In Conversation with Carolyn Mooney, CEO, Nextmv

Our business lives are full of optimization problems – scheduling, time management, resource planning, pricing, routing, risk management, network optimization, financial engineering, etc. Simply defined, optimization is the science of making the best decision possible, given a set of constraints.

Historically, optimization has been the province of PhDs with deep backgrounds in mathematics, using a generation of software that was developed for academia and large defense contractors.

Enter Nextmv (proncounded “Next Move”), a company in which I’m a proud investor. Nextmv is reinventing the space for the cloud era, making optimization and simulation technologies available to every developer.

It was great to welcome Nextmv’s CEO, Carolyn Mooney, at our most recent Data Driven NYC to talk abotu the space and the company.

We covered:

  • What is decision intelligence, and how does it differ from business intelligence and data science?
  • What is the overlap with the area known as “operations research”?
  • How decision intelligence is broadly horizontal area
  • How Nextmv is democratizing decision intelligence with its cloud product
  • Bonus: Nextmv’s policy of radical transparency on team compensation

Below is the video and full transcript.

(As always, Data Driven NYC is a team effort – many thanks to my FirstMark colleagues Jack Cohen, Karissa Domondon Diego Guttierez)

VIDEO:

TRANSCRIPT [edited for clarity and brevity]

[Matt Turck] You are the CEO and co-founder of Nextmv, which is a decision automation platform for developers. Is full disclosure, I have the privilege and honor of being a very proud investor and I’d love to start the conversation with a little bit of your background and your journey to starting the company.

[Carolyn Mooney] My background is in systems engineering. I started out working at Lockheed Martin in the ballistic missile space and radar space, which is not a common startup background to be sure. But it was really interesting, because after spending a bunch of years there doing modeling and simulation for the Navy, I ended up landing at a startup in Philadelphia called Zoomer, which is where I met one of my co-founders Ryan. He and I were building out simulation and dispatch services for Zoomer at the time, which was meal delivery. Very shortly after, we got picked up by GrubHub and started building out what we called the decision engineering team at GrubHub. This existed in the space around decision intelligence and how basically to make decisions at scale. We were covering everything from forecasting and scheduling to automated dispatch, to ETA management and real-time supply and demand shaping.

It was really an informative time because we realized there was just not a lot of good infrastructure for building the kinds of systems that we needed to automate. That was how we started Nextmv. We said, “Wow, this is really hard. And it’s taking a lot of engineering resources and a lot of data scientists to build these algorithms and to deploy them. So, let’s build a platform around this and give the engineer a better infrastructure.”

Let’s talk about the space in general. I remember when I first came across you guys and Nextmv, despite all the time I spend in the data world, I was reasonably unaware that this whole [decision intelligence] space existed and it was fascinating to realize how large and broadly horizontal that space is. Do you want to talk about what decision intelligence, decision automation, what all of that means?

It’s easy to start with the fact that billions have been invested in both data science and AI, and this is a fact and people have been talking about this space for a while.

You build on the digitization era, which was like, “Make sure we have event data for everything.”

And then from there you went to BI, which answers questions like “what’s happening in my world?”, “Can I understand what happened in the last week?” etc.

And then you went to data science, which was fundamentally answering the question, “what is possibly going to happen?”. That’s predictive modeling.

Where decision intelligence sits is the next layer on top of that. Decision intelligence answers the question: “[I have a sense for what’s going to happen,] what should I do about it?”.

That’s why we picked the name Nextmv, so what’s your next move? It’s really that space.

Decision intelligence is kind of the next evolution of a data stack.

What would be an example of decision intelligence at play?

A simple example could be a subscription box. I’m a user of Stitch Fix or Birchbox. And these types of companies, they may have a data science group that’s working on, what is the likelihood that I’m going to like an item? So they’re trying to predict if I’m going to like this sweater. And that’s awesome, except they have tons of subscribers and they have a very limited inventory of this sweater. So how do they decide which person gets this inventory? And so they have limited inventory, they have lots of people. How do you decide who gets what items? And you probably want to try to maximize your ROI. I mean, every company’s trying to impact that bottom line. And so in a company like that, you need to make that network level decision. That could be something that we are doing manually. If it’s easy, it’s five people and I have 10 items, cool, we can probably do that on paper. But when you start to have thousands of things and your planning space explodes to thousands or millions of options, we should be using compute power for that.

That’s really where decision intelligence comes in. So it’s basically what should I do about all this data, insights and predictions that I have?

There is a lot of overlap with the space known as operations research. Is that correct?

The space around operations research is really around optimization as a technology and simulation as a technology. These are what we refer to as decision-making paradigms.

Operations research technologies have been around for a while and they are used to make decisions, but they are typically thought of in very academic communities and implemented on legacy tech stacks. And we use some of those platforms around that space.

For example, in the optimization space, you have “solvers”. Solvers are a fancy way of saying, “Hey, I’m going to generate all the possible plans and pick the best one based on my criteria.” My criteria could be, I care about not being late for a delivery service or in that Stitch Fix example, I care about what is my maximum return on that allocation. So these are KPIs if you want to think about it in the most general sense.

Operations researchers are these PhDs that are used to thinking about that space in mathematical terms. They take all of that business context that we just talked about and they boil it down to a matrix map. Effectively, they’re translators — translators that are very well educated obviously, and trained to use these legacy tech stacks and they come into a business and they’re trying to optimize, they’re trying to get that benefit of a 10% improvement on margin for a delivery company, or improving how you allocate marketing budget or something like that. So, that’s operations research and we hire operations researchers on our team. My co-founder is one of them. But we feel it’s amiss to not have software developers in the mix also.

What are some examples of these legacy software or products?

Some of the legacy players in this space are IBM, which has a product called CPLEX. FICO has a product called Xpress that we actually use in our prior work. And then also Gurobi is a big one. They spun off of IBM a few years back.

These are platforms that people use in their academic research and then take over into industry. They’ve been used in the DOD space. They’ve been using manufacturing, even scheduling airline traffic. So these are the types of problems that they’ve tackled traditionally.

Again, it was a discovery to me that in many modern tech companies, like DoorDash, Instacart, or Uber, in those data teams where you have a bunch of data scientists, you also have those operations research specialists that do those complex calculations and sit next to data sciencists.

Definitely, in a funny way I think they’ve taken on the helm of calling themselves data scientists also. So in a way they’re a very, very specialized niche community within data science.

But yes, and they’re starting to rebrand into this decision science.

We actually hire decision scientists within Nextmv. We feel that’s the space around how do you build, use and implement these types of systems.

To finish on decision science, the other discovery for me was how broadly horizontal a space it is. One typical example is all the routing problems and delivery and logistics and all those things. But that’s actually a small part of the very wide range of different use cases. Can you talk about some of those use cases across the enterprise?

There’s been some interesting ones recently. We were working on a project right now with some folks around how to basically route different x-rays to providers to give feedback. So like call center routing. So people don’t think about this as an optimization problem, but I thought this one was kind of interesting. If I had a horrible break and I’m looking for someone to read my x-ray, that was a problem that came up recently that we were working on.

We have another customer who’s doing humanitarian aid. How do you allocate different resources to provide aid in the fastest manner and to cover the most needs, the quickest. And so, there’s some different, interesting applications there.

I talked a little bit about the matching and allocation problem around inventory with subscription boxes. So there’s that case, but there’s also things around pricing, price optimization, marketplace matching. So how do I efficiently match supply and demand?

Really at the end of the day we think about this as very horizontal, this should be how you represent decisions for your operation as code. And that can be any operations decision, like how to allocate marketing spend into different channels, anything like that.

The fundamental premise of Nextmv is to democratize this whole area, which has been the province of math PhDs and older software platforms. How do you go about that? What’s the sort of the thinking and ethos behind the product and platform?

We think about making every engineer a decision engineer.

So in the same way that Twilio gave a bunch of engineers the primitives around how to create messaging, we’re giving people the primitives around how to make decisions. You and I should be able to sit here and define a new decision for our operation. Whether that’s allocating marketing budget, or that is creating a dispatch service. We can define what the input output is, how we think about caring about it and what the business rules are for possible plans.

Our platform enables all those steps. You can build a model from scratch, defining input and output. You can push it into deployment. So deploy it via something like serverless. Or into our cloud architecture. And then you can also define what you care about. So that’s the value function or fundamentally what is guiding your decision. So defining that is the KPI that you care about. So that’s how we think about going about building it and really that’s the, we think about it as an end to end, decision automation platform, which is model management, which is the workbench basically for building and creating these things.

Let’s double click on that and what are the different parts of the platform and what can you do. What kind of skills do you need to have as a developer to be able to use the platform?

The things that you can do and, I started with this, you can build any custom decision. So we think about decisions being plans. And so how do you generate the plan that is addressing your business need? As a developer, you are typically already going to be thinking about this in the context of your business problem. You’re going to say, “Okay, what is the input data that I can use to make this decision? What is required?” So in the inventory example, for Stitch Fix, I have to have all my potential inventory and I have to have scores for inventory that matches to the people that I already have subscribed. That would be your input data. You’re saying, “Hey, this is my contract. This is my data contract between the model and my services.”

And so you can define that, you can define your output contract. Your output contract is what you’re going to go operate on. What your system needs to go effectively, make that plan a reality. So when we were at GrubHub, that would be, what is the route for a driver so that I can send it to a driver application? Because they need to see it. And so those assignments, that sort of thing. You can do both of those things, you can also define what’s possible. So I touched on planning and building this space of possible plans. you can literally guide that. You can say, “I want to build plans by adding one delivery to one driver every single time or adding one of those items to a box and starting to build those boxes.”

It’s almost like a a state-based approach. You’re building these plans iteratively, and that’s what’s allowing you to to think about that space in a more business practical manner versus having to think about it in matrix math. Does that make sense?

You have a cloud product? Talk about what it does and how developers can get invovled with it.

We have our console. You can actually go to that today. You can either access it through our website on nextmv.io, or you can go directly to our cloud.nextmv.io. That is where you can basically sign up for a free account. You can start testing us. There you can see a demo model, which is just around routing. So you can play with that and get a sense for what it means to have that JSON in JSON out structure, what you can configure, you can configure different run profiles, stuff like that. And so we think about that cloud platform and that console as our workbench area, that’s where you go to configure a decision, that’s where you go to manage your account, et cetera.

What we’re really excited for is over the next few months we’ll be releasing our second generation of that console and tying that to our Nextmv Cloud, which will allow us to do more customization than we have today and bring the full power of our SDK into our cloud platform. So that would be for building any custom decision. And so if anyone’s interested in that, we have a waitlist going, so that wait list is also on our website, nextmv.io/waitlist.

We talked about some broad use cases, at a theoretical level for the space, but let’s talk about more concretely some of the customer use cases that you guys have experienced so far?

Yeah, we’ve had a couple interesting ones. We have one customer who is working with us on a bus scheduling algorithm. So thinking about how to efficiently schedule workers and that there’s basically different limitations around labor laws and all that stuff that go into that shift planning. So that is one interesting use case that we come across.

I mentioned humanitarian aid already. I just thought that one was fascinating. They were literally demoing a use case about delivering aid to Haiti during a hurricane. And I just thought that’s a real world problem that our software could have a major impact on. We also have some customers that are doing some interesting things around delivery ecosystems.

Another customer is rethinking the space around hospitality. They are taking kitchens that you would normally have at the hotel you’re staying at and centralizing them for a bunch of hotels and giving them an upgrade in terms of chef and quality and that sort of stuff and then delivering from there. So some really interesting things and all of these systems come with their own unique business rules and challenges and all of the operations are unique. I think that’s what drove them to use Nextmv is they want to be able to consider their uniqueness when they’re building their algorithm, instead of being tied into just configuration.

What’s next for the company in the next year or two? What’s on the roadmap? You alluded to some of this, but what’s on the roadmap and what do you want to be able to do?

We’ve recently launched in our cloud console the understanding of configuration. So being able to configure different models. What we’re excited about, and we’re already prototyping now is the ability to, like I said, create that custom decision and to push that up into our console environment to collaborate on that with other users and to really build on top of that. We’re excited about that for a few reasons. We’re really intrigued by what developers will create and what our users will create on top of this. We’ve been in this space, we’ve lived optimization simulation tech for a long time. There’s this academic mindset around that for problems that it can apply to.

But I think the cool part about a platform is you’re really building generative technology. So we’re excited about how people stitch these pieces together from the IO perspective to what their decision is.

Ryan, one of my co-founders built a Sudoku solver on top of our platform. I mean, that’s silly and not really business relevant, but it’s really interesting.

Can you do Wordle? [laughs]

I really want a Wordle solver, we’ve been talking about it. I would like to be able to solve Wordle in milliseconds, that would be fantastic.

So yeah, I think that’s what we get really excited about. I mentioned that next generation like console, cloud, clients at SDK, that really is where we get to open this up to a broader community in a freemium version and allow people to really be creative and start building on top of this. So that’s what’s coming next for us is starting to be a little bit more out there in the community, potentially running some events like that and hackathons and getting people creating. I think that’s the most exciting thing coming in the next year.

All right, very cool. So shifting gears a little bit, one question from the group, and then another question from me more about the company building aspect of the Nextmv story. So question, how hard was it to convince VCs that building a horizontal data platform was a good idea? Sounds like Matt was unaware of this field to begin with, and I found that VCs are quite skeptical of companies building horizontal from day zero.

Yeah, you should field this one [laughs]. No, I’m just kidding. No, I think it was challenging in the beginning and part of it was storytelling. And I think that’s actually a huge part of founding in general is being able to tell your story. There were a couple things playing in our favor. One, we had done this at GrubHub and I think that’s a really big part of it, right? Understanding very intimately the challenges that we faced building and scaling these systems at GrubHub, played nicely into that. And then the other aspect was just painting the picture. I mean, I think, and Matt, you can feel free to chime in here if you disagree, but you know, people are aware of these data trends.

And they’re also aware that still, I think there’s some absurd statistic that Forbes published, which is like 87% of data science models don’t make it into production. So there’s a pipeline issue there that is code based. So I think when you start talking about, “Hey, we still have a gap between where we are from a data and insights perspective and predictions perspective to where operations is within a company.” I think that really resonates with anyone that’s been in the operating space. And so that was ultimately, I think what got a lot of the folks that we have on our platform over the hump is that they’ve really seen that play out in both their portfolio companies and maybe in their past jobs as well.

Yeah, absolutely. I would concur with all of this and the reality is that you’ve also been honing on one sort of beachhead area initially around logistics and delivery and that type of stuff, following the good old principle that you need to be a tool before you become a platform and that you have to expand over time.

Exactly. And basically not boil the ocean. I mean, that was our tack really early on, is like, “Okay, we’re trying to create a new category here and really open this up to a whole new set of users.” So those users fundamentally need an education aspect to it. What is it and what can it do and how powerful is it? And so you need to start with something very concrete. That was why we started with supply chain logistics, routing assignment problems, scheduling problems, et cetera.

Then just one last question, because I think that’s so cool and interesting for all of us working in the world of startups. Nextmv has a policy of radical transparency when it comes to compensation, everybody knows what everybody else is making, what’s the story there? The thinking, the pros and cons?

Yeah, I think, when we started the company, we talked about this for a couple reasons. One, myself, as a female founder, I found it really important just to mitigate the wage gap. I think a large part of the reason the wage gap exists is that women and minorities in technology are not necessarily willing or confident to negotiate. I’ll throw myself out there as an example. I negotiated one salary in my entire career and it wasn’t even that much of a negotiation. It was just like, “I kind of maybe sort of want this salary.” And they said, “Okay.” So I think that’s a big reason. We wanted to take that away from the discussion and say, “This is how we value that work within Nextmv.”

And so everyone that is doing that job that we’re paying for the work product. And the work product is the same, regardless of if you live in Kansas City and you come in as a decision engineer one, or if you live in Germany and you come in as a decision engineer one. So we took that tack really early on and just said, “We pay the same for the same work product.” And so that was part of our ethos. We built a distributed team at GrubHub. So we saw this play out across a lot of different cities and we just wanted it to be clear and we wanted people to understand the risk profile too. So not only do we do transparent salaries, but we actually do a tiered salary thing where we offer people mid, high or low salaries that correspond to the inverse for equity.

So you can basically pick your risk profile. And I think, as someone who only worked at one startup before, I really didn’t understand what questions to ask when I first went into that space about things like runway and what am I making, what are other people making? What does this equity mean? We wanted to make it accessible for anybody coming into the startup ecosystem and work for us. And fundamentally that just drives more diversity and more creativity on our side.

Great, I love everything about this. So cool. That feels like a very nice place to end the conversation. But thanks so much for coming by and telling us about the Nextmv story and obviously excited to see what’s next. So thanks again.

Thank you, Matt.

Leave a Reply

Your email address will not be published.