Positioning a Generative AI Startup: Erik Severinghaus, Founder and CEO of Bloomfilter (Part 6)

Sramana Mitra: So coming back to what you just said on the how, let’s double click down one more level. Take me through examples of how you use generative AI to stitch together these disparate systems in the software development process.

Erik Severinghaus: When we take a customer, we connect into the systems that they’re already using. Let’s imagine customer uses JIRA to do project management and track the process flow of what’s happening. They use GitHub to store their code, and that’s got a lot of process metadata associated with it. They using a testing and integration system like CircleCI or Jenkins. They use Figma on the front end to do their designs. Elastium estimates that in a typical software development lifecycle, there’re fifteen different tools. Let’s just start with these four systems.

In order to understand how work is moving across those systems right now, there’s no real way to get an integrated view of how the design in Figma relates to this card in JIRA, or how this card in JIRA aligns with this code in GitHub, or how that ultimately gets deployed into something like a deployment management system.

If you try to map all of this together manually, you’re dealing with tens of thousands of tasks in each of these different systems. So, trying to map these things together and understand how the process is flowing becomes virtually impossible. This is where we use generative AI to understand what this written code relates to, or how that tracks to the project management, or how that ultimately then tracks to this design. It allows us to get an end-to-end process view of what’s happening within all these different systems. Does that make sense?

Sramana Mitra: Yes. And then what is actionable based on this analysis and this mapping?

Erik Severinghaus: So we find the most fascinating things. We can go to a customer and say, “Did you know that 40% of your tasks that are going into production are actually not being tested before they go into production? By the way, of that 40%, those tasks are five times as likely to have a production defect opened against them as the tasks that do go through a testing cycle.”

Or we can show them that if they don’t have the requirements in test cases for a certain design, then it’s three times more likely that that you’re going to have to redo that work.

Because it’s going to get into testing, it’s going to be rejected, we’re going to have to go and rework it. So once we can map the end-to-end process flow together, we can show people what’s happening within the process flow, and then we can call out where people aren’t following their own process.

Remember how I told you, 70% of stuff’s laid over budget or doesn’t ship at all. What happens is a lot of these companies are not following their own process. It would be like if you got on to an airplane and the pilot wasn’t going down his checklist before he took off because he was afraid the plane was going to run late. We wouldn’t want to do that because we know that it would increase the risk of error, it would increase the risk of a crash, and it would increase the risk of unacceptable outcomes.

That same dynamic holds in the world of software development. We have a lot of software development tasks that are running late, and so everybody cuts corners. They try to get there as fast as possible. They skip their process, and then it ends up making the entire program worse and it ends up with waste and rework.

Once we can map the process and people understand what’s happening, then you can start to flag where there’re exceptions, where there’re problems, where people aren’t following their process, and we can start to improve that process. So we can start to say, “Let’s take these ten tasks and have a meeting about it. Let’s understand why it’s not going through a testing phase before it goes into production.”

Sramana Mitra: How do you price this? Is it a SaaS model?

Erik Severinghaus: Yeah, it’s a consumption-based software as a service model. So, we’re looking at how many people are involved in the software development process. Then we’re looking at how many different tools are we going to integrate into in order to pull all this information out. So those two things basically become a consumption unit for us. Then we charge based on that consumption unit.

So typical pricing may start as little as $10K a year, depending on the size of the organization you’re talking to. It may scale up to a quarter of a million dollars or a half a million dollars a year, depending on the size and scale.

Sramana Mitra: Is there a way that you have figured out how to show ROI with this pricing model?

Erik Severinghaus: Yes, we have to. We have a pretty strict rule. We won’t give pricing without an ROI analysis. Sometimes we use industry standard numbers. If a customer will let us, what we’d love to do is profile their own system so we can connect into their systems. We can do an expected ROI analysis by evaluating how much waste they have or how much work is being redone.

So we’re typically targeting a ten X plus ROI in terms of the amount of waste that we can remove from the software development lifecycle.

This segment is part 6 in the series : Positioning a Generative AI Startup: Erik Severinghaus, Founder and CEO of Bloomfilter
1 2 3 4 5 6 7