Arena: Market Conditions Before Datadog
Software development and operations teams moving to the cloud · 2010 · United States
Changing technical requirementsDisconnected workflows
Software teams moving to cloud servers could add and remove machines faster than monitoring built around fixed servers could follow. Developers and operations staff watched the same application through separate tools for server metrics, application performance and logs, and when something failed each side tended to blame the other. Finding the cause of an outage meant switching between those tools and assembling the picture by hand, and teams that ran open-source monitoring themselves also had to maintain it DD1 DD6 DD10.
How each step happened
Step 1 of 5 · 2010–12
Infrastructure first: where outage diagnosis starts
At Wireless Generation, a New York software company, Olivier Pomel ran development and Alexis Lê-Quôc ran operations, and their teams blamed each other whenever something broke DD19 DD25 DD6. In 2010 they set out to give both teams the same data during a failure, at a time when monitoring, application tracing and logs were bought as separate tools DD19 DD10.
They started with infrastructure monitoring for cloud hosts, servers and instances that can appear and disappear within hours, and launched it in 2012 DD1. That put the product at the first question of any outage investigation: which host is unhealthy, and where. It also put the agent, the program that collects the data, onto nearly every host, because every host needs basic monitoring while not every host needs tracing or log analysis DD10. Pomel calls it the best product to land with because it touches more of a customer's environment than anything else DD21. One data model tied every signal together with tags, labels such as service, region or team DD19, so the next steps of an investigation, from a host to its requests and their logs, could later sit on the same data.
Rivals New Relic launched in 2008 as a performance tool for Ruby on Rails applications and was usually installed only on application servers; it bought Opsmatic for infrastructure monitoring in November 2015, and its own infrastructure product entered private beta in August 2016 DD28 DD10 DDX-1 DD29. Open-source tools such as Nagios and Graphite covered servers without a view shared with developers DD10.
End-to-End WorkflowFounder Domain Expertise
Step 2 of 5 · 2010–15
Open agent and integrations fit any stack
The agent was built alongside the first product, so this step also starts in 2010. The founders spent six months talking to potential users before writing code, and a wide beta taught them more than a hand-picked alpha because users could select themselves DD20. The agent sat in a public repository from December 2010, and all of Datadog's collectors were open source DD31 DD19, so an engineer could read the code before running it on the company's machines.
Integrations carried the same collection to each system a customer ran: AWS, Azure, HP Cloud, OpenShift and OpenStack by 2014, Docker and CoreOS by 2015 and more than 350 by 2019 DD24 DD26 DD1. Each let another kind of customer install the product without professional services DD1, and Pomel traces product/market fit to them DD27. Integration breadth is what let a low price per host reach every host.
Rivals New Relic's agents were written per language, a Rails plug-in in 2008 and a separate Java agent in 2009, and Peter Offringa says their design assumed static servers DD28 NR15 DD10. New Relic released its agents and integrations under an open-source license only in July 2020 DDX-2.
Drop-In AdoptionCustomer Discovery
Step 3 of 5 · 2012–19
Land and expand on per-host pricing
A price meant to reach every host
Pomel says the first objective was to be deployed as widely as possible, not in one corner with the rest left out for lack of budget, so the price was sized as a fraction of what customers spent on infrastructure DD19. The 2014 terms were a 14-day free trial and $15 per host a month DD24. The charge followed hosts, not users, so every engineer could see the same data at no extra cost DD24 DD1. There was no freemium plan, which Pomel says has no upsell path for a product that must be everywhere DD19. Contracts ran month to month, so a wrong product would show up quickly as lost customers DD20.
Selling after use
The first head of product also ran sales, and product teams still lead the first sales of each new product DD6. Tech-savvy midmarket companies bought first; the board concluded that enterprises moving to the cloud would come to Datadog DD19 DD9. When they did, around 2016, an enterprise sales force followed, but adoption inside them stayed bottom-up: the CIO passed the product to teams, and sales came back later with an order form DD19.
Accounts that grew from the first host
Zendesk started with infrastructure monitoring in 2013 and added application tracing in 2018 DD1 DD24. The 2019 filing calls the model land and expand. Its measure, dollar-based net retention, divides this year's recurring revenue from a group of customers by what they paid a year earlier; it was 146 percent in mid-2019 DD1. Existing customers supplied about 75 percent of the revenue increase in 2022 DD18. Usage pricing works in both directions: when customers cut cloud spending in 2023, second-quarter growth slowed to 25 percent DD16 DD17.
Rivals New Relic priced by server, with tiers at $200 a month per server and a free Lite tier in 2008, and later moved larger accounts to site licenses and field sales DD32 NR11 NR12. Its net expansion rate was 109 to 112 percent in 2019, against Datadog's 146 percent, and its July 2020 move to prices per user and per gigabyte promised that customers would no longer have to sample which applications they monitor DDX-6 DDX-5 DD1 DD30.
Product-Led SalesFreemium ModelUsage-Based Pricing
Step 4 of 5 · 2017–19
One agent and tag model across products
Datadog sold only infrastructure monitoring until that product fit DD19. Application tracing, which follows each request through the code, became generally available in February 2017. Its traces were grouped by the same tags as the infrastructure data, so engineers could move from an unhealthy host to the requests running on it, and one Enterprise price of $23 per host covered both DD23. In September 2017 Datadog bought Logmatic.io, a French log-management company, for its product and team; Pomel wrote that the greatest value lay in moving between metrics, traces and logs DD22. Log management launched in 2018 DD1.
Each new product reused what the first had put in place: the agent on nearly every host, the integrations, the tags and engineers already working in the service DD10 DD19 DD1. A new product arrived already installed, already labeled and already familiar. Customers using more than one product rose from about 10 percent in mid-2018 to about 40 percent a year later DD1. At Zendesk, separate tools had divided teams; one tool let engineers connect application behavior to infrastructure and logs, reuse dashboards and train colleagues once DD14. The expansion paid for research and development, which reached 45 percent of revenue in 2022 DD11 DD18.
Rivals New Relic added infrastructure, logs and other products that Offringa says felt bolted on and not fully integrated until New Relic One in September 2019; logs linked to traces across all its application agents in January 2020 DD10 DDX-4. Splunk priced log analysis by volume, so customers sent it only their high-value logs DD10.
Scope economiesMulti-Product
Step 5 of 5 · 2018–
Switching costs
Being deployed everywhere and used by everyone put Datadog's dashboards and alerts into every team's daily work, across three products DD1 DD19 DD14. Leaving means rebuilding them and retraining the engineers who rely on them. A financial infrastructure company that moved about 40 dashboards and 50 alerts to Grafana took about a year DD13.
The same cost once protected the incumbent. Cloudbeds' move from New Relic to Datadog was technically simple, but retraining developers stretched a planned two months to about six DD12. The barrier delays a switch and does not stop one when the alternative is worth enough: Cloudbeds left for stronger logs, the Grafana customer for cost visibility and open source DD12 DD13. Consolidation also concentrates risk; Datadog's March 2023 outage affected all services across multiple regions DD3. The defense rests on what steps 1 and 3 created: a product on every host, priced so that every engineer uses it.
Rivals New Relic's installed position delayed Cloudbeds but did not keep it, even after a discount of nearly 90 percent off its new list price; New Relic was taken private in 2023 DD12 NR5. Grafana Labs won a Datadog customer that accepted a year of migration work DD13.
Switching costs