This startup used design to make industrial A.I. trustworthy to humans
This is the story of a startup called Phaidra, who makes an advanced A.I. that helps industrial plants save huge amounts on their energy bills. Last year my VC firm Character invested in Phaidra, then ran Design Sprints to help them find product-market fit. Today, Phaidra is announcing a $25M Series A. So this post is a bit of a brag because I’m stoked that the sprints worked and I’m proud as heck of the Phaidra team. But more than anything, I hope this behind-the-scenes case study might be useful to anyone who’d like to use Design Sprints for product-market fit.
I’ve led a lot of Design Sprints over the years and I always have a great time, but this sprint with Phaidra stands out as one of the really special ones. There are a few reasons:
- An exceptional team
- An exceptionally difficult problem
- A happy ending with product-market fit
But I should chill out on the Cliff’s notes and just tell you the story. Check it out:
The backstory: From Google to factory floor
Two of the founders, Jim Gao and Veda Panneershelvam, are world-class artificial intelligence experts. They used to work together at Google’s DeepMind. (If you haven’t heard of DeepMind, it’s the real deal — basically the 1996 Chicago Bulls of A.I.)
During their time at DeepMind, Jim and Veda ran some of the coolest projects there, including the A.I. that beat a world champion at Go and the A.I. that reduced Google’s data center energy costs by 40%. Jim and Veda began thinking about starting their own company, and they were particularly intrigued by the success of that data center project. Where else could A.I. save huge amounts of energy?
Industrial manufacturing.
Factories and industrial plants use tons of electricity on heating and cooling and operating all kinds of equipment. The more they thought about it, the more Jim and Veda became convinced that they could radically improve industrial performance. Building the A.I. would be really hard, but Jim and Veda had built impossible things at DeepMind.
However, writing amazing code wouldn’t be enough. Every factory and industrial plant has different machinery cranking out different products. It’s a much messier landscape than the tightly-designed Google data centers Jim and Veda were familiar with. They needed a legit industrial expert — someone with the expertise to help them figure out technical challenges and the credibility to build relationships with customers.
Enter Katie Hoffman. Katie had worked on engineering and innovation at Ingersoll Rand, Trane, and Raytheon. She’d spent years in a hard hat. She knew industrial manufacturing inside and out. And she was just as excited and optimistic about the potential for A.I. as Veda and Jim.
So together, the three of them founded Phaidra. Jim’s the CEO, Veda the CTO, and Katie the COO. Now they just had to build the product.
A quick side story
Around the same time that Phaidra was forming, we were starting our venture fund, Character. I say “we” but the truth is the whole idea was cooked up by my co-founders John Zeratsky and Eli Blee-Goldman. The first I heard of it was on a phone call with John, which went more or less like this:
JZ: “So I’ve got this idea to start a venture fund based on Design Sprints.”
Me: “What do you know about running a venture fund?”
JZ: “Dude, don’t forget we worked at Google Ventures for all those years.”
Me: “Yeah but we weren’t running it. I didn’t hardly know what was going on. Plus, where are you going to get all the money? Plus, how are you going to convince top-tier founders to go with a VC they’ve never heard of?”
JZ: “Well, I was paying attention at GV. Also, my co-founder Eli knows everything about venture capital and he’s kind of a genius. We’ve figured out how to raise the money. And I’ve already started prototyping the idea by talking to founders. I’m telling you, it’s going to work.”
Me: “Um, do you need another co-founder?”
JZ: “Maybe.”
A few months after this conversation, the fund was launched, and John called me up and told me everything you’ve read above about Phaidra. He also told me he wanted to invest in Phaidra’s Seed round, making them the very first investment in our portfolio. I was stoked. And when I found out that Phaidra wanted to do a Design Sprint, I got even more excited. I’m a sci-fi nerd, and I remembered reading those DeepMind stories and feeling like the future was pretty cool. Now I’d get to work with those guys!
Okay, back to the real story:
The missing piece
In our next conversation with the Phaidra founders, Katie described the company’s biggest, most pressing challenge. In a nutshell, she said, “It doesn’t matter how good the technology is if our customers don’t understand it or trust it.”
Here was the situation: Jim and Veda were having great success building the back-end tech that could learn how any given industrial system worked and make minute-by-minute adjustments to improve efficiency. But Katie saw a problem: If the engineers who ran the plant didn’t trust Phaidra, they might shut it off. It wasn’t enough for Phaidra’s A.I. to work — the human operators needed to know why it worked.
These workers were right to be skeptical — they are experts, after all, and most things marketed as “artificial intelligence” these days aren’t especially intelligent. Phaidra needed to build an interface that could help the A.I. explain and prove itself. If they couldn’t create this missing piece, they couldn’t win over new customers and their brilliant technology wouldn’t matter. That’s where the Design Sprint came in.
Zero to simulation in five days
The Design Sprint was five days long. The team included all three Phaidra founders, plus a designer, a software engineer, and two colleagues from customer-facing technical experts. Katie was the Decider — as the person in charge of the project, she would make all key decisions in the sprint.
Now I’m gonna take you inside the Design Sprint. Get ready for a lot of images and nitty gritty details. You can skim if you want, of course, but if you’re running a startup yourself, or just nerdy like me, maybe you’ll dig it:
Monday
On the first day, the team made a high-level map of how this new software interface would work for various customer types:
Katie identified the key questions we needed to address:
- Would people understand and trust what Phaidra was doing?
- Could the software demonstrate the business value of Phaidra’s artificial intelligence?
At the end of the day, Katie chose a target. The most important customer type, she decided, were energy management engineers who were responsible for the cooling systems in their facilities. And the most crucial moment for those engineers was checking the stability of their systems to make sure Phaidra’s A.I. was prepared for upcoming changes — and wasn’t doing anything wacky.
From there, the rest of the week’s plan was clear. The team would recruit a few energy management engineers from different factories and plants for a test on Friday. And we would spend Tuesday and Wednesday figuring out what that “stability check” should look like and how it should work, then create a realistic prototype on Thursday, just in time to show the engineers.
Tuesday
The team sketched competing proposals for how the “stability” interface might look:
Wednesday
The team reviewed the proposed solutions and had some structured arguments using Miro.
Next, Katie chose the approach she believed would work best for engineers:
Then we created a storyboard — a kind of blueprint for the prototype we would build the next day:
Thursday
Now the team had just one day to build a prototype. Following the storyboard, Mandi (Phaidra’s design lead) built this in Figma:
The prototype would test all of Katie’s key questions about product/market fit:
Friday
On Friday morning, the Phaidra team was ready with a realistic simulation of what the interface could look like. Throughout the day, they tested the prototype with four industrial plant engineers on four separate Zoom calls.
The results were mixed, but promising:
Take a look at those top two questions — those were the keys to product/market fit that Katie identified on day one. It appeared that the “Time Machine” solution was good enough to establish trust with engineers. The team could’ve switched out of Design Sprint mode and started building this new front-end.
But Katie, Veda, Jim, and the rest of the team looked at the scoreboard and didn’t want to stop. They already had hunches about how to close the gaps on those yellows and reds.
Design Sprint 2
So they followed up the next week, sketched new solutions based on their improved intuition, and Katie chose a new approach for their second iteration:
Then the team turned the “Inspector General” into a new and improved prototype:
Finally, they tested with four different engineers. Here’s how the scoreboard changed from sprint 1 to sprint 2:
Now, Katie and the team had momentum and confidence in the solution. As Katie put it: “We accelerated 3–4 months of product progress into 5 days. More importantly, we de-risked our product by testing with real customers.”
They switched into execution mode and started building the “Inspector General”.
Outcome: The Inspector General rides on
Today, Phaidra’s artificial intelligence works as well as Jim and Veda first guessed it would — or maybe even better. So far, Phaidra’s A.I. reduces energy usage at their customers’ factories and industrial plants by 15 to 30%, and does it while going easier on the equipment — which means fewer breakdowns and other headaches for the facility teams to deal with.
As for the console and building trust with customers? The “Inspector General” sketch is now real live software:
The other cool thing is how this UI is helping Phaidra pitch and close deals with customers.
Remember, the second key question Katie picked during the Design Sprint was “could the software demonstrate the value of Phaidra’s artificial intelligence?” The answer is a big fat YES. The team uses screenshots and live demos of the front-end in their marketing and sales process to show the value of Phaidra in a concrete way.
The other secret reason we run Design Sprints
Here’s a secret: we don’t just run Design Sprints with our portfolio companies because we want to improve their chances of success. I mean, first and foremost, yeah, that’s why we do it, and we think it’s great, obviously.
But we have another sneaky motivation as investors: In a Design Sprint, we get to see the team in action, and understand the risks and potential of their business, in a way that most investors don’t get to do.
This is huge. In these two sprints, we spent more than 40 hours with the Phaidra team, and saw customers react to their product live. So when Katie, Jim, and Veda asked if we wanted to invest more in their Series A, it was a totally easy decision. The team is even more awesome than we thought, and we’re even more excited about the product now that we’ve seen inside. So we tripled our investment as part of their new round of funding.
One last ramble from Jake
Design Sprints help teams gain confidence in their own solution, and that’s exactly what happened here with Phaidra — they started the sprints with more questions than answers, and ended with clarity about what they need to build to be successful with customers.
But in parallel, each sprint is helping us at Character to build confidence in our solution for our customers, who are startup founders. We’ve run many sprints since this one with Phaidra, but Phaidra was our first investment at Character and our first full Design Sprint with a portfolio company. We’ve got this crazy idea that a venture fund based on Design Sprints can work, and each successful sprint gives us more confidence that it will.
For me, this whole thing was really fun to see. On one hand, sure, I’m confident about Design Sprints. Like I said, I’ve led a lot of them. I now expect that two back-to-back sprints will usually get a strong team from zero to on-track, and I’m used to watching key questions flip from red to green as the team learns and hones their intuition. Hey, if we didn’t believe Design Sprints help teams find product/market fit, we’d have no place starting a venture fund on the premise. I should be confident, right?
But on the other hand, there were a lot of reasons to feel nervous at the start of this sprint. Our first sprint as Character. A hardcore computer science problem in an unfamiliar context. Brilliant founders on a mission to transform multiple industries… and here’s me, an art school dropout, ready to waste a week of everyone’s lives. It could’ve gone wrong in so many ways, so it’s pretty rad that the magic worked.
