The founding team of Apache Kafka worked at LInkedIn before leaving to start Confluent. Apache Kafka is used by thousands of companies for message streaming, such as LinkedIn, where a publish/subscribe model allows applications to share and create data in a serverless and microservices architecture. What Kafka solved for is the ingestion of events data in real-time with low latency.
At the time that Kafka was developed, LinkedIn was ingesting 1 billion events a day. The company is now ingesting 1 trillion per day. Kafka does this through a log that writes messages to a topic and is able to retain messages for a long time. Kafka is also used in stream processing by parallelizing the pipelines. Kafka Streams were built to increase simplicity while retaining the same amount of performance as a Spark streaming job.
As with Spark and other open-source projects, there is a marketplace for making the frameworks easier to use. Confluent Kafka opens up the amount of data that can be integrated, for example, to combine transactional data (orders, inventory) with sentiment-driven data (likes, page clicks). This helps with predictive analytics and also machine learning because the “data flow” allows for algorithms to work as they are intended to. This is what is meant by the title slide of the S-1 filing “Set Data in Motion.” In order for data to be in motion, Confluent’s platform connects data from many different sources.
Note: We first covered Confluent here and there is more additional information in that write-up on Spark, etc.covered Confluent here and there is more additional information in that write-up on Spark, etc.
Confluent created a Community License to help stave off companies like Amazon from commercializing its contributions. The Confluent license is a re-architected cloud-scale Kafka framework that is compatible with and improves on existing Kafka systems. Kafka sits in the middle of data analytics/warehousing and databases, which is technically above the operation layer and below the application layer. Rather than focus on data in rest, Confluent is reimagining what data in motion and in real-time looks like by introducing tools, connectors and a stream processing layer for Kafka workloads.
This is a good quote from the S-1 filing: “It is not just that companies are using more software–in a very real sense, they are actually becoming software.” By becoming more software driven, more businesses will rely on real-time data. Confluent believes that data in rest is not able to meet the current and future demands of software-driven businesses. Daily batch processing and static real-time queries or “point-in-time” queries with stored data lead to an unnecessarily large and tangled architecture that is not capable of data flow between applications.
Here's a quote from the most recent earnings call:
“I would say there's really no reason you want things to be processed in batch. Like it doesn't — there's nothing in the world that happens batch. Things in the world happen all the time, continuously and in real time” … “I touched on this a little bit in the call, is it's kind of just the digitization of business, like as the actual operation of the business moves into software, not just the analysis, not just the report you get in the morning, but the actual carrying out of the business as it plugs into e-commerce things, as it drives operations out in the world, as IoT bridges into other parts of the world, as machine learning kind of closes the loop on some of the decision-making and processes, that's really where you have to do it.”there's really no reason you want things to be processed in batch. Like it doesn't — there's nothing in the world that happens batch. Things in the world happen all the time, continuously and in real time” … “I touched on this a little bit in the call, is it's kind of just the digitization of business, like as the actual operation of the business moves into software, not just the analysis, not just the report you get in the morning, but the actual carrying out of the business as it plugs into e-commerce things, as it drives operations out in the world, as IoT bridges into other parts of the world, as machine learning kind of closes the loop on some of the decision-making and processes, that's really where you have to do it.”
Big data is more relevant for companies like LinkedIn where Kafka was developed. Big data is not the thesis rather the thesis is the increase in the number of companies that will need real-time data processing and real-time data analytics due to the increase in software driven architectures. The idea is that “data in motion” will replace data at rest, or batch data processing from traditional databases. This is also important for the real-time data streams that machine learning requires.
Confluent/Kafka is Built for the Constant Flow of Data (i.e., Event Streaming)
In our write-up on Big Data, Analytics (and ML) we discussed event streaming and the importance of Apache Kafka. Data streams are created from real-time events, such as messages, transactions or traffic conditions. Confluent opens up the amount of data that can be integrated, to combine transactional data (orders, inventory) with sentiment-driven data (likes, page clicks). This allows large amounts of data to be moved quickly especially for machine learning algorithms that are very data hungry for training models. Specifically, building an analytic model that is trained and makes predictions requires current information, will make predictions on new events in real time, and requires monitoring the infrastructure for accuracy and errors. With Confluent, this can be done by building streaming analytics on top of Kafka using compatible languages, such as Java, .NET or Python, or data scientists can use the SQL engine.
The goal is to reduce operational complexity, deliver optimal user experience and increase accuracy. Kafka Streams through Confluent do not require a cluster to be spun up, offers a single framework for streams of events, and reduces the number of pieces in a stream architecture. Kafka is already a popular framework that can handle trillions of events in one day, has over 5 million downloads from developers, and is used by Big Tech, stock exchanges and car manufacturers.
Source: Benchmarks from Confluent
Kafka is used by 70% of enterprises and this is helpful for Confluent, as the company offers a managed Kafka system. The company believes Kafka is the correct framework to lead a new data flow due to its high throughput of 600 megabytes per second and 5 milliseconds of latency. The maturity of the Kafka ecosystem is also large in terms of number of developers and partners. This important as it prevents Pulsar from taking market share from Kafka.
The three main improvements that the Confluent team made are:
1. to re-engineer Kafka to be cloud-native; especially easy management of elastic quotas (i.e., reads and writes on usage) and offering security with multitenancy.
2. offer a more complete Kafka ecosystem with over 120 connectors and a control plane that automates limit handling; also a SQL layer to bridge current database skills with streaming.
3. is more geographically inclined through cluster linking and other improvements for geographically distributed data; cluster linking allows Kafka clusters in different geographies to be connected for more real-time data flow.
Amazon MSK is a competing managed streaming service that is a good option for developers provisioning a Kafka cluster and a new streaming platform may not be needed in this case. Rather than re-architect Kafka to be cloud-native, Amazon MSK cloud-enabled it as provisioned infrastructure. This means Confluent is stronger than MSK with scaling elastically by offering elastic quotas, which eliminates the need to size clusters for spikes. It’s also stronger on multi-tenancy security. Amazon MSK also does not offer Kafka Connect or Kafka Streams. For more enterprise uses where Kafka Connect or Kafka Streams is required, then Confluent is more likely to be used to save development time and learning curve in writing Kafka Connects sinks and source.
The majority of the information above focuses on Confluent creating a new kind of data pipeline for streaming, yet Confluent will have another catalyst down the line when more streaming applications come to the market. The company wants to disrupt data at rest, and on the other hand, Confluent wants to reimagine the kind of applications that are available once data is in motion. The obvious example is machine learning applications yet the Metaverse will also need data to be in motion.
Confluent is partnered with nearly every database and data warehouse on the market, including Snowflake, Databricks, Google’s database and warehouse products like BigQuery and BigTable, Amazon DynamoDB and RedShift, Microsoft Azure Synapse, Cosmos Data Lake Storage and SQL Server, MongoDB Atlas, Redis, Oracle, My SQL, etcetera. For observability, Confluent is partnered with Datadog. Most recently, the company announced a partnership with Alibaba Cloud – which a bit surprising considering China’s hard stance against United States-based tech.
Q3 Earnings Overview
By Bradley Cipriano, CFA, CPA
In Q3, revenue grew 67% year-over-year to $102.6 million which is an acceleration from 64% growth last quarter. Confluent Cloud drove this acceleration with cloud revenue growing from 200% in the previous quarter compared to 245% in the current quarter. Remaining performance obligations also showed a slight acceleration from the previous quarter at 75% in Q3 compared to 72% in Q2.
The above $1 million ARR customer segment grew 90% year-over-year. Customers above $100,000 slightly decelerated from 51% growth to 48% growth. Due to the size of companies that need a managed solution for Kafka, this is an important key metric to determining Confluent’s growth in the future. Confluent does have a pay-as-you-go option yet managed services for Kafka will likely attract a higher paying customer.
Covid may have affected Confluent’s net retention rate, which declined from 134% in 2019 to 125% in 2020. This number has accelerated to being greater than 130% in Q3. The company is providing near-term targets of 120% and long-term targets of 130% because of Confluent Cloud’s consumption model. Here’s a quote regarding a higher target for NRR in the future: “We're actually at our longer-term target of above 130% and the profile of like in the dynamics of NRR as it relates to our two main products, Confluent Platform and Confluent Cloud, our thesis is that Confluent Cloud will have a higher NRR profile over time because it's elastic, it's consumption-based and there's very little friction in terms of expansion, whereas with Confluent Platform, we're renegotiating deals, etc.”
Adjusted operating losses increased to ($42.6) million from ($19.7) million in adjusted operating losses in the year-ago quarter. The fiscal year estimate for operating losses is ($169) million or adjusted EPS of ($0.91). Free cash flow decreased from ($10.3) million to ($20.6) million.
Confluent Cloud can impact gross margin, at 69.4% down from 71.6%, and subscription margin at 76.8% down from 78.9% a year ago. As the company reaches scale, this is expected to remain around 70%. The company had $1.3 billion in cash as of Q3, which was relatively flat QoQ, and no long-term debt outside of $34 million in lease liabilities. Confluent also receives a significant portion of its revenue upfront in cash, which helps pay for working capital and reduces the need for immediate outside financing. By netting AR from DR (net DR), we can see that upfront cash payments (proxied by net DR) slightly increased QoQ from 83% to 84% of quarterly sales.
Furthermore, there was a degree of conservatism in Confluent’s deferred revenue balance. Revenue recognized from deferred revenue was $118 million YTD, or 74% of beginning deferred revenue. This was down from 77% in the prior quarter, signaling that the recent acceleration in sales was organic and was not driven by pulling forward deferred revenue recognition. This trend also provides more support for future sales, as there is relatively more deferred revenue on the balance sheet, providing a ‘floor’ for future growth.
Confluent also states that RPO is an important metric to monitor in order to measure the health of the sales pipeline. In Confluent’s first conference call as a public company (Q2), CFO Steffan Tomlinson explained that:
“Given the various revenue components and billing terms in our model, remaining performance obligations or RPO and current RPO rather than billings, are important metrics to measure the health of the business. RPO provides insight into the organic momentum of our business as it represents contractually committed revenue to be recognized in the future regardless of billing terms and variability in cloud consumption pattern. RPO provides insight into the organic momentum of our business as it represents contractually committed revenue to be recognized in the future regardless of billing terms and variability in cloud consumption pattern”
As mentioned above, RPO was up 75% YoY to $385 million and RPO to be recognized in the NTM was up 65% YoY to $258 million, both of which represented an acceleration from the 72% and 63% YoY growth rate in the prior quarter, respectively. The acceleration in RPO provides support for future sales. Coupled with the relatively higher rates of upfront cash receipts, Confluent appears well positioned to continue to grow strongly in the near term.
Looking forward, management guided that Q4 revenue will rise 55% YoY $109 million, which would mark a deacceleration from the most recent growth rate of 67% YoY growth. However, this estimate is likely conservative, as management guided that Q3 sales would grow 46% YoY to $90 million and actual Q3 sales grew 67% YoY to $103 million. If we assume that Confluent beats it guide by a similar amount in Q4 as it did in Q3 ($13 million), then Q4 sales growth will accelerate to 73% YoY (this is merely an observation – no guarantees). To be complete, management also guided for Q4 EPS to be a loss of-$0.22.
For fiscal year 2021, the company is guiding for revenues to grow 60% at the midpoint to $377 million, up $27 million from its prior guide of $350 million. For fiscal year 2022, the company is guiding for sales to grow 36% YoY to $511 million, and for Q1 ’22 sales to be flat on a sequential basis, while Q4 ’22 is expected to be the strongest quarter.
Management’s back-end weighted guidance is due to two key trends: 1) the timing of large enterprise deals which are clustered near year-end and are recognized in the quarter of signing, skewing revenue and 2) the recent ramp in sales hiring, which management noted that it takes “roughly four quarters for folks to get fully productive”.
We suspect that management’s forward guide is likely conservative. For instance, cash support for sales increased and RPO also accelerated. Looking forward, Confluent has 64% of its NTM sales guide already secured via RPO commitments, which is high relative to peers. For example, MongoDB’s NTM RPO is 20% of its NTM sales estimate. Furthermore, 70% of Confluent’s NTM RPO is locked-in with upfront cash payments (deferred revenue), further increasing the quality of its forward guide.
Confluent is being conservative with their forward guide likely to ensure that they beat expectations as a new company. While there is limited financial history, there are signs of conservatism as upfront cash receipts increased and RPO accelerated. Confluent remains well positioned to benefit from the secular tailwinds driving “data in motion” in the cloud environment and should continue to grow strongly in 2022.