Atlassian Corporation (TEAM) Earnings Call Transcript & Summary

So I've got the great pleasure of introducing our next talk here because I'm sure you probably all know, but at Camunda we really love hearing stories about what developers are building with Camunda, and so that's why I'm so pleased to welcome Brajesh Bharti from Atlassian. Brajesh is going to describe the many opportunities that Atlassian developers gained by deploying Camunda, and using the workflow and BPMN -- workflow orchestration and BPMN capabilities that came with it. So together, we'll hear how Camunda served as a centralized orchestration platform for all of Atlassian's support tickets in the form of a very cool automation bot that they have named Suzie. And so with that, Brajesh, please do take it away.

Thank you so much. Welcome to the CamundaCon. I'm Brajesh, Senior Engineer in Intelligence Automation team of Atlassian. Earlier, I was part of the Camunda on-boarding team and has helped operationalize Camunda in Atlassian in Intelligent Automation team. As you have heard in Camunda keynote sessions [Indiscernible] yesterday, as he mentioned, Atlassian has drastically reduced the wait time for important processes by leveraging hyper-automation capability in Camunda -- in automation in bot as mentioned we're using the Suzie bot. We are using Camunda at the heart of our ticket supporting bot which is known as Suzie, enabling us to connect with different endpoints, including various RPA systems. In this session, we're going to explore more on our journey how we evolved Suzie, and how we leverage the hyper-automation capability of Camunda in Suzie bot. As I'm going to talk about our beloved bot Suzie, which is helping customer and partner to directly work with sales and field operations. In Suzie bot which is helping sales and finance related things, like quoting or subscription processes end-to-end, where Suzie have lots of capability, but I'm going to highlight some key features from Suzie which is one of our bill bot, which is helping us in many of our capabilities but going to highlight in here 4 pointer. First is the configuring policy for the business user where the business user can configure policy at the right time, inquire DMN or the rule configuration is very easy, and that we can easily configure well and that how it helps the Suzie helping to configuring the policy for business users. Second thing, the quoting processes where Atlassian products have different product and different type of subscription, where we required the quoting. So partner and customer can request for the quoting request and that helping end-to-end processes via Suzie bot, which helps to quoting processes for the Atlassian product. Third is activation and deactivation for the subscription. As the customer required a subscription for one of the product or multiple product, which require end-to-end checking the subscription retail, creating order, from order to go to subscription, or a complete subscription to go to the quoting, and then payment invoice end-to-end. So that's the activation and deactivation Suzie helping out us to do the activation and deactivation subscriptions for customers and partners. And last not least, the auditing and license report. So if any auditing happening in any organization, they can request us to get the license report for the -- at demand. Suzie provides the license report to them to handling into the auditing processes. Suzie have lots of features, lots of things we are doing in Suzie, but these are the key features we are highlighting in this session. Next is, I'm going to -- so Suzie, earlier we implemented in RPA. It's fully RPA-based implementation automation. Now when we introduced Camunda, we leveraged the best part of RPA and Camunda as well. And then we integrated how -- improve our -- the wait time or quoting processes improvement, the end-to-end processes, reduce the wait time for end customer basically. I'm going to talk about what all the challenges we found in Suzie bot, and then we move -- and how we integrated RPA with Camunda. So these are the key challenges we found in RPA. First is out of the box, in orchestration layer we don't have in RPA and that will be basically for the end customer or business user not able to visualize to what the workflow end to end, and that will be one of the challenges we found and -- which is RPA not provide out of the box orchestration for this. Second is pull-based schedule RPA had, which is basically -- in this, we are not able to do real-time processing because you have to configure the time base scheduling in RPA. And that will -- the challenge is, we are not able to handle the real-time processing for any of the processes in the Suzie bot. Third is, it's not suitable for the long-running processes like one of the use cases like quoting for approval is needed, and you require some approval. That approval need to be go into the different department, different users, and it will take time to approve for your processes, approve for your quoting things or invoicing or payment, different type of business use cases, and that will be not supporting by the RPA. Fourth one is substantial amount of maintenance as in RPA, if orchestrated layer or that you have to do the maintenance with your -- if any changes comes, do the weekly maintenance, we have to do into the RPA system. Fifth one is dependence on developers dimension. In RPA what we face in the challenges like if any workflow need to be changed, we have to depend on the developer. As I mentioned in the policy configuration, we can -- on the fly, we can able to do into the DMN layer, and we are not dependent on the developer because we can change directly from the DMN via business user or immediately we can change and then on the fly we can deploy. But in RPA, we have to be dependent on the developer and the developer can able to make the changes for this configuration and then we have to deploy. So these are the challenges we faced and then we thought we can combine the best uses of RPA and best uses of Camunda, and then we can integrate with both of the best world from the RPA and Camunda and use for the Suzie bot. Now going to discuss about like how Camunda help to avoid these challenges and improve the Suzie bot. So Camunda as providing the orchestration layer for the workflow, and it's very easy to track end-to-end workflow from the business user and visualize -- anyone can visualize end-to-end workflow orchestration, what all the event is triggering happening, what all the system impacted upstream, downstream and it's very easy to -- workflow can be defined without decision layer. As I mentioned, in RPA, we are facing -- we have to configure into the schedule based configuration. Now in Camunda, we can put the real-time -- kind of real-time response because Camunda provide the API based or the subscription based or you can configure the real-time processing end-to-end processes. I'll come to the point where we implement it to the real-time processing and how it change impact happen in the upcoming slide. Third is the flexible architecture enabling integration with other system. As I mentioned, we are using low-code/no code, some of the platform like MuleSoft, Workato and different -- Oracle, NetSuite, right? So different type of system we are using and that we are very easy to integrate with this architecture to change this or enabling this integration with this system, using the Camunda. Developer-friendly as I mentioned, the DMN can easily change on the fly, where if any required in BPMN immediately you can change, and that will be -- you can integrate with your CI/CD and that will change the impact happen to your -- inside the CI/CD pipeline. So it's really -- and the main thing for the developer, you can use any tech stack like Java, React, NodeJS or Python. So whichever your stable tech stack that you can use into -- with the Camunda agent. Visualization of business processes, as Camunda provides Optimize, we are using Optimize for the visualization for end-to-end processes as we are using the Suzie bot for ticketing support system so we can easily visualize to how many tickets we have processed in month, for the year and quarter and where all the business exception, what all the technical exception where we can miss the data. So these things we can easily visualize end-to-end process and very easy to track for the management purpose or developer purpose, or business purpose end-to-end. This is the Suzie workflow, where it's a start from the subscription base. So we implemented AWS SQS subscription base, where Suzie consumed all the ticket. So if any ticket created in Atlassian support system, where we get the event trigger and that event we are listing here, subscribe. And then it go -- we configure the DMN on the basis of the business criteria, which event need to be called that we can configure into the SQS based level, and as well as the DMN layer. So in DMN, we configure this event and then it call to the -- based on the business tool, valid for the event and that ticket to go to that event. So you can see here, it's lots of features in Suzie. It's each event calling to the one end-to-end processes like quoting is one end-to-end processes, subscription is one end-to-end processes. So you can see here lots of features and we are using end-to-end processes here. I'll go to details in 2 of the use cases, how we are using and how it's going to the event trigger happening to this workflow. The first case study I'm going to talk about the quoting processes. As what all the challenges we faced when we implemented in RPA, what -- how Camunda provided the solution and what the outcomes impact happened using the Camunda and RPA. So first challenge is, we are not able to provide the quick response to the customer because in RPA, we configure the schedule base trigger, and that will be picked every 45 or 1 hour, have to wait for, if any, even trigger or any ticket created that has to wait for some certain time. And after that time, it will be -- pick that ticket and then process for that customer. So the first challenge is, the wait time is longer in using the RPA for the quoting processes. What all the solutions, so we implement the Camunda, whenever the event trigger, as I mentioned in the earlier slide, whenever the ticket created our Suzie bot subscribes that event and then process immediately. So we are able to handle the event-based trigger rather than schedule-based trigger. And outcome is Suzie can handle now it's 80% of ticket and that the turnaround time, it's reducing almost 93%, where 45 minute to 2 minute now we are -- within a 2 minute we are able to provide the quoting processes end-to-end. As the second challenge is, not able to close the business deal immediately, and that will be impacting us to business processes and delay for the customer and partner response. So we integrated RPA with the Camunda orchestrator where we are calling different type of RPA system, wherever the UI automation where we use the RPA and API-based or the call based or event based where we use the Camunda. So we use the Camunda orchestrated for all the call and then we implemented in the RPA with Camunda. And that will be, again, impact to the turnaround time is reduced, and it's very faster to end-to-end processes, and we are easily able to track end-to-end processes from the event trigger and the response from RPA or other systems. Lack of tooling for the holistic view. As I mentioned, more business insight, what happened, where it delayed or what will be the impact. That will be difficult into the RPA to track that part. So in Camunda, as we can put the timer or we can put the tool processes, and that will be transitioned start from there that point. So we are able to implement the long-running processes wherever the approval is required or waiting from the response from the either Oracle side or the NetSuite or RPA side, we can wait there and then we can process from that -- the transactions and we are able to implement end-to-end. So Camunda provide the long-running processes, and that will be implemented. And then through the Optimize, it's very easy to track your report where and what -- why exception happened and that will be easy to analyze your impact or basically end-to-end. And that is -- outcome is basically winning through the trust for the potential customer. As customer get the quick response and then it's easy to decide, go with this quoting or not, that will be very easier for them. Because waiting for the 45 minute to get the response within a 2 minute for the quoting for all the product that will be easy to deciding that, and that will impact ultimately the business quoting processes. Next slide, I'm going to talk about human-in-the-loop, how it will help. But before that, I'll go to the how quoting processes integrated with the Camunda with RPA. So as you can see here, the first slide is the quoting processes in Camunda where we get the ticket. We add the event, call the event for which either is a direct customer or partner customer and then send back to the customer response. Here again, we are using the Camunda leverage the DMN, where in DMN you can configure which type of customer and on the basis of customer you can send back to the response. And then, well, for calling to the RPA, we are adding into the queue in RPA queue, and then we are processing end-to-end the RPA things. The second case study is human-in-the-loop, again I'll go to the what challenges and what ticket we are not able to handle with RPA and then how Camunda helping us to soliciting with NLP and ML part and then what the outcome impacted us during these processes. So the second case is human-in-the-loop. The challenge is the input data that bot could not understand, like, let's see one of the example you put the address -- customer address, right? But that address is not valid and that bot can't be able to understand your pin code or your address or city or country, some misspell that. So that part is the data not able to understand, where how Camunda helping us here. We build using the NLP and ML, we are able to handle unstructured data. We are able to read that data. And then accordingly, we integrate the other API through the Camunda and then we're able to read and processes wherever the data is could not understand from the bot, which impacted is where we are able to 80% of ticket earlier, now we improved our -- the accuracy for the handling the ticket where they [Indiscernible] as a resolution, so we are able to handle end-to-end ticket more rather than without human-in-the-loop. Second thing, the processes where human intervention is necessary such as manual review, like some quoting or some processes need to be reviewed before going to the customer, like the authorized and like let's say you require the license audit report, and that authorization later is not valid from the customer or partner that need to be reviewed with your -- some of the department or some of the member need to review for that part where manual intervention is required that we are not able to handle in the RPA. But using the Camunda, what we did, we put the waiting time there, timer we put, they review, approve, come to that same transaction and then again, end-to-end process happened with able to complete all the processes, where the user and group able to approve the quote, refund the invoicing, trade note, lots of use cases where you can able to get the responses. Third is the business exception, which need like -- I'll just give one example where the customer put some data, and that data is not valid or it's required some business -- as per business scenario, it's not valid, the data or the scenario and that need to be processed. But as bot -- Suzie bot, not able to understand the data and it's so the business exception because as for business rule, it's not valid, where what human can do even check with the customer, get the correct data, fill them back to the upstream or downstream system and then able to handle that part using the little human-in-the-loop in between the [Indiscernible] and basically able to -- so impact is, outcome is, it's simplified the exception handling, and we are able to handle the exception scenario as well. Obviously, little human intervention required in between in the flow. How it's flow work? You can see here the customer partner request to the Atlassian portal, create a data ticket, the subscription base Suzie bot where the Camunda listen if any event trigger or any ticket created, it processes. But if required any manual intervention, it identifies the ticket, do the action, which is required and what needs to be done and then finally send back the response back to the customer. Now I'm going to talk about how technical architecture in Atlassian, and how we are using the Camunda in Atlassian. So -- you can see here the technical architecture one part is RPA, and another is Observability where we are using the Splunk, LaunchDarkly, Opsgenie, Slack, multiple -- you can see for the Observability purpose. RPA is -- we have lots of bot where we have lots of queue for event-based processes and then each event added it here. The Camunda, we are using the AWS platform, where we deploy on to top of AWS EC2 instances. You can see here the Camunda server and the authentication mechanism, we are validating through the Staff ID, which is Atlassian-based SM where we can create a group that Camunda group, the developer group or BSA group, or that admin group. So there are different type of group you can create. You can add that the Staff ID there. And then as Camunda provide validating through SSAM, you can validate your user, which group belongs to and then accordingly that user gets the access for that Camunda purpose, Camunda cockpit and admin UI. So you can see here the Camunda server where we can configure that SSAM. That SSAM configure the microscope that is a UI-based where we configure the group and using the group you can add multiple user and that user is your Staff ID and that Staff ID can be validated with that group and then accessible with the Camunda. You can see here the server center. You can see the alarm, you can configure if anything, a record or if any goes wrong, your EC2 instance is up or the data is loading or anything happen, it will alarm your Camunda layer. And it's basically configured in the Opsgenie. Coming to the Observability part, we are using the Splunk for logging, we are using the Opsgenie for alerting mechanism. And alerting mechanism, we are using Opsgenie and as well as Slack. We are using the LaunchDarkly for the feature flag. So without the feature flag, what -- so we had -- we have to wait for deployment for the upstream and downstream system need to be ready. But during the feature flag, you can deploy your things on prod but you can turn off that feature and then whenever the upstream or downstream system ready, you can just on the feature flag and then you can go ahead with the deployment processes. So you don't need to wait for the -- as a middleware or integration layer we have always wait for the upstream and downstream, so here, you don't need to wait for the system and you can deploy on prod and enable your feature on the basis of demand whenever required. Next is going to talk about the journey, Camunda with Suzie journey in Atlassian. So we started in November 2019, and we -- the platform team identified the BPMN, which is Camunda. We evaluate lots of BPMN and then we found the best fit for the micro services or the tech start, which is best suited for the Camunda. We operationalized it in April 2020, where somewhere around 2020, we evolved the Suzie as well, where we started with the subscription and activation and deactivation subscription in the Suzie bot. And then it go with adding the multiple capability like quoting, auditing, human-in-the-loop and that export control policy configuration, so lots of features and capability we added in the last 2 years. And now it's almost 20 automated processes in Atlassian, 20 means here I'm giving the invoice process as one process, credit, refund, taxation and lots of processes we automated. And that one automated process is actually, internally, it's calling multiple, event multiple processes, and that will be handle the capability. Like Suzie is one automated and that capability -- lots of capability inside in Suzie. Next stop is what the satisfaction from the customer as the Suzie, the faster response, one of the customer mentioned here and that support -- whenever you create a ticket or support system, if you get the response immediately that will be happiest moment for the end customer, basically resolve the issue. As I mentioned, the subscription, activation and deactivation -- like your site is deactivate and you create a ticket, but it's due to some technical or some -- the error and your site is deactivated, bot identified that deactivation, what happened and then it's resolved within a minute and then provide the response back to you. So faster response and you get the response back to this. So here, the closing note here, what we can take from this session. If you want to implement your RPA system with the Camunda you can leverage the Camunda as a heart of orchestration -- heart of automation, where use orchestration layer, you call the multiple RPA system, call to lots of the end system where you can integrate with API-based or event-based or subscription-based, you can -- like we -- one of the use cases, we use the Salesforce and we integrate with the Salesforce where we are getting all the sales automation like approval, quoting and all these things, where we get the approval from Salesforce, and from Salesforce we directly call to the Camunda system. So technical point here is you can integrate with RPA system with Camunda. And then if you want to -- that schedule base to event base, you can use the Camunda, leveraging Camunda. If you want to do your organization, the quoting, invoicing, the finance system overall end-to-end or end-to-end sales and automation system, best part is use the Camunda as orchestration layer and then leverage -- get all benefit from the Camunda which is provided. So that's all from the session. Yes. Thank you so much. These are the things where you can feel free to ping or if any clarification, you can mail me or chat with me.

That's wonderful. Everyone, can we give Brajesh a round of applause, please? Thank you. Such a great talk. I really love that. And so again, just a friendly reminder, folks in the studio here, and of course, online or at home, please use Slido. CC Stage 2 is for our track here, but you can upvote questions that you definitely want Brajesh's input on. And of course, you can add new ones as well. So please use that because it does help us a lot.

But I don't know. I thought it was fascinating that not only did you just -- you had this great tech stack that is playing to the strengths of each of the pieces that you have there. And then again, of course, orchestrating it all together with Camunda, so much so that people think that Suzie is real. Like that's just remarkable to me. That uncanny valley situation that's happening, right?

Correct, correct. Even customers sometimes know like Suzie is real.

Yes. That's great. Okay. Well, let's take a look at some of these questions here on Slido. We've got a decent amount. So again, folks, please do use the upvoting function there. But which RPA solution have you used in your environment? That's one of the first questions here.

We are using the UI products in RPA soliciting, and that is integrated with the Camunda.

Great. What is the scale in terms of processes per day that you're handling?

Scale wise, currently, we are handling the 8,000 in the Suzie bot, the support system, 8,000 monthly. But like lots of automation we have, like invoicing and credit note where we -- millions of -- in monthly, we are getting the data and then we scale that. So we configure the EC2 instances for scale. And that will be that engine layer, each layer, it will get the instances -- same instances, and that will be handled to the scaled part.

Great. How many people use Optimize? And where does the responsibility kind of lie in the company or on your team?

Yes, yes. So in team, everybody is using Optimize in all the automation configure with Optimize. Some of the business users using, but only thing, one of the challenges we are facing with the access part in the Optimize. But for the reporting purpose, for all the automation, using the BSA. So we have a couple of groups like BSA group. One is the admin group and then the developer for the viewing and the end-to-end processes. And then we have -- we have created a separate report for the management, which is handle the quarterly what all the responses and what all the ticket handled for Suzie. And even that invoicing -- if any invoicing that impact, if anything wrong or anything in data mismatch that will we get the report from the Optimize side.

All right. Great. Okay. Can you elaborate more on how you're using machine learning and natural language processing tech?

Yes. So in quoting processes, we get a couple of tickets where we are not able to handle due to that -- the customer retain, as it's a free text fill in the Jira ticket. And when they wanted like the Jira product and Jira Cloud where 500 users, big bucket is 200 user, but it's not in sync or it's not in the proper tags. Where we use the NLP, where we build the NLP API, NLP services, where we send this data and then NLP extract, get the human sense, I mean, the machine sense where get back to the response in the respective format, these are the product, and the users wanted for this product. And that's how we use NLP and ML to handle this scenario and provide the quoting -- quote to the end customer and partner.

Great. Have you been able to quantify the benefits like increased sales, operational savings or reduced time to revenue?

Yes. Definitely -- we reduced -- in Camunda, we are using different tech stack, but almost one -- yes, I mean, 1 lakh are saved in last year in the automation, where Camunda, helping us do the manual efforts, saving almost 50,000 around in last year -- manual effort, and as well as impacted as the business purpose as the quoting -- the quoting processes improve, customer getting immediately invoicing, quoting, that auditing report sent back. So it will be impact to the customer satisfaction and then impact to the business as well, as we are able to close immediately in business purpose, the quoting things, and that will be, again, adding to the dollar value into that, our bucket.

Yes. I mean even the software side of it, you're just increasing that trust between the customers because it is just so rapid and timely and everything like that.

Correct, yes.

That's great. Okay. Do you have a standard way of organizing workflows at Atlassian?

Yes. We have a translational workflow, where different type of event if you want to use the queue-based -- subscription-based. So we build one of the -- our common library, where you can configure that event based, the queue name, event name and then it automatically configure your -- the calling to the events. So that workflow, you have to view that queue-based workflow and then it will be use the subscription-based event. And that's how it's go end-to-end workflow. But if you want to API-based or that event based that will call to the real-time API in the Camunda that have a different workflow and need to implement that queue base and you can go to the API layer, API implements that workflow, you can implement where you can define the DMN, they have also rule for the -- where we use the pool base or not use the pool, where we can use the looping and all these things. So yes, we have -- that they have a rule where and what need to be the workflow, how we can use the workflow in which scenario. So it depends on to the business scenario based.

Perfect. That makes sense. A lot of questions, but wanting to see the bot in action here. But before we get to that one because I think maybe you just need to get a quote from Atlassian and that's how they can see the bot in action, right? I think this is a great question, is -- what's your approach to process optimization? Are you benchmarking performance or customer satisfaction? Or are you just collecting these e-mails where people think Suzie is a real person.