Liberty Mutual, one of the largest insurance companies in the U.S., set off on a digital transformation journey in 2014. Fast forward to present day and the company has since started vacating data centers for the public cloud, and in last four years adopted a serverless architecture based on Amazon Web Services' (AWS) Lambda platform.

Dave Anderson, director of technology at Liberty Mutual Insurance IT (virtually) sat down with SDxCentral to look back on the journey to serverless computing and discuss how going serverless introduced new perspectives on how IT teams at Liberty Mutual work.

Take me to the starting line: How did the Liberty Mutual team discover serverless? At what point was someone like “hey, maybe this is a technology that we should look into.” 

Anderson: One of my team was actually at AWS re:Invent in 2014 when Lambda was launched. They came back and were just blown away because we had already been on this journey of trying to figure out those principles. The idea of Lambda, and what was early functions-as-a-service (FaaS), just sat really well. 

Was that more or less Liberty Mutuals serverless adoption strategy? How has the platform changed or evolved since you initially began using it?

Anderson: We had already been on a microservices journey before that so we were already used to breaking things down into smaller components. 

When you start to get used to the cloud service environment, you realize you have to be event driven and that's a very different approach. So teams naturally start building smaller functions, start thinking about contracts between functions, etc. And then you start thinking about observability and how we can actually watch the system in production.

There's a whole different level of engineering that you come to when you start working in a serverless environment.

And what processes are you currently using a serverless platform for?

Anderson: It started off relatively conservative. We would do things like generate a back-end downstream process or maybe an integration. We did a few early things like that to get used to the technology.

Then we started doing a lot of serverless integrations to orchestrate between two different systems. We wrote quite a large financial system that performed a serverless extract transform and load (ETL). The system processes millions of financial transactions monthly. It's a perfect use case for serverless because that type of system is dormant most of the month and then there’s a big batch run, so it just scales up, runs a huge amount of throughput, then turns off. And there's minimal cost for about 28 days a month.

Liberty Mutual has now had a serverless architecture deployed for more than a year, what changes would you like to see from the platform? What problem would you like to see it tackle next? 

Anderson: The big challenge is probably distributed tracing. If you have a hybrid architecture with serverless and maybe other types of components, there's still quite a challenge in trying to trace traffic through lots of different systems.

Cloud observability is a very difficult thing to get right, and that's always being worked on. I think there’s a constant focus there.

What have been some of the biggest challenges in moving forward to expand the use?

Anderson: I think just the change in mindset was a huge challenge. Taking engineering teams who are used to writing applications and saying “okay now I need you to design a system,” that's a very different challenge. 

What was your antidote to that? Was it a culture change within the engineering teams to try and get that perspective or mind shift?

Anderson: Yeah, completely. Code is a liability. If an engineering team writes a million lines of code, that's usually not a good thing. That's a lot of risk and liability. We focused on tackling the mindset of pride in the amount of code engineers wrote.

The second is the idea of evolutionary architecture and creating a fast feedback cycle. If you create a one-way door in the architecture, you're basically blocking off a future solution. You have to create two-way doors; if you make a decision make sure you can come back from it if you need to. So trying to avoid making decisions that can't be reversed. Encouraging teams to move in an evolutionary way was very different to traditional teams with big, upfront plans of yesteryear.

What recommendations would you have for vendors that are providing serverless platforms to better serve customers?

Anderson: It's got to be ease of use. It's got to be removing friction. You've effectively got two types of customers for serverless: There are the engineers who just need to build something, which requires a low friction journey. The second customer is really sort of the business stakeholder or the actual end user. They just want business functionality fast. I've had conversations with my business peers and they're like, “I don't care if it's Lambda or whatever, I just need to solve this problem quickly." Reducing the complexity to reach the business problem, for me, is a key message.

Vendors that overcomplicate things don’t help. It’s not a simplicity message, it's a clarity of message. I think sometimes vendors try to confuse the audience with acronyms and different terminology and it doesn't help.

For enterprises looking to adopt a more modern infrastructure or perhaps even rounding out their digital transformation, what recommendations or advice would you have for those enterprises looking to implement serverless platforms into their operations pipeline?

Anderson: Start with the customer. Start with the user. Start at the top and work your way back into large areas. If you start at the bottom, it's a much longer journey. Compute is a commodity. The business stakeholders don't care if computers are running in a data center, in an image, or in a serverless Lambda. Look at the value chain from your customer and improve the overall experiece.

Too often the same companies will try to follow a framework or a process and not really sit down and think about its value proposition. I think there's a specific mindset where you ask “what is our differentiator?” “what's our value proposition to our business users or customers?” and focus on that. This is more than someone else's data center. This is a whole new paradigm.