Ask anyone how they feel about their dashboards, and you’re likely to get an earful of complaints. Despite everyone agreeing that dashboards ought to be mission critical tools, which monitor the beating hearts of their businesses — metrics related to revenue, user behavior, and application performance — business dashboards remain hard to build, difficult to maintain, and ultimately painful to use. We believe this isn’t the fault of data teams: it’s that the tools and workflows for how dashboards are built must be reimagined.
How we got here
Rill was founded just over two years ago, built on a team and technology that was spun out of Snap. In that time, we’ve validated that business users prefer our fast, flexible, simple dashboards to more complex BI tools. Our fully-hosted Rill Cloud offering now has 1000s of users at companies like Comcast, AT&T, and Ericsson with use cases in digital advertising, e-commerce, and application performance monitoring.
However, even though our customers loved our dashboards, we had a realization: the journey of onboarding their data into our service — of profiling and transforming data into metrics — remained an excruciating challenge that no existing product solved well.
Today, we’re excited to share we’ve raised $12mm of capital from a supergroup of investors to tackle this pain point head on, inspired by our own experiences.
Our new product, Rill Developer is a free, open-source tool that radically simplifies the data to dashboard journey, enabling analysts to effortlessly shape data into the metrics that power business dashboards. We hope Rill Developer will make it easy for analysts to build dashboards that their business users will actually use.
It’s 2022, why do business dashboards look like PDFs?
For better or worse, today’s business dashboards are a direct descendant of an era of desktop publishing tools, where analysts embedded bar charts in a report. Somehow, despite geometric increases in computing, graphics, and storage — business dashboards remain one step removed from a high-end printed PDF from Kinkos.
This lineage has constrained our imagination of what dashboards could actually be, and cut us off from considering entire new classes of interactions that modern software tools have embraced.
Why?
Dashboards running on data warehouses are sloooow. When you need a cup of sugar, do you drive to Costco? Like real warehouses, cloud data warehouses are optimized for cheap storage, not for the fast service that dashboard applications deserve. Not only is querying event-level data from cloud warehouses slow, like driving 15 minutes for a cup of sugar, it’s needlessly expensive. OLAP databases (Clickhouse, Apache Druid, Apache Pinot, and DuckDB) which aggregate data and keep it in-memory for speed, are an ideal serving layer for dashboards, but orchestration of data into these engines is a complex task for data teams to manage.
You shouldn’t need a Modern Data Stack to build a dashboard. An analyst’s journey in the last mile from data to dashboard is much harder than commonly acknowledged, but it need not require signing up an entire stack of vendors. The first step is exploratory data analytics of some data source (”What’s in this S3 bucket / database table?”) Analysts write dozens of profiling queries to understand cardinalities, null percentages, and frequency statistics. Then begins the testing of various data transformations (”What happens when I truncate this OS version string?”) Finally there’s the switch to a BI tool where metrics definitions, formatting, and layout decisions are made (”Does this average revenue per user trend make sense?”) Because of the multiple steps, programming languages, and dependencies, analysts fear making breaking changes, and dashboards grow stale.
Dashboard UX is a dumpster fire. Most BI tools lack strong opinions for how to display data, forcing analysts to become visual designers, fussing with layouts and chart types. This is a distinctly different job than what they were hired to do, which is shaping data and designing metrics. This lack of consistent UX confounds business users, for whom each dashboard is a new kaleidoscope of charts, when all they want is a quick answer to a simple question like “How many orders did we ship to Phoenix last week?”
Why analysts and business users love Rill dashboards
Rill dashboards are joyful because they’re fast. Whole new classes of interactions are possible when dashboards are tightly connected to the data they visualize. For practical reasons, most BI tools are thin applications with no data engine of their own, and only as fast as the database they sit atop. Rill, on the other hand, is a thick application that comes with its own embedded in-memory OLAP engine, and remains fast even with data sets in the billions. Rill does not merely copy data from its sources, it transforms it on ingestion into a modeled, aggregated version of data that is typically 10x or 100x smaller than the original. This modeling step is done with business requirements in mind, and enforces a level of discipline (”what do we need to store” versus a “let’s keep everything” approach) which reduces data footprint and costs [1], but most importantly makes dashboards fast.
Rill’s data to dashboard journey is all in one tool, with just SQL. Rill Developer allows an analyst to rapidly iterate at each step of the data to dashboard journey, beginning with exploratory data analysis (”What’s in this Parquet file?”), continuing with data transformation, and ending with the design of metrics and dashboards — within one tool, with just one programming language, SQL. Changes to any transformation or metric definition are updated keystroke-by-keystroke — there is no run button — and instantly reflected in dashboards. The result is a conversation-fast experience for building dashboards, that empowers analysts to make frequent updates with confidence.
Rill’s dashboard UX is simple yet flexible, and thus usable. Analysts using Rill do not actually design dashboards, they design the metrics model that underlies these dashboards. Rill then renders these metrics within a standard layout of line charts, tables, and UX elements. As a result, all Rill dashboards have the same look and feel, differing only in the data presented to end users. The benefit of this universal interface is that it enables users to filter, sort, and pivot to an entire sequence of questions — a single flexible dashboard application instead of many reports. A product manager at a digital audio marketplace, who has over 100 partner-facing dashboards in production, quipped: “Rill requires little to no training.”
What is Rill not doing?
We’re not tackling complex ETL pipelines. Developers using Rill are expected to meet our product with a form of structured data already existing in a warehouse, data lake, or stream. Some of the early adopters of Rill Developer are using DBT for their pipelines, exporting a Parquet file as the last step. For our enterprise customers, Rill offers our expertise to help build and run these ETL pipelines.
Why did we raise from this Data Supergroup?
Beyond the capital from a group of leading venture capital firms, we also reached out to people behind the companies, products, and technologies we admire, and we’re incredibly grateful to have a true Data Supergroup investing their capital and wisdom.
Institutional Investors
- Bloomberg BETA
- DCVC | Data Collective
- Flex Capital
- Park West Asset Management
- Sierra Ventures
- True Ventures
- 468 Capital
Angel Investors
- Jim McKelvey, founder of Block fka Square
- Wes McKinney, creator of Pandas, founder of Voltron Data
- Lukas Biewald, founder of Weights & Biases
- JJ Allaire, founder of RStudio
- DJ Patil, former US Chief Data Scientist
- Jeff Hammerbacher, founder of Cloudera
- Hilary Mason, founder of Fast Forward Labs
- Anthony Goldbloom, founder of Kaggle
- Caesar Djavaherian, founder of Carbon Health
- Felix Feng, UCSF Professor, founder of Artera
- Mike Franklin, CS Professor, founder of AMPLab
- Sean Taylor, creator of Prophet
- Abdhur Chowdhury, founder of Summize, Twitter Chief Scientist
- Sean Gourley, founder of Primer AI
- Pete Skomoroch & Sam Shah, founders of Skipflag
- Scott Yara, founder of Greenplum
- Kaz Ota & Hiro Yoshikawa, founders of Treasure Data
- John Hughes & Brett Wilson, founders of TubeMogul & Swift.vc
- Amar Goel, founder of Pubmatic
- Katrin Ribant, founder of Dataroma
- Kevin Weil, Chief Product Officer at Planet Labs
- Florian Leibert & Toby Knaup, founders of D2iQ fka Mesosphere
- Jim Payne & Nafis Jamal, founders of MoPub
- Auren Hoffman, CEO of Safegraph
- Gil Elbaz, CEO of Factual
Where do we go from here?
Rill aims to unlock a lower friction, higher velocity customer adoption path, and grow into the fast-expanding market for real-time data applications, of which metrics dashboards are the first canonical example.
Footnotes
- [1] Before adopting Rill, one of our customers was spending tens of thousands per month on warehouse queries to generate hundreds of static reports; after Rill, they reduced their data footprint by 10x, cut their overall BI costs in half.
