Amid global economic slowdown and an ongoing labor shortage, companies around the world are looking for novel, cost-effective solutions to a seemingly infinite number of problems. Forrester estimates that enterprise investment in robotic process automation (RPA) is set to hit $22 billion by 2025, and artificial intelligence (AI) is forecasted to reach $37B in the same period. These estimates mark 20X and 3X increases respectively over a period of just six years, partially in response to broader macroeconomic headwinds.
From process discovery to Intelligent Document Processing (IDP) and Unstructured Data Processing (UDP), the value of Intelligent Automation (IA) continues to expand and evolve. Within this context, many business leaders have been contemplating the role of IA amid the broader economic slowdown. I recently sat down with Jasper Masemann, Partner at HV Capital, a premier European venture capital firm, to understand his perspective. We covered a range of topics, including:
From Jaspers perspective as a venture capitalist, valuations appeared out of hand over the last 12 months, with some companies raising funding at 100X their forward-looking revenue. Public company valuations have reverted back to pre-pandemic levels, which on the surface might feel bad, but ultimately is a more accurate reflection of what these companies are actually worth. HV Capital has accelerated its pace of investment, as the market dynamics have presented unique investment opportunities. The firm has been fortunate that most of their portfolio companies did not raise money at crazy valuations, and are better positioned to ride out the anticipated recession.
As with many of the terms used in the world of business technology, there is often confusion around the distinction between “digitization,” “digitalization,” and “digital transformation.” Digitization, which refers to creating a digital representation of physical objects or attributes, has been ongoing since the 1960s. Digitalization refers to enabling or improving processes by leveraging technologies and digitized data. Finally, digital transformation is the business transformation enabled by digitalization, where digital technology is integrated into all areas of a business, fundamentally changing how it operates and delivers value.
Jasper believes digitalization and digital transformation were rapidly accelerated by the pandemic in two major ways:
The first wave of RPA was like a gold rush, where the gold was in the river and you could see it from the shore. In the beginning, the data used for process automation was structured, so practitioners could easily understand the rules-based scripts that drove its rapid adoption. Unfortunately, consultants over positioned RPA—claiming it could automate any and every process—when in reality it was heavily dependent on structured inputs that were processed according to clear rules.
Consequently, most of the process automations involving unstructured data have stalled. Documents were the first type of unstructured data that RPA users sought to process using optical character recognition (OCR). However, most OCRs involved difficult setup that required creating a template for each new document format the users wanted to process.
Five years ago, companies like Automation Anywhere saw the opportunity to reinvent OCR. I helped the company launch IQ Bot and create Intelligent Data Processing (IDP) as a category. When I first joined super.AI, I thought the document problem was solved—with 75+ well-funded startups and large software vendors focused on it. However, users are finding first-generation IDP solutions hard to use for a few different reasons:
For the simple reason that unstructured data is everywhere, we encounter a wide variety of use cases at super.AI. Recently, we have seen companies adopt AI-powered automation for the following applications:
Customers seeking to automate the remaining 20-40% of documents they’ve struggled to process with existing IDP solutions have approached us due to our platform’s flexibility. Our approach involves humans working with AI to process 100% of complex documents (e.g., unstructured, handwritten, featuring stamps) at scale. We accomplish this using our Data Processing Crowd, an on-demand resource pool of human workers that process data with usage-based pricing.
Substantial edge cases remain in document processing, even for common documents like invoices and purchase orders. There are also opportunities to process unstructured documents such as contracts using natural language understanding (NLU) and natural language processing (NLP), or emails to enhance marketing efforts. Ultimately, what every company needs is an AI that works for their specific use cases. Enterprises are looking for AI solutions that are accessible to any business user. Products that eliminate the need to select suitable AI models and hire data scientists to optimize and maintain them. This enables non-technical workers to bring automation to the highly specific use cases they know best, and when automation isn’t possible easily leverage skilled human workers to close the gap.
Vast quantities of often unexpected data enrichment challenges exist across enterprises, and several super.AI customers use our platform to enrich and improve their existing data.
For example, it is common for charges on bank statements to show billing descriptors that are confusing or completely unrecognizable. While some merchants use clear names, many (especially small businesses) simply don’t. As a result, when people look at their credit card statements they often see unrecognizable charges. One of the largest financial services companies in the world partnered with super.AI to identify patterns buried in unstructured billing descriptor data, then automatically transform unrecognizable merchant names into clear transaction records. Both card issuers and cardholders now benefit from having easily recognizable merchant names attached to each transaction.
Due to increasingly strict data privacy laws like the General Data Protection Regulation (GDPR) that covers the European Union (EU), customers have started approaching us to automate image, document, and video redaction at scale. Often in an effort to protect personally identifiable information (PII) or otherwise sensitive information, use cases include face blurring, Social Security number redaction, license plate redaction, and more.
Today, visual inspection is highly manual and very paper-driven process that involves people taking pictures on-site, then manually analyzing them to create inspection reports that document damage, cracks, corrosion, and more. Using AI, it is possible to automate not just visual inspection reporting, but expand monitoring so that it’s always-on, providing feedback in real-time and detecting damage at an early stage.
According to two Gartner reports, 85% of AI and machine learning projects fail to deliver, and only 53% of projects make it from prototypes to production. In our conversation, Jasper and I discussed how that is beginning change.
Jasper believes that AI projects of the past involved a lot of experimentation. People were trying to figure out whether AI would work with their data, what kind of models work best, and what kind of team they would need to build. Today we know what models to use. We know the process for data labeling. We know how to deploy models in production. We know what people to hire. That means AI projects today are focused around execution instead of experimentation, resulting in much higher success rates.
Customers are also realizing that emerging AI platforms like ours make it possible take any business problem involving unstructured data and quickly create an enterprise-grade AI application that automates data processing, and ultimately increasingly complex tasks. These platforms are becoming the next step beyond the current automation platform of their choice (e.g., RPA, low-code, BPM).
Amid the slowdown, venture capitalists around the world have been asking their portfolio companies to slow their burn rate to create 24-36 months of runway. Companies (like super.AI) that raised funds last year, or earlier this year, are in a good place with sufficient runway. According to Jasper, advising business operators to extend their runway to 24-36 month is like telling a runner to carry as much water as possible to finish a race. Instead, start with a few questions: Where do you want to run? How far away is it? How much water do you need to get there? For a business, this sounds like:
Once these questions are answered, the goal should be to scale it fast and raise the next round of funding. The question then becomes: How long will that take? In truth, it is far more difficult to raise money now than it was last year. Likely, raising the same amount of capital will require companies to achieve a lot more than what was asked of them in recent history. This means it will probably take longer to achieve goals, requiring more runway. Intuitively, this all makes sense, but the question is more about what a company has to do to reach the next funding round, and how long it will take to get there (rather than blanket advice based on an arbitrary timeline).
Startups should also check the costs associated with customer acquisition. For instance, if a startup is getting $1 back for every $5 spent, something is wrong. Previously, this loss might have been justified as a cost of growth, but investors are now looking for companies that can generate profits not just customers.