Siqi Chen, CEO of Runway, on building browser-based collaborative FP&A

When COVID hit and all of our revenue went down to zero, Andreessen Horowitz was triaging all of their companies to decide which ones to save first. We presented how we were going to survive. We had 400 employees at the time and 8 digits of revenue, and it was all going to go down to zero.

We created a bunch of Google Sheet variations for 3, 6, 12, and 24 months of COVID. They were the only VC firm on the planet who thought that COVID was going to last longer than 6 months; they said 2 years. Two years of zero revenue was not going to work for us, so we had to lay off 95% of the company. I laid myself off and the entire leadership team. The founder Steve hibernated the company for about a year and a half, though now it’s growing quickly and profitable with a thousand employees.

Thirty minutes after I laid myself off and most of the company, I talked to our now ex CFO and asked if there was something better we could have used for all that work we were doing on spreadsheets. He said no, this is just what you use. I asked about enterprise tools, but he said no one likes using those. I suggested maybe someone should build something good here, and he said he thought I’d be a good person to do it.

I talked to the Andreessen team, and they said they'd been looking for an investment in this space. The problem was that everyone who came to them had a finance background but not an actual consumer product background. They thought that for a really disruptive company to exist in this space, it needed someone who understands the pain, has operated in it, and is actually a good product and design person who wants to work on financial software. That's a really small Venn diagram of people, and I raised my hand and said that's exactly what I want to do. That's what led to Runway.

What's the substance of the complaints about Anaplan, Oracle's FP&A product, and similar tools? What do you hear from those people who do the real work?

Very similar to you, I've talked to probably close to 100 people who have used these kinds of enterprise products before. Every single one of them says they still do their modeling and forecasting on a spreadsheet and copy-paste the results into this platform. None of them seem to like it.

The one exception was when I talked to a Fortune 50 CFO. He said he had a really great experience with Anaplan and gave credit to the person who implemented it. He offered to introduce me to this person. I said, "That's amazing, yes, I would like to talk to him." I got on the phone with him and said, "You guys have done a successful implementation. It sounds like it was really good. I would love to learn how you did that. It sounds really unique." He replied, "What are you talking about? It was awful." I thought I finally found an exception, and it turns out it wasn’t.

So why is that the case?

I think it’s worth first talking about what people do enjoy out of these products because you don't pay that much money for software without it doing something. If you think about the priorities of a finance team, the number one thing is to have an accurate forecast, and it is extremely time-consuming to create that forecast. Instead of having to pull data from a bunch of manual places, it connects to a lot of data sources and automates that. So actually, the model does legitimately become a lot faster to update, and you save a bunch of time that way.

The other thing is you want to make your workflows faster. One of the primary values you get after you get your model built out on a finance platform is streamlining your budgeting process. If you don't use a tool and you have spreadsheets, you literally will email empty spreadsheet templates or budget templates to your different department heads, have them fill them out, and send them back to you.

If they don't, because this is paperwork no one wants to do, you ping them and bug them until they do. Maybe you get it back and it's haphazardly inputted because who cares? They don't even know what it's for. It's totally for you; it's just extra work that they have to do that has no impact on their performance reviews. So who cares? And then you paste it back into your platform or into your own spreadsheets.

You can make that faster though if you had a platform. You can create different pages and people can fill out the form. You can track the workflows and status. The system will remind people, and then you can merge it very quickly. So you save time that way.

And it turns out that if you just do those two things - automating data integrations and making the budgeting process faster, it's pretty valuable. You have extremely highly paid executives, and if you can save time and make their lives easier once a quarter, people are willing to pay hundreds of thousands of dollars for that.

However, I think you hit the nail on the head, which is: why is it still the case that not just the people outside of the finance team, but the product teams, the sales teams, and the engineering teams still have their own models to forecast the future? Why is finance still doing that? Our answer is that it’s because you can be so successful in building a workflow and automation tool, you don't actually need to build the thing that solves the other problem of modeling and planning and being the tool of choice for business thinking.

The main complaint with these incumbent platforms, and I would argue also other platforms that exist today, even the newer generation platforms, is that they are not designed to be tools for thought. Tools for thought have to be fast and flexible.

I'm going to go a little bit philosophical here. I think the story of information technology in general is a story of new tools for thought. That is what writing is, and what languages are.

Language allows you to offload thoughts from your own head and communicate externally to other people. With the invention of language, you can coordinate tribes, and eventually language evolved into writing. Writing allows you to coordinate groups of people outside of earshot, and that's what enabled civilizations and nations. The telegraph enabled our ability to coordinate people across vast distances.

A spreadsheet is very similar to those forms of technologies. It lets you offload very complicated relationships and thoughts and data into an external medium, allowing you to communicate and manipulate them faster than what you can do in your head.

That's why it's powerful. If you want to build something that people actually use to model and plan, it has to be a tool for thought so good that people reach for it by default, even more than Excel, so that it accurately represents how people actually think about their department, how a product team operates, how growth operates, how marketing operates. It has to let you interact with it in ways faster than what you can do in your head. That is a very high bar for design and engineering execution. If you don't have the culture or expertise to build that, you don't build that product.

Every other product that exists is a much simpler product, which is: here's a form, fill it out, and I'll track the status of it, and we can merge it back. Where you actually are doing your thinking, I don’t even care. Do it on your own spreadsheet. And that's how you end up with 40 fragmented spreadsheets that don't talk to each other.

In short, how would you describe Runway and the approach you took here?

Our approach is that the things existing FP&A platforms do are table stakes - they're necessary, but insufficient.

The baseline theory of how you address the pains and priorities of a finance team is that forecasts have to be accurate, they have to collaborate with other departments, and report out accurate numbers. The way we're going to solve this problem is by saving time through automation and improved workflows, as well as version control. All of that is table stakes, but it's not enough for finance to have the strategic impact that's expected of them.

For strategic impact to actually happen, collaboration needs to occur. For collaboration to happen, people outside of finance need to understand how the business works, the context of the business, and how it all fits together. Imagine if your entire model lives in either a locked-down platform that you don't have access to, or even a spreadsheet.

There’s important business context inside that spreadsheet. Even if you're an amazingly progressive company and decide, "I want everyone to collaborate on this, so I'm going to share out the spreadsheet and give everyone access to it," you can't do that. Why? Because all calculations on Google Sheets or Excel are on the frontend, so there's no way for you to lock down or encrypt a column called "Salaries." It's literally not designed for that use case - it is physically impossible for you to do that securely.

But even if you could share it securely, it doesn't matter if people don't understand what they're looking at. If you share a spreadsheet and people don't understand how the model works, collaboration won't happen and strategic impact won't occur. We have a joke about models that we say in every sales call to finance people, and we get a laugh every time: "The only person who understands the model is you."

So what we're saying is that saving time is not enough. What is actually required is something that delivers the clarity of understanding that a finance person has about a business to the rest of the company. The way we deliver that clarity is by building a better tool for thought, so that they can offload their own intentions and mental models about how the business works into one place. Then you can connect it all, and now you have a single pane of glass view of your business.

The finance and FP&A space has seen an explosion of new startups in recent years, like Digits, Equals, Mosaic, Causal and others. What is the key thing driving so much founder interest in this space?

I think, looking at my personal experience and the things that inspire me, we now have experienced founders with higher expectations for user experience. Many large enterprise companies were founded before the iPhone existed, which really changed how people thought about the value user experience could deliver and what was possible to build. People had a hard time even imagining a phone that felt that good to use before the iPhone.

That's one component: you have founders who have operated before, seen this problem, experienced the pain and frustration, and recognize it as a valuable problem to solve. It's usually a second-time founder or a pretty experienced operator building these companies. If you're a first-time founder, you're going to work on either a dating app or a travel thing, as those are the things you care about and are familiar with.

The other component is exactly what you mentioned. Products like Airtable, while not a direct inspiration, expanded what I thought was possible in terms of flexibility, power, and accessibility. We're far more inspired by Notion and Figma in particular. Those products have been very thoughtful about building powerful abstractions.

My favorite abstraction in products is components in Figma. If you're not familiar with components in Figma, they let you abstract out the idea of a button. You design a button once, and if you want to make buttons in all of your designs curved or purple instead of red, you can just change the master component, and it updates everywhere. This is much easier than in Photoshop, where you might have 20 screens and have to change the button manually each time.

What's really interesting and powerful about this abstraction is that it exposes the power of software engineering principles like polymorphism and inheritance almost invisibly to an end user. When you choose the right abstractions, it can better contain and offload intent in people's heads. With a 40-screen app, you don't think of it as 40 screens with 100 individual buttons; you have an idea of what a button should be in this app, and now you have an abstraction that maps closer to what's in your head.

It gives you the power to manipulate it the way you would want to in your head, but externally. In your head, you think, "I want to make all the buttons purple," and now you can just make all the buttons purple, not 400 individual buttons. The power of the right abstraction allows you to think better, and these are the kinds of things necessary if you want to create a really good tool for thought, which is what we think has been missing in this space.

For a head of engineering, what's the draw that brings them into a tool like Runway, versus it being a siloed finance tool as FP&A traditionally is?

There are two entry points. The first one is pretty standard—it's a standard budgeting experience. For example, I need to update our annual plan and reconcile the headcount requirements and marketing expenses between different departments. You can create pages and say, "Tell me what your plan is. Can you model this out?"

As people edit these pages, they're in personal drafts, which will easily become scenarios. These scenarios are then submitted as essentially a pull request to the finance team. So that's the direct way people get pulled in.

The second way people get pulled in doesn't start with finance at all. Maybe you just want to create a good sales forecast or a good product model directly, and you want to get there sooner because you want to start with real data.

We integrate with just about everything under the sun, so now you can pull in your Hubspot or Salesforce or your Amplitude, Snowflake, whatever, and shape it into a form that's ready for forecasting. Because we have robust forecast modeling, as most people know it, it's easier to maintain and understand how the model works. That's the other method by which people get pulled in.

The idea, which is reflected in the weird situation that we've become almost blind to, is that we think modeling in these spreadsheets and planning is solely the domain of the finance team. But in actual reality, every department has a spreadsheet that models the future. The fact that this is happening and these models are not connected with the source of truth for where the company is going is directly impacting the priority pain point of a finance team: you want to have accurate forecasts.

If the model of the forecast doesn't reflect how the people running different parts of the company actually think about how their parts of the business work—how the growth of a game works, how marketing campaigns get reflected and become leads, how leads get converted to revenue—if that's not actually accurate, then your overall business forecast isn't going to be accurate. And it isn't accurate today, because the real source of truth is that spreadsheet that the head of sales or marketing or engineering maintains themselves. It's what they use to actually run things.

We believe this state of the world exists because the tools do not serve that function today. If you can just build that tool, then you can change that and have an accurate forecast that actually represents how the different parts of the company work.

Can you talk a little bit about the core abstractions that you've built for Runway and how they rethink this process?

I'll start with two concepts that people are generally familiar with: dimensions and drivers. For those who aren't familiar with modeling tools, Airtable is a fantastic example. The core abstraction in Excel is a cell, and you have cell references and sheets. What Airtable has done, for example, is extract the database use case from Excel by making the row the primary unit.

Before Airtable, there were products like Lotus Improv and Quantrix, and before that IBM TM1. These tools allowed you to model things with drivers and dimensions. Instead of having a cell that says "cash" and then calculating "cash this month minus burn last month" and forecasting it out, you could just have cash as a row.

The relationship would be that every cell is just going to be cash this month minus cash last month minus expenses this month. You can reference other rows, not just cells.

So we have that functionality too. It's not a new invention.

Dimensions allow you to segment your model. If you have revenue or marketing spend, you can break it out by channels, SKUs, and other factors. This becomes really useful especially when you have complex businesses. If you want to determine the marketing spend for the enterprise segment in California for a particular product, you can do that through dimensionality. Then you can sum it up without having to maintain a million different formulas.

One of our main innovations is the idea of plans. When we think about our first principles of how we can make this clear and understandable to the people that finance teams work with, we ask ourselves: How do people outside of finance teams think about the business? How do people inside the finance team think about their work?

One of our insights was that they probably don't think about their work in the form of a wall of numbers. You can't put that into your head. Instead, you think about it in terms of things you will do in the future to impact your business. That's what a product roadmap is. You have engineers working on a product because they increase its value, leading people to pay more money for it or increase conversion rates, retention, engagement, or growth. Otherwise, you wouldn't do it.

The same applies to marketing plans. It's something that happens in the future: you're going to spend more money, which will impact cash negatively, but you're going to get more leads, which will impact future cash in some positive way.

We call these things roadmaps, marketing plans, and fundraising plans. All of this context lives completely outside of the spreadsheet, disconnected in places like Notion and Google Docs.

We thought if we could find an abstraction that can connect context with your models for the first time, it would make a much better tool for thought. Now, when you look at your model, you can see the plans that are already in your head and visually see their impact on your business. You can immediately see how they affect margin and growth two years down the line.

Because you have new abstractions, you can have more powerful interactions. You can easily adjust conversion rates, pricing levels, marketing investment, or headcount plans. You can quickly move the timing of your plans earlier or later because they exist as abstractions.

If you think about the physical action of changing the timing of a series of plans on a spreadsheet, it's incredibly annoying. You have to find the delta of your previous plans, paste the deltas forward, and then paste all formulas back, doing it for every row impacted by that plan's timing. With a better abstraction, you can literally just move it on the timeline.

These are a few examples of the kinds of thinking that we believe will lead to a better tool for thought, and it's why our customers are quite excited about our approach.

Do you see an opportunity for Runway to move into spend management by offering virtual cards tied to budgets that can automatically reconcile in your FP&A product? More broadly, how do you think about expanding into financial services like lending by leveraging your visibility into a company's finances?

Just like Brex and Ramp are really focused, we try to do so too, and the value that is still uncaptured today is being able to correlate and make useful all the disparate sources of data and see how they fit together.

Whether it's Brex or Ramp or NetSuite or Snowflake or Jira or Linear, we want to be a place where you can see how all of those are connected.

It will be very interesting for us, for example - and you could basically do this today - you could integrate with Linear, get your tickets into plans in Runway, and then you can say this particular project or marketing plan is going to cost this much money in credit card expenses. We're going to get this many more leads in our CRM, and then as time passes, you can actually see if that's what happened.

You can categorize your expenses correctly, you can attribute leads to certain features or certain channels or certain marketing campaigns, and now you can do an A/B test against that.

I think that is the way in which we're going to focus on adding value. I think there are enough credit cards; who needs another?

Runway has 700+ integrations into payroll tools, HRIS, CRM tools, etc. Can you talk about how you got so many integrations created and the value in covering the long tail of integrations vs. concentrating on the top 15-20 tools?

We work with multiple integration partners and also develop our own integrations. Different partners work better or worse depending on the service, so it's a pretty wide range. Integrations are super important to what we do for two key reasons. First, the most important value we deliver for our customers is getting their data in without them having to constantly update their model by hand, which is incredibly time-consuming.

Second, so much value lies outside of finance in terms of data. It's about understanding the interaction between your HR hiring plans, your CRM pipeline, your product roadmap, and financial outcomes. That's why we have to be pretty broad and robust in who we support.

How do you think about LLMs as a foundational technology (or not) for Runway and what does it enable you to build that couldn't exist before? Can you talk about your vision for AI in FP&A and what that might look like inside of Runway, particularly as we get closer to AI agents that can reflect and get better at taking action?

So we have this page called Ambient Intelligence that describes how we think about AI at Runway. I believe the two biggest opportunities in AI in general are: 1) context and 2) proactiveness. If you use ChatGPT, it doesn't know much about you, your business, or your intentions, and it doesn't do anything unless you talk to it.

If you reverse that approach, there's a lot of creative potential. Imagine if it knows all about your business, your intentions, your plans, how it works, who the people in it are, and what they do. Then it's able to help you think while you're working, instead of waiting for you to ask it a question in a chat interface.

A related point is the design expression of AI in your product. The default design expression for AI that most people are familiar with is basically treating it as a creature outside of you - an independent entity or agent. When you talk with ChatGPT, you're talking to this other thing. But when you think about the actual economically productive use of AI today, it's probably something like GitHub Copilot.

That interface is very different - it's not this external thing, it's helping you think as you're working, almost in the background. It's not in your face; there's no sparkly emoji. It's just doing things to speed you along as you’re working in your natural workflows, and we think that is what a good expression of AI should look like for the current generation of language models.

When we connect that idea to what we're building that's different, specifically, we believe we're building a tool for thought that enables understanding for the entire company. The use of AI in that product should be an AI that helps you think, not one that thinks for you.

For example, one of the things that people don't even notice when they use Runway is that when you create your model and examine a driver, you can trace through the calculations and there's a plain english description for that driver. When you open it up, it's just a description that explains what this driver is - that's actually AI-generated, but it doesn’t seem that way. We don't have a chat thing or a sparkle thing; it's just plain text. That’s a subtle but illustrative example of what our AI philosophy is at Runway.

People kind of use AI as a marketing term, but it's as mundane as saying "Salesforce: powered by JavaScript." AI is just another tool in your belt to build useful products that solve problems; it's what everything will be made out of. It should be embedded in every pixel and every piece of text that you have, and it should be proactively helping you think.

So that's the opportunity, that's how we think about AI philosophically. The specific opportunity with Runway is that context is important, as I mentioned. If you combine that design expression and proactiveness with access to more context than probably any other product you can imagine in a company, I think that positions us well to build something really valuable and useful to customers.

If everything goes right for Runway over the next 5 years, what does the company look like and how is the world different?

To expand on a previous analogy, I think about Figma and Amplitude a lot. I was an early investor in Amplitude as well. One of the things I’ve realized is that when a product is very impactful and useful, it changes the culture of the team that adopts it. We tend to underrate just how big of an impact products can have on behavior and culture.

Before tools like Amplitude and Mixpanel, data was very sensitive. If you needed access, you had to go through a data team and an analyst who would author the right query to give you that report.

Now, we take it for granted that a PM, or a designer can just go in and understand how many users you have, what retention looks like, and so on. That was not the case before, and we don't really talk about that. It's because tools impact culture tremendously.

You mentioned Figma—PMs are in there now. People forget that, not very long ago, it was considered rude to even look over the shoulder of a designer while they were working. But now it's just okay that everyone's in Figma. That's actually quite remarkable.

If we are good at our jobs and execute on our opportunity well, the impact that I hope to have is that everyone, by default, understands how their business works. They should know why money is spent, for what purpose, what you get for it, and how to make it better. That is not the norm in companies today, but it can be.

This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.