and have this workflow proactively driven inside the Chatbot and inside the dashboards where the data was readily available to them.  

The bots in this particular ecosystem had declined, upsell the business worth of $90 million.  Help the user retention.  The bot applications allowed the analytics team to automate a lot of tasks which were happening on spreadsheets.  

So the details of all of these is available.  There's an article in the blog.  I'll show it with you on the slides.  And I think I'm open for the questions.  

>> ANDREW OLSON:  Great.  Yeah, if anybody has any questions, please put them in the chat, and I can read them off.  

While we're waiting, I also put a link to the session.  Feel free to give feedback on the session, and also put your questions in the chat.  

I put the link in the chat.  We'll give it another minute or two.  See if there's any questions.  

Do you have any other comments you want to say?  Otherwise, I can stop the recording and we can take questions into the hallway.  And contact information the on the screen.  Feel free to reach out direct.  

Do you have any last comments?  

>> KAMALPREET KAUR:  Thank you, guys.  Any feedback, if any, would be really appreciated.