Taking data to the cloud

With the explosion of data, every application is now a data application. Learn why this has hundreds of software teams building their applications in the cloud across multiple use cases and industries.

Vic.ai, a startup developing software to automate accounting and financial processes, today announced that it closed a $50 million series B financing round with participation from GGV Capital, Cowboy Ventures, and Costanoa Ventures. Pivoting away from its initial focus on accounting professionals, Vic says it’ll use the capital to expand its existing enterprise offering as well as build out its financial intelligence engine.

Experts believe that automation holds enormous potential in the accounting industry. According to an Ernst & Young survey of finance leaders, 53% believe that more than half of finance tasks currently handled by people could be performed by AI over the next three years. In a separate study conducted by cloud accounting software company Sage, 45% of certified public accountants (CPAs) said that they intend to automate tasks including data entry and number-crunching, while 40% said they plan to automate invoicing and accounts payable processes and workflows.

Vic, which was founded in 2017 by Norwegian entrepreneurs Alexander Hagerup and Kristoffer Roil, combines the two pillars of (1) invoice processing and (2) optimizing business decisions while applying reasoning to the data that’s processed. The platform delivers features like the AI-powered Autopilot, which selects invoices and expenses that meet a certain confidence threshold and automates them so that they’re immediately sent to approvers without requiring human review. Another feature, Autonomous Approval Flows, determines the number of steps in an invoice approval process and automatically decides which employee needs to review each step.

“No matter the industry, every company needs accounting, which is inheritably tedious and time-consuming. Accounting tasks have always been managed by using legacy systems that are based on various predefined rules and templates, like Excel,” CEO Hagerup told VentureBeat via email. “Kris and I wanted to infuse intelligence into accounting through an AI solution that can reason like humans and make accounting decisions so that employees can focus on more high-value activities … Today, Vic delivers fully autonomous AI systems that make finance and accounting teams more efficient, accurate, and intelligent.”

AI under the hood

Hagerup and Roil built the first iteration of Vic by training the platform’s AI on historical accounting data and processes from more than 30,000 companies. The training dataset contained over 200 million accounting documents and corresponding journal entries, amounting to 300 million records, which were reviewed by accountants at consultancy firms including PricewaterhouseCoopers, BDO, and KPMG. This “live usage” helped to train Vic’s machine learning algorithm over time, enabling it to provide nearly “complete autonomy” for transaction processing, according to Hagerup.

“There is less bias to mitigate in accounting data, as the results are audited following official standards, and the audited data is fed back to the algorithms. Since we make predictions across 4,500 entities in multiple regions with thousands of accountants reviewing, any client or accounting firm preference — i.e., bias — is eliminated mainly by the size of the data and the review by external auditors,” Hagerup said.

Craig Le Clair, principal analyst at Forrester, cautioned that a potential Achilles’ heel for AI in finance and accounting is the extensive variation in process across firms that prevents AI solutions from being developed in a packaged and repeatable way. In short, he says, AI for finance and accounting is “being oversold a bit today.”

“Use cases such as account reconciliation, monthly close, invoice processing, financial planning and analysis, and adjacent areas like audit and contract analytics rely on extensive numbers and data and are prime candidates for AI, but do not provide value today. Based on the maturity of used AI components and adoption characteristics, AI usage today is not widespread,” Le Clair told VentureBeat via email. “Glancing at Vic.ai, it seems they do some invoice extraction. Also, they seem to do ‘reconciliation,’ but [they’re] somewhat vague on the ‘insight’ piece — [it’s unclear] if they have a financial planning and analysis solution that competes with specialists in finance and accounting.”

Be this as it may, New York-based Vic has processed more than 535 million invoices to date for more than 4,000 corporate finance and accounting clients, among them HSB, Intercom, and Armanino. Annual recurring revenue reached $5 million, growing three times year-over-year. And Vic expects to nearly double the size of its full-time 35-person workforce to 67 by the end of the year.

“From a customer and market adoption perspective, the pandemic has only accelerated the usage of cloud-based platforms for managing accounting work, and we see demand for our solutions,” Hagerup said. “Vic AI algorithm can perform reasoning intelligently and adapt to new requirements, making the accounting process autonomous.”

With this series B, Vic has raised $63 million in venture capital.

Taking data to the cloud

With the explosion of data, every application is now a data application. Learn why this has hundreds of software teams building their applications in the cloud across multiple use cases and industries.

Presented by MongoDB

Founded in New York City in 2009, Rent the Runway is disrupting the trillion-dollar fashion industry and inspiring women with its vision for a more joyful, financially savvy, and sustainable way to feel their best every day. Through Rent the Runway, customers can rent, buy, or subscribe to designer apparel, accessories, and home decor from over 750 brand partners. While Rent the Runway aims for a seamless process for its customers, the back-end technology powering this one-of-a-kind logistics operation is complex.

The company has two distribution facilities — one in New Jersey and one in Texas — that receive, clean, and repair items before swiftly packing and shipping them out again. This involves a significant investment in AI, radio ID tags, and robots to more efficiently sort, clean, and ship the garments.

Ahead of his presentation at MongoDB’s annual developer conference, Rent the Runway’s Director of Engineering, Mike Liberant, discussed the company’s data strategy and code-first approach and how that factors into the company’s end-user experience.

MongoDB: Can you describe what goes on behind-the-scenes at your distribution facilities when a garment is returned by a customer?

ML: First, every garment returned to our warehouses must be cleaned and sorted before it’s available to rent again, so our goal is to automate the movement of goods where possible to streamline this process.

Worn garments are first ushered onto a conveyor belt before they go through an X-ray machine. Why an X-ray machine? Quite frequently, our members will accidentally leave items in the pockets of their rented garments (lipstick is a frequent offender) that we need to catch before they go into a washing machine.

After it passes through the X-ray — hopefully makeup-free — our software sorts items into one of 20+ different bins for cleaning. Adding to the challenge is the fact that these garments require different cleaning methods, meaning that the robotic arms managing the sorting process have to understand which bins correlate to which cleaning method. These solutions are important to scaling our operations, and success comes down to equipping our development teams with tools that help them make sense of the massive amounts of data generated.

MongoDB: How does this massive amount of data impact your development teams and data model?

ML: Given our tight timeline, using a relational database was a no-go due to the upfront requirements you need to account for in terms of designing schema properly and the corresponding data model. We knew there would be a large learning curve and we would have to consistently iterate on that data model, which entails changing your schema for every iteration when using a relational database.

This means a database administrator has to coordinate with a development team to add columns in, and the developers then have to go back and update their code to match the corresponding data. It’s a unique challenge for our engineers because we’re working with a variety of different garments and we’re trying to build software that can clean them correctly.

MongoDB: How did your database choice impact the overall outcome?

ML: Using MongoDB’s document data model was key to reducing our developers’ cognitive loads. It doesn’t require the upfront work of designing our schemas and it allows us to stick to a code-first approach.

As object oriented software developers, we can conceptualize the world around us as objects easily. For example, a car is an object with certain attributes and a garment is an object with certain attributes. When that garment has a specific cleaning method attached to it, it’s just another attribute. The document model allows us to deploy, go live, and add attributes later on with very little coordination between database administrators. This means our developers write code exactly how it appears in their heads, instead of having to normalize the data into multiple tables.

MongoDB: Does this data strategy stretch across the company?

ML: Since joining Rent the Runway in 2019, I have focused on defining our North Star vision for building applications, including how we architect our systems. We have been able to separate each business domain using microservices backended by MongoDB Atlas, Kotlin, and Spring Boot, to provide a modern tech stack.

During the pandemic, we focused on increasing efficiency through enhanced automation while inbound and outbound volume was lower. Minimizing our inbound processing time is not only good for our business, it helps us free up inventory faster so our customers can rent or buy it. When a garment is picked up by a robotic arm, the arm scans the garment’s RFID tag and determines what bin it needs to be sorted into. The robotic arm also tells us when a bin is full and needs to be moved to a washing machine, when it’s empty and all these other sorts of data points that are useful. Anything data-related also gets copied to our data warehousing system.

Any downtime in our warehouse creates a delay for our customers, so we proactively took measures to de-risk this by implementing triggers — which enable us to execute application and database logic automatically — either in response to events or on a predefined schedule. Using Realm Triggers within Atlas to pipe data to our data warehouse is essentially a no-code solution that helped to further de-risk our entire system, allowing us to extract the value of this data for future forecasting of warehouse workloads.

MongoDB: What were the benefits of using a multi-cloud database service?

ML: Using Atlas, we reached our time to market for the warehouse automation rollout in half the time of our legacy tech stack. We achieved this purely through this code-first, dynamic data modeling approach and were able to improve the efficiency of our inbound sortation dramatically. At the end of the day, this plays a key role in improving inventory availability and, therefore, creating a superior customer experience.

Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. Content produced by our editorial team is never influenced by advertisers or sponsors in any way. For more information, contact [email protected].