Data quality software startup Anomalo wants to help customers watch their data for weirdness.
Anomalo has raised $38.95 million in funding so far, with a $33 million series A round in October. The founders, Elliot Shmukler, CEO, and Jeremy Stanley, CTO, met at Instacart where they encountered the problems that Anomalo is now trying to solve. The company already counts BuzzFeed, Discover Financial Services and Substack as customers, and claims 7-figures of annualized recurring revenue.
Data is often messy, and weird, and flawed. Transcription errors, and incompatible APIs, simple mistakes and just the complexity of modern supply chains all make automated data processing a challenging task. A traditional rules-based approach is slow and painful at the best of times, and when dealing with the massive, fast moving datasets common to modern enterprises (particular in industries like fast moving consumer goods) it can be hard to keep up.
“To set up rules, you have to know what to look for already,” said Shmukler. The challenge for teams is that knowing what to look for in advance in a rapidly changing sea of data becomes an intractably difficult problem to solve. It's the novel, the strange, the anomalies that indicate both where the problems are, but also the opportunities. Just checking everything adheres to the known rules misses all of the novelty that should be the focus.
Anomalo co-founders Jeremy Stanley, CTO, (left) and Elliot Shmukler, CEO (right)
By feeding an existing set of historical data into Anomalo, the system learns what normal looks like, including all the variability, seasonality, and regular changes that normally occur. “Anomalo learns about the data itself, but also the relationships between types of data,” says Shmukler. This enables Anomalo to see how certain segments of data relate to one another, and pick up correlations beyond “revenues go up when sales volumes increase”.
“Anomalo provides traceability back to the source data that triggered the issue,” says Stanley. Beyond mere rows of data, Anomalo provides context about the common characteristics, such as the segment of the data that is experiencing the weirdness. The founders provided examples of assumed constraints such as “suppliers will never forget to include meat in customer orders from two states” that are difficult to write rules for in advance. Knowing why a certain subset of customers aren't getting their full order, and then figuring out why requires more than transaction-level monitoring.
Importantly, Anomalo doesn't go so far as to say “this is definitely the cause, stop thinking now”, potentially leading the human decision makers to rely too heavily on the machinery instead of using their own brains. Instead, Anomalo points a spotlight at the best place to begin reasoning about the problem. It acts in concert with human beings, enhancing their abilities, not replacing them. Anomalo, in essence, asks the humans “Huh, that's funny. Does this look weird to you?”
Customers tend to be skeptical at first (which is a good thing when it comes to overhyped technology fashions like AI and ML) but they are soon convinced, according to Anomalo.
“The system earns their trust,” says Shmukler. Compared to their own home-grown methods (or nothing at all), customers see Anomalo working in clear, practical ways on a specific dataset, and so they start to task it with new datasets to watch.
“Each new problem is a new experience that you can't easily write rules for in advance,” says Shmukler.
The straightforward practicality of Anomalo's approach impresses me far more than the breathless claims made across the industry, and I hope to see more of it.
