Yellowbrick Data lets you keep the keys to your data

At ACI Worldwide, a company whose software helps move money, the fraud team had a scheduling problem. Some analytical work had to wait until the end of the day to avoid disturbing customers. There is something faintly absurd about a system for understanding risk that asks its users to come back after hours. In Yellowbrick’s customer account, the replacement warehouse let the team run those queries when it needed them. The purchase changed the timetable as well as the technology.

01The query that had to wait

ACI’s published case study describes more than 100 terabytes spread across 30 billion rows and over a thousand columns. Holiday payment surges add another complication. A warehouse must ingest data, support operational work and answer difficult questions while demand keeps moving. A stopwatch beside one query cannot tell you whether all those jobs will get along.

“We don’t have to wait until the end of the day”

Radu Medesan · ACI Worldwide
Excerpt from its Yellowbrick customer account

Yellowbrick Data sells an answer to that crowded-room problem. Its commercial SQL platform serves data engineers, analysts and application teams working with enterprise-scale information. Customers use it for fraud analysis, advertising measurement, business reporting and other work where an answer loses value while it waits. The company competes in the warehouse market, but its pitch pays unusual attention to who owns the room.

02Keep the cloud. Keep the keys.

Neil Carson, Mark Brinicombe and Jim Dawson founded Yellowbrick in 2014. Carson had been CTO at flash-storage company Fusion-io. The original warehouse became generally available in September 2017; its public unveiling followed in July 2018. It was a compact appliance built around flash storage, with production customers already aboard. The founders’ expertise extended from database software down into the machinery that moves data.

NEIL CARSON

MARK BRINICOMBE

JIM DAWSON

By June 2022, Yellowbrick had announced an elastic cloud-native release. CTO Mark Cusack’s accompanying explanation described a journey from hosted optimized hardware into public cloud, driven partly by customers wanting to avoid concentration in a single cloud. Kubernetes supplied portability. The product’s address became more flexible while its emphasis on deployment choice continued.

Today it can run in customer-controlled AWS, Azure and Google Cloud accounts, as well as on-premises. Yellowbrick calls the arrangement a Private Data Cloud. Storage and the management plane sit inside the customer’s environment. An insurer or government organization can choose its location and apply its own access policies rather than make geography an afterthought.

That distinction has practical appeal wherever sensitive information and residency requirements shape purchasing. It also creates work: somebody must manage the cloud account, permissions and infrastructure budget. A deployment boundary helps establish control; competent operations still have to exercise it.

03A familiar door, a different engine

Under the familiar SQL surface is a purpose-built parallel engine. Yellowbrick uses column-oriented storage for large analytical scans and a row-oriented store for recent streaming data. Separate storage and compute let organizations change processing capacity without treating stored data as part of the same indivisible purchase. Workload management allocates resources among jobs competing for attention.

The front door speaks PostgreSQL-compatible protocols. Existing drivers and tools can therefore feel familiar, although compatibility is not a promise that every feature or query behaves identically. Yellowbrick names Next Pathway and Datometry among its migration partners. Moving a warehouse includes schemas, pipelines, reports and testing. Familiar syntax is a useful beginning; the unglamorous checking is still required.

Catalina Marketing supplies a concrete example. Aging Netezza hardware, slow queries and rising costs prompted a search. A proof of concept helped persuade Catalina to switch. Its Yellowbrick-published account reports 182 times faster queries and eight times faster query concurrency. More revealingly, media teams stopped having to organize their work around the old system’s restrictions. Those are results for that customer’s environment, not a forecast for every buyer.

Zurich’s selection announcement puts price for performance and ease of migration at the center of its decision. Its reporting improvement offers a useful reminder: enterprise software earns its keep against a business deadline. The worthwhile benchmark is the one attached to a job people actually need to finish.

04The price has two parts

Yellowbrick sells software through subscriptions and metered on-demand usage. The infrastructure has its own bill. A customer running in its cloud account pays the cloud provider under its own arrangements, then pays Yellowbrick for the database software. This makes the arithmetic visible, even though the final answer still depends on storage, compute, networking and how often the system runs.

For illustration, 64 vCPUs on the one-year rate produce a $39,232 annual software bill. The three-year rate gives an annual equivalent of $30,848. Sixty-four vCPUs running on demand for 100 hours cost $1,792 in software. These calculations share a capacity figure, but different usage periods; they are examples, not deployment quotations.

A subscription suits sustained demand differently from an occasional burst. A sensible assessment counts busy hours and quiet hours, then includes migration and operational effort. The appeal of a clear rate card is that a finance team can question the assumptions without first decoding a proprietary unit of currency.

05The rival in the next room

Yellowbrick overlaps with Snowflake, Redshift, BigQuery, Databricks SQL and established enterprise warehouses. Its positioning combines analytical performance with customer-controlled deployment. Yet customers need not choose one vendor for every job. NCSolutions uses Yellowbrick across AWS and its data center to compute advertising insights, then distributes derived data through Snowflake Marketplace. The apparent competitor becomes a route to customers.

This is a useful way to read the market. Buying a compute engine and choosing a distribution network are separate decisions. NCSolutions keeps commercially sensitive source information under its control while sharing the resulting insights. For another company, a tightly integrated managed service might be the more convenient arrangement. Workload, distribution needs and operational capacity should decide.

Investors have funded Yellowbrick’s attempt to win that enterprise niche. Four announced rounds total $248 million, including an $81 million Series C in 2019 and a $75 million Series C1 in 2021. Citadel was identified as both a new investor and a customer in the latter round. A 2025 TechForward award added recognition; its public developer repositories show continued activity in 2026.

06Test the queue, then buy the engine

The lesson a reader can borrow is a purchasing method. Run a proof of concept with real data and concurrent users. Keep ingestion running. Include the awkward reports, check the outputs, and price the whole environment. Catalina’s evaluation and ACI’s scheduling problem both suggest looking at how a system behaves when work overlaps.

Yellowbrick’s proposition fits organizations with demanding SQL workloads and a reason to control deployment. A small team with modest data and little appetite for infrastructure ownership may find a simpler hosted service sufficient. Teams expecting complete PostgreSQL equivalence should test their dependencies. There is no useful shortcut around those conditions.

Even Yellowbrick occasionally allows itself a joke. Its 2022 conference release mocked pre-canned demonstrations and promised cats on horses. Database marketing can be a solemn business. The more consequential question remains refreshingly ordinary: when someone needs an answer, must they wait until everyone else has gone home?