Why Companies Are Moving to Serverless Cloud Technology

Thinking about going serverless?

Serverless architecture has become the go-to technology for companies like Coca-Cola, Netflix, and Airbnb. Why? Because you can hand-off server management and quickly scale to millions of requests in seconds, all without blowing your budget.

If you’re toying with the idea of building a serverless app or thinking about migrating an existing project, you’ve no doubt doing your research and that’s why you’re here. (If you are new to serverless, check out: What is Serverless? The 60 second answer.)

Reading tutorials and technical documentation can help you get up to speed on what it’s like to develop on. But to really understand how it works from a business perspective, you need to get the behind-the-scenes lowdown from companies that have already done the hard yards.

In this post, we’ll take a look at how businesses big and small have built their production workloads with real users on serverless architecture. We’ll dig into the challenges they were experiencing prior to going serverless, how making the switch solved many of their problems, and how serverless has helped their business.

Why Companies are Moving to Serverless Technology

There are four overarching benefits of serverless architecture when compared to other hosting solution options: cost, server management, scalability, and faster time to market.

Read on to learn how businesses have benefited from using serverless technology in each of these categories.


1. Heavywater

When Heavywater’s AWS bill started pushing $30,000 just three months after launch, the FinTech company knew it had an expensive problem on its hands.

The company pioneering research into machine learning and artificial intelligence has helped the mortgage industry outsource processing. Basically, the company has trained its platform to read and comprehend business content to mimic human thinking, building expertise over time.

Behind the scenes, this involves processing a ton of batch files, and originally the company’s orchestration architecture was built using SWF and EC2 instances on AWS.

This approach made sense when the product was first launched, but the development team quickly cottoned on to its drawbacks. For a start, the batch processing jobs controlled by SWF were being executed and monitored 24/7—relying on the same EC2 instances used by the company’s microservices.

In just four months, Heavywater’s monthly AWS bill grew from $10,000 to a stagging $30,000 with over 1,000 EC2 instances running.

Cost wasn’t the only issue—the development team found throughput was also a problem, with an average processing rate of only 4,000 files every 24 hours. To make matters words, SWF was failing inexplicably.

At first, as software engineer Mohsiur Rahman explains, the team thought code was at the heart of their troubles. But after further investigation by AWS, they were encouraged to consider moving to AWS Lambda.

After a massive overhaul of the company’s infrastructure, starting with converting all of its microservices to functions and designing a new infrastructure that used asynchronous functions, Heavywater settled into using serverless architecture.

The results were immediate. Costs plummeted from $30,000 to just $4,000 in just three months. Human resources devoted to batch processing dropped from 24 hours to 16 hours, and the number of EC2 instances decreased to just 221.


It’s not unusually for companies—and in particular agencies—to release free products out of the goodness of their hearts. But for New York-based agency Postlight, their “free” Readability Parser API was costing them $10,000 a month.

The company is best known for its Mercury AMP Converter, a tool that makes any website Google AMP-ready with one line of code, and its Mercury Reader Chrome extension, which removes ads from articles. Both products used the Readability Parser API.

Tasked with rewriting the API, Director of Engineering Adam Pash writes that he was faced with a couple of tough challenges:

  • The API was old and had become “brittle” over the years. Parsed results were stored in a database that had grown to store a “massive slice of the internet.” It had become next to impossible to perform even slightly complex queries, which meant the company didn’t really know what was happening with the API. He needed to produce a functionally equivalent library that would return the same or better results than the original.
  • He needed to reduce the $10,000 monthly cost.

The company settled on JavaScript for the rewrite and made one other important decision: it would switch to serverless.

After releasing the updated Mercury Web Parser API to replace the original API, Postlight’s cost dropped to just $370 a month. Pash highlights that the company’s previous costs had included database expenses, which they chose to forego, instead opting for short-term caching.

The new API now supports thousands of developers who use Mercury every day to extract structured content from any article on the web. All up, that’s 39 million requests every month, costing Postlight just $10 for every million requests.


Have you ever thought about the backend technology that powers Coca-Cola’s vending machines?

At AWS Re-invent, Michael Connor, the person in charge of Coca-Cola’s cloud migration strategy, talked about the tools and strategies the company used to develop its next generation of marketing applications on serverless architecture.

Vending machines around the world use an integrated communications system that tells Coca-Cola headquarters whenever a machine is running low on a particular beverage or needs servicing. The company’s marketing teams also use the system to create campaigns to attract more purchases, like “buy one get one free.”

With the company’s oldest machines operating on IaaS and pushing 12 years-of-age, Coca-Cola found its AWS bills were getting expensive—six EC2 t2.medium machines were costing $12,864 a year to operate. This included automation, elastic load balancing, management.

After switching to serverless technology, Coca-Cola’s costs dropped to $4,490 a year. This accounted for around 30 million requests.

The logic behind how its vending machines work using serverless technology is straightforward: When a customer buys a drink, the vending machine calls the payment gateway to verify the purchase, which in turn makes a REST API call to the AWS API Gateway, and then triggers a Lambda function.

The Lambda function handles all the business logic behind the transaction. If the customer uses a mobile device to make their purchase, a fifth step is added—a push notification to their phone submitting the information to Android Pay or Apple Pay.

Server Management

The Seattle Times

With 7 million unique visitors a month, The Seattle Times website needs to process a lot of images for different devices, especially when breaking news hits.

After maintaining on-premises hardware and custom publishing software for nearly 20 years, the newspaper’s software engineering team decided it was time to upgrade to a more modern content management platform and outsource server management—and find a faster way to resize images.

Initially, the team switched to a managed solution but ran into issues around scalability. Going back to the drawing board, the team considered alternative hosting options, including returning to self-hosting, most flexible managed hosting options, and various cloud providers.

They eventually settled on AWS, which immediately solved their scalability issues. The team also implemented a serverless function using AWS Lambda to automatically resize images on different devices, such as desktop, tablets, and smartphones.

When an image is uploaded to AWS S3, the serverless function is triggered, which then resizes the image.

News images can now be rapidly resized for different viewing environments, allowing breaking news stories to reach readers faster. Previously, if an image needed to be resized in 10 different sizes it would happen serially. Now, all 10 images can be created at the same time, speeding up image delivery without the need for server maintenance.


With more than 130,000 photographers on its books, PhotoVogue is the place to be for aspiring photographers hoping to get one of their images into Vogue. But for Marco Viganò, Head of Digital Development at Condé Nast Italia, the online photography platform presents a scalability challenge.

Launched in 2011 and part of Vogue Italia, PhotoVogue allows upcoming photographers to showcase their work. Each picture that’s submitted is carefully reviewed by Vogue Italia’s editorial staff, ensuring only the highest quality images appear online.

With the number of photos being submitted and published growing by the day—there’s currently a collection of more 400,000 photos, each of which can be up to 50 MB in size—Viganò found the company’s existing infrastructure was restricting growth. His team couldn’t provision resources quickly enough.

After migrating to AWS, the team implemented a serverless function using Lambda. Now, when a photographer uploads an image to Amazon S3, the function is triggered, which automatically converts the upload photo to various digital formats—such as GIF, JPEG< PNG, and TIFF—allowing the images to be edited by PhotoVogue staff.

Viganò says migrating to serverless architecture not only allowed the PhotoVogue team to hand-off server management so it could spend more time developing new services, but sped up the user experience by up to 90%, both for photographers uploading images and the editorial team processing them.


With 80 million readers every month—the largest audience of any female-focused digital media company in the world—Bustle.com gets a heck of a lot of viral, spiky traffic.

While massive, unpredictable loads might send some engineering teams into a panic, the developers at Bustle don’t think twice about their infrastructure—their serverless site scales infinitely.

The Brooklyn-based company’s site originally ran on a third-party PaaS, which involved a lot of server management, automation, and monitoring. It also raised the barrier of entry for new engineers to roll out new code changes.

Moreover, the engineering team wanted to focus on rolling out new features for users rather than wrangling infrastructure.

So the team migrated the Bustle site to AWS. Initially, they started using serverless architecture to process high volumes of site metric data from Amazon Kinesis Data Streams in real-time. This allowed the team to get data more quickly so they could understand how new site features affected usage, allowing for better data-driven decisions.

Then they decided to explore running an entirely serverless website. First, they built Romper.com using Ember.js and Riot.js running on a serverless backend. Then they built the Bustle iOS app on serverless. Eventually, Bustle.com was migrated to a serverless backend.

To help developers easily to integration tests and deployments when they were ready to release their code into production, the team built their own serverless software-delivery tool.

Now with serverless, the engineering team puts zero thought into scaling applications—no one has to deal with infrastructure management. The team can remain small with only half the people normally required to build and operate a site of Bustle’s scale.

But most important, Bustle engineers can now focus on building out new front-end features rather than dealing with backend operations.


Dragon Quest X is a hugely popular MMORPG (Massively Multiplayer Online Role-Playing Game) with hundreds of thousands of players from around the world.

The Japanese company behind it, Square Enix, designed solid infrastructure to provide a good gameplay experience at all times. But one popular feature—in-game screenshots—was proving a scalability challenge.

When players took a screenshot, the image was sent to a server where processing, such as thumbnail creation and the addition of a copywriting watermark, took place. But the processing was resource intensive and there were spikes in server load.

Usually, 200-300 images per minute would be received for processing every minute. But during holidays and New Year’s Eve in-game events, this number would climb to 6,000 images per minute, taking up to four hours for each image to process.

The company turned to serverless architecture to speed things up. Now, when a screenshot is taken, it’s uploaded to Amazon S3, which in turn triggers an AWS Lambda serverless function for image processing. After this, queues containing the binary image data output are imported into the on-premises servers to save the processed image data.

With serverless, the time it now takes to process an image has been dramatically reduced. At the first New Year’s Eve events after the migration, more than 6,000 screenshots were taken and submitted for processing every minute. What previously took hours to finish was completed in just over 10 seconds.

Engineers are Square Enix expect that the in-game screenshot feature will cause even larger spikes over time as player numbers grow. But with serverless technology, they can scale up without limit.


San Francisco-based life science software company Benchling is helping scientists unlock the mysteries behind CRISPR, a breakthrough technique used to modify parts of a genome with extreme precision.

With thousands of customers using its platform, including researchers from biotech, pharmaceutical firms, academics, and governments labs using CRISPR to build disease models and screen for drug targets, it’s important Benchling’s platform is fast.

Benchling allows scientists to quickly identify possible accidental matches to select the best candidate sequences for their CRISPR experiments. As Vineet Gopal, an engineering manager at Benchling explains, his team wanted to speed up searches across hundreds of genomes so researchers could design better experiments with less effort.

The engineering team was finding it difficult to support its growing user base and genomes with its old IT environment, which used several servers to process CRISPR search tasks, with each server storing the genomes on disk. Each search was taking about 30 seconds and the team wanted to to get that number down to just a few seconds.

Gopal says the platform was already supporting hundreds of thousands of CRISPR searches each month, but Benchling couldn’t spin up new servers fast enough when demand was high.

The engineering team moved the platform to serverless and split up CRISPR searches across several serverless tasks to reduce costs and boost scalability.

Now, whenever a researcher conducts a CRISPR search on a specific genome, Benchling’s web server splits up the genome into smaller tasks and invokes an AWS Lambda serverless function for each task. Lambda then downloads the genome data stored in Amazon S3, performs the query, combines and results, and returns them to the researcher.

Using serverless, CRISPR searches are now parallelized, solving Benchling’s scaling challenge and eliminating the need to maintain several servers to perform searches.

Searches are now 90% faster and scaled to hundreds of genomes, allowing scientists to spend less time searching and most time focusing on science. Gopal says the industry standard for these types of searches is “minutes to hours.” Benchling has been able to able to improve on this dramatically, setting new industry benchmarks.

Faster Time to Market

A Cloud Guru

A Cloud Guru “teaches the world to cloud” with more than 600,000 students learning AWS, Azure, and other popular cloud technologies.

But the online learning platform doesn’t just talk the talk, it walks the walk—its website backend runs completely on serverless cloud architecture.

Brothers Ryan and Sam Kroonenburg launched A Cloud Guru after growing frustrated by the lack of accessible and affordable cloud training available for software engineers. After a cloud training course Ryan initially submitted to another website quickly grew to 8,000 students, the brothers set out to build an online cloud computing school.

While on holiday with his family, Sam locked himself away in a bedroom and built A Cloud Guru in just 30 days, including an online learning platform, discussion forum, and payment processing.

Using an entirely serverless system, and without having to build a backend, the brothers got to market quickly within just four weeks of starting the build. They were able to get users on the platform quickly, which helped inform early development and, ultimately, helped the pair build a better product.

Serverless technology’s insane speed to market helped fuel the company’s growth. The team uses a mix of different cloud services, including AWS, Auth0, and Intercom, and pays monthly fees.

This model has enabled A Cloud Guru to deliver training with a disruptively low-cost pricing model and attract a large user base quickly, helping to fuel the platform’s fast growth.

In just six months, the company grew to 40,000 users. Today, the company has more than 600,000 members—all without a single server.

As Sam explains, since the company doesn’t run servers or infrastructure, they don’t have to employ people to maintain and monitor them. There are no servers to be patched, no auto-scaling groups to manage, and no compute performance metrics to worry about.


Serverless architecture, as we’ve explored in this article, has helped many companies and big brands push applications to market faster, while at the same time solving complex issues around scalability and server management.

Cost has also played a big role in serverless technology’s industry-wide adoption with sub-second metering enabling companies to keep their costs low and, ultimately, deliver greater value to users.

Without the distraction of infrastructure management or concerns about scaling, serverless allows developers and businesses to focus on product development and pushing new features to market faster than ever.