What Makes MoEngage Reliable at Scale?
MoEngage evolves into an elastic software by balancing resource utilization dynamically and ensuring no failures.
Reading Time: 6 minutes
MoEngage processes over 250 billion events every month, reaching speeds of up to 500 million events per hour at the rate of 250K per second.
If you were to travel at this speed (250K miles per second), you would reach the moon before you could say one Mississippi!
Brands worldwide use MoEngage to send more than 3 billion messages in a day, including 2.5 billion push notifications, over 250 million emails, and 20 million WhatsApp messages.
Here are some graphics to help you visualize the scale at which MoEngage operates:
Why is reliability important?
Let’s say you decide to run a flash sale in your retail outlet in Queens, New York. After all, sales are great for building customer engagement, generating a burst of revenue, and capturing shopping patterns, aren’t they?
You want to make as much noise as possible about this sale. You want to send out emails, push notifications, and text messages. You also want to show ads on TikTok, Instagram, or Facebook and show banners of this sale to all visitors coming to your website or mobile app from New York.
You plan to send out comms about this sale only to specific shoppers who have visited your store in Queens in the past year. You also decide to promote this sale to customers who haven’t shopped online or in your stores for over six months but made their last purchase from New York – frequent buyers don’t need a sale to drive a purchase, do they?
So, while you run ads to bring visitors from New York to your website, only the audience with infrequent interactions with your brand will see details about the sale.
Setting this up needs a lot of data and a reliable customer engagement platform (CEP). Your CEP needs a fail-proof data architecture to fetch this data from multiple sources, convert it into a format it understands, interpret it, and help you build customer cohorts. The platform also needs to be able to filter the audience in real time based on either location or past behavior.
A single failure can lead to a domino effect on your flash sale, directly impacting your revenue and deviating from your projections. Some common failures look like this:
- Unable to send messages, emails, or push notifications due to the size of the audience
- Missing audience interactions such as email opens or clicks on social media ads
- Poor or incorrect campaign optimization due to missed engagement
- Incorrect or incomplete data ingested from your websites or multiple individual sources
- Wrong insights, analytics, or reporting
Why do these failures occur?
While there can be several reasons for the failure of large-scale data and campaign operations, one of the most common is the lack of your customer engagement platform’s capabilities to deal with unpredictable patterns.
You’re not the only brand that may decide to run a flash sale. Maybe a large bank needs to process millions of data points at the start of the new fiscal year. Perhaps that video streaming platform has a spike in traffic because of the FIFA World Cup tournament.
But you shouldn’t suffer because your customer engagement platform is unreliable. Your revenue should not be impacted because your customer engagement platform cannot deal with unpredictable spikes in traffic or data operations.
How did MoEngage solve this challenge?
Our fundamental objective and guiding principle has been to achieve true elasticity across all of our systems.
Building an infrastructure that navigates the failures mentioned above requires 3 critical steps:
- Data ingestion needs to happen in real-time at a scale of billions
- The ingested data must be processed and prepared for use in campaigns
- Comms over billions of emails, push notifications, text messages, in-app messages, website banners, and more need to be successfully sent
While the team had successfully solved the BIG challenge of ingesting data in real-time, the next step was to ensure that our customers could efficiently utilize this ingested data to set up real-time campaigns.
This meant ensuring every component in our infrastructure (processing, memory, storage, and others) could adapt dynamically to changing demand–achieving true elasticity!
But attaining elasticity is easier said than done, especially after our ingestion speeds increased by 10x!
The engineering team had to guarantee that MoEngage could automatically scale resources up or down in real-time based on workload demands while ensuring consistent performance and SLA guarantees. The team also figured they needed to implement an automated monitoring and trigger-based scaling system that adjusted resources without human intervention.
This mammoth task did not deter our superstars.
Our engineers rolled up their sleeves and came up with a unique solution to efficiently process customer interactions (like clicks, searches, or purchases) from multiple sources while handling unpredictable traffic patterns and maintaining consistent processing times.
Btw, your tech and engineering buddies will love this, so make sure you show them what we built!
The solution built by our engineering team works through a sequence of components:
- Events are first captured via APIs and stored in a ‘message store.’
- A monitoring component continuously polls this store to analyze traffic patterns and predict resource needs.
- When processing capacity needs to be increased (for example, when ingesting many data points or sending out millions of comms at once, like in your flash sale), a ‘clone orchestrator’ dynamically creates additional processing instances to handle the workload.
Neat, isn’t it?
The system’s ability to scale resources up and down based on real-time demand and historical traffic patterns makes it innovative and unique. Our engineering team used asynchronous, non-blocking methods to write data efficiently and then distribute events across resources to prevent a single application from monopolizing system resources.
How does MoEngage’s elastic infrastructure help you?
Our innovative solution predicts traffic patterns and dynamically scales our resources to maintain service level agreements (SLAs) even during unexpected traffic spikes.
So, a video streaming platform needing extra resources from MoEngage to help it drive more viewers to the FIFA World Cup tournament will not impact your plan of launching a flash sale!
No incomplete data ingestion attempts.
No missing customer interactions.
No incorrect analytics.
No failed comms.
Oh, and I forgot to mention – ALL of this happens while keeping your data safe and secure, thanks to our patented technology that tokenizes PII information!
If you’re interested in reading more about how this system works, check out this detailed explanation below:
An advanced explanation of how this system worksOur unique patented system implements a dynamic resource allocation strategy based on the predictive scaling of data ingestor instances. The architecture consists of several key components:
From an implementation perspective, the system employs several notable techniques:
Our system addresses the classical distributed systems challenge of maintaining SLA compliance under variable load conditions while optimizing resource utilization. It implements a feedback loop between observed traffic patterns, predicted demand, and resource allocation decisions. The architecture allows for tenant-specific QoS guarantees through prioritized resource allocation while maintaining system-wide fairness in resource distribution. We use a combination of reactive and proactive scaling strategies, with the latter being particularly valuable for accommodating predictable traffic patterns without incurring the latency penalties associated with purely reactive approaches. |
Key takeaways
As a brand that caters to millions of customers, you must ensure smooth, efficient, and successful data operations and campaign management.
A reliable customer engagement platform (CEP) is crucial to ensuring data is ingested from all available sources and is prepped and ready for use in building campaigns. It also ensures comms are delivered to the right audience without snafus. Most importantly, your CEP must not ‘break’ when multiple brands scale up their operations and needs at the same time.
A single failure in any step leads to missed opportunities, lost revenue, and a negative impact on your brand reputation.
MoEngage understands the need for a reliable CEP, and we’ve addressed the challenges via a 3-step framework:
- Real-time data ingestion at a scale of billions (read more about it here)
- Processing and preparation of trillions of data points for use in campaigns
- Successful delivery of millions of customer comms over multiple touchpoints
While this article delves deep into how we address the second step with a unique and innovative approach, watch this space for how we successfully deliver comms on multiple channels in our next article!
