There is a particular kind of embarrassment familiar to anyone who has run a data team. The dashboard says one thing. The billing system says another. Marketing wants an answer by lunch. Engineering has a queue of requests and no appetite for yet another tracking SDK. At 8x8, Soumyadeb Mitra was supposed to build a secure customer data platform at enterprise scale. He tried buying parts of the answer and building the rest. The project, he later wrote, never quite met expectations.
- RudderStack collects customer events, cleans and routes them, then uses the customer’s warehouse as the central record.
- Its first sale followed an open-source Hacker News launch, to a company whose existing CDP was struggling to scale.
- Customers such as Crate & Barrel and Khatabook use it for distinct reasons: faster activation, cleaner identities, or lower event costs.
- The useful idea to borrow: settle ownership and event definitions before buying another dashboard.
That failure supplied a sharper brief than a marketing survey might have. Mitra wanted a system that could collect the click, the purchase, the app action and the support interaction, then deliver each to the right place without trapping the whole lot inside a vendor’s private database. He and a team built a prototype called Rudder and put the code on GitHub. In 2019, RudderStack was born.
The first customer already had a CDP
The early company presented itself plainly on Hacker News as an open-source alternative to Segment. When its launch reached the front page, Mitra said, the repository gained hundreds of stars and the hosted app got a wave of signups. One became its first paying customer. That customer already used a competitor, but the system was not scaling and the team was considering an internal replacement. RudderStack arrived at exactly the point where curiosity had become a budget line.
The sequence matters. Open code let engineers inspect the machinery. The managed version gave them a way to avoid operating all of it. That is the business model in miniature: make the plumbing visible, then charge for a supported, hosted service and enterprise capabilities. In 2020, RudderStack raised a $5 million seed round and acquired Blendo, which moved data from cloud services into databases. Mitra said the acquisition spared his team the peculiar labor of handling every cloud API and schema themselves. A $21 million Series A followed in 2021; a $56 million Series B in 2022 brought total funding to $82 million.
“I explored every option on the market, ended up both buying and building solutions, and the project never fully met expectations.”Soumyadeb Mitra, recalling the 8x8 project
A diagram of where the data goes
Customer data platforms often sell a finished portrait of the customer. RudderStack sells more of the route to that portrait. An SDK captures an event in an app or site. Event Stream sends it toward the warehouse and other destinations. Transformations can rename, filter or enrich it on the way. Tracking plans and governance controls can catch malformed or sensitive data. Profiles resolves identities and models the customer in the warehouse. Reverse ETL sends useful traits and audiences back out to marketing, analytics and service tools.
This is where the comparison with Segment and other packaged CDPs gets specific. RudderStack argues that an enterprise already has billing, transaction and operational data in its own systems; sending every record into yet another vendor store is awkward and can complicate governance. Its architecture gives the warehouse a starring role. Snowflake and Databricks integrations make that position practical. The price is that a buyer needs a capable data team and a well-run warehouse. A small business wanting only an instant marketing audience may prefer a more prescriptive product.
What the customer actually bought
Crate & Barrel’s account is an unusually clear example. The retailer wanted timely, personalized campaigns, but its event pipelines were brittle and customer identities incomplete. It rebuilt event collection with RudderStack SDKs, used Transformations to standardize data, and moved modeled warehouse traits into tools including Salesforce Marketing Cloud. The company reports a tenfold increase in personalized campaigns. The number is a customer-reported outcome, not a controlled experiment; the useful detail is the order of operations. Better messages came after better records.
Khatabook, an Indian financial app, faced another kind of problem: volume. Its case study describes four billion monthly user events and says moving from Segment to RudderStack cut CDP spend by 90%. Its engineers spent 15 days validating datasets across sources and destinations. That detail prevents the savings figure from becoming a magic trick. The software changed the economics, but people still checked whether the events and identities survived the move.
The first two figures are RudderStack’s own 2025 totals. The third is a single customer’s reported result and should not be treated as a typical saving.
The price of clean pipes
RudderStack’s public cloud pricing in September 2026 puts its Free plan at 250,000 events a month. Growth starts at $265 a month for one million events. Enterprise contracts are quoted individually and add more advanced governance, security, support and access to Profiles and Data Apps. The self-hosted project remains available on GitHub under Elastic License 2.0. Those details make the buying question concrete: count events, check which capabilities the team needs, and price the warehouse and engineering work alongside the subscription.
The company’s technical taste is equally revealing. In an engineering post, Mitra explained why RudderStack used PostgreSQL rather than Kafka for an early event queue. Failed deliveries needed metadata, repeated retries and per-user ordering; a mutable database made those cases easier to express. It is a small decision with a broad lesson: choose infrastructure for the failure you must handle, not for the logo you wish to put on an architecture diagram.
Now the agents want a turn
In 2026, RudderStack added RudderAI, MCP tools and Lookout, an AI analytics and instrumentation workspace introduced in private beta. These tools aim to let teams ask why a pipeline failed, inspect tracking, build dashboards and propose instrumentation changes. The company says its MCP layer exposes more than 30 tools. The persuasive use case is a narrow one: an assistant can trace a missing event to a bad schema or destination error and give an engineer a reviewable fix. An agent cannot rescue a company that has never agreed what “customer” or “purchase” means.
That is the pleasingly unglamorous thesis beneath RudderStack’s growth. The company says it delivered 3.3 trillion events for more than 4,000 organizations in 2025. Impressive traffic, certainly. But the original 8x8 frustration was never a shortage of data. It was the inability to make the pieces agree. RudderStack’s bet is that a company should own the place where those pieces meet, understand the route each took, and be able to change the route when it breaks. For anyone still reconciling three dashboards before lunch, that is a more useful promise than another pretty portrait of the customer.