Exploring Oceans of Firehose Data
When Vertically-Focused Machine Learning Startups are Valuable
With enough distance, any surface appears infinitely uniform.
The ocean? From our seats on the plane it’s just a bunch of blue. The Milky Way? Pfft — not even a grain of sand in the great big multiverse.
Thinking this way can be a bit terrifying — I stopped reading about the Big Bang after losing too much sleep staring at the ceiling, questioning the cosmic meaning of awkward BART rides — but it is an almost essential mindset for machine learning founders looking to build value.
Prediction and automation services working with surface level data result in undifferentiated offerings and step-wise value creation. More importantly, they miss out of the exciting things that come when you dive in, get dirty, and keep going deeper than ever before.
If you believe this — like we do at Comet Labs — it follows that startups creating the most value using machine learning technologies are those that enter markets with:
- A minimum viable product based on some kind of automated service leveraging rich, industry specific data (i.e. fast entry and stickiness); and
- A productization strategy that improves operational efficiencies over time (keeps getting better and better).
One of our recent investments — PingThings — is an excellent example of a strong, focused, warrior team not only solving industry-specific problems but creating original value by journeying deeper than ever before with early industry customers.
We invested in PingThings because they’re the best at what they do.
PingThings is valuable because what they do is challenging, tedious, and boring — and it sets their utility customers on a path to improve their operations by orders of magnitude and in previously impossible, unknown ways.
“We have so much data, we don’t know what to do with it”
Utilities. Auto OEMs. Appliance manufacturers. Governments. Insurance providers. Banks.
If I had a nickel for every time I’ve heard large entities say “we have the data, but we don’t know what to do with it,” over the past decade, I’d have about $3.35. Which isn’t a ton of money, but it is a lot of people reflecting on the real value data access has generated for their operations.
Take utility smart meters.
Once upon a time, when smart meters were first being deployed at scale, the conversation was about how capturing energy usage data at a finer scale would revolutionize utility operations. Fewer losses! Improved demand response! A higher penetration of distributed energy!
In reality, the dominant value of smart meters in industry is…wait for it…the elimination of truck rolls to check meters.
Some dirty math:
- Install cost of utility smart meter (all in) — say $100
- Hourly wage for meter checker — $25/hr
- Minutes needed to check each meter — 3 minutes on average, so thats $1.25/check.
- 12 checks per year, so $15 saved/meter/yr
- Payback = 6 yrs and 8 months, which fits in pretty darn well with utility asset financing.
So — at the risk of being stupidly outspoken — here’s what I’m thinking:
When the value of a deployed data technology can be quantified in the payback period on the physical asset — that’s IoT.
When a deployed data technology is valued by the automated services it creates -that’s industrial machine learning.
Silver Springs and Itron probably want to become utility machine learning companies. They could, but they won’t — and for a couple good reasons:
- They’re already good at what they do, and don’t really need to.
- The sampling rate for most utility smart meters is < 1 Hz, and for utility operations that’s not granular enough to provide value from machine learning today.
PingThings and High Frequency Time Series Databases
Utility synchrophasors, on the other hand, are capable of sample rates greater than 1000 Hz. Thats a data deluge when scaled to thousands of sensors across a utility network, so much so that just acquiring the data breaks current state of the art providers to utility markets.
So, the PingThings dudes out of Berkeley built one of the world’s best databases optimized to ingest high frequency time series data inputs.
And while they entered their pilots hoping for moons and stars — as most startups do — what they discovered was an extremely strong demand for the voltage sag detection capabilities of their platform.
Pingthings are the only company in the world offering a searchable, explorable database capable of capturing and tracking millisecond-timescale, transient events — like a voltage sag — over thousands of sensors and years of operation.
Hmmm — thats a mouthful. And at first there wasn’t a lot of good understanding or agreement about why this new capability is valuable.
But now that the capability exists, the probability that just this single capability becomes a reporting requirement for all utilities is greater than zero, and it is already under review by regulators around the world.
There’s your gateway, and there’s no better stickiness than being told to use a technology.
But what’s the growth opportunity? . According to one of their utility beta customers, voltage sag represents less than 1% of the capabilities of the Pingthings platform, and utilities currently know only about 10% of what they believe the full value will be.
And that’s an exciting point to build from.
KISS and “The Big Deal”
The future of electricity operation combines high precision, automated grid management with high levels of both distributed generation and electric vehicles working in economic synchrony.
The technology to do all this exists today, and the process of implementing it at scale is a function of understanding, coordination, investment, and will. But the amount of operational effort to make this fully functioning dream is massive, with a design that’s currently undefined.
As I noodle on this with the PingThings team, I keep thinking back to one of my earliest mentors, Doug Wert . Heaven got a doozy with old Doug, and when he wasn’t being an outspoken, brash, over-zealous project finance savant, he was browbeating a young Andy to ignore complexities: “keep it simple, stupid!”
KISS — an early lesson in the value of simplicity, and one of the hardest things for technology-bent early stage startups to master.
PingThings has mastered simplicity in their first product offering, one that plays a small but critical role in the complicated, data-driven, multifaceted future of electricity generation and delivery.
More importantly, PingThings is not just knocking on the door, they’ve already walked in.
That’s incredibly valuable, and an exciting first step towards more reliable, safe, distributed, sustainable networks — electric and otherwise.
