Founded 2016 | HQ Norderstedt, Germany | 40 employees (approx.) | $7M revenue (est.)
There is no question that ScaleHub checks the innovation box with a bold marker. The use of crowd platforms to augment AI-based cognitive capture makes a great deal of sense; although AI certainly improves the speed of capture processing and accuracy rates, it has its limitations.
ScaleHub AG, founded in 2016, is headquartered in Norderstedt, Germany. The company is led by Olaf Malchow, Ralf Goebel, Dan Dubiner, and Torsten Malchow and has received around $1.75 million in grant funding from the European Union. In 2016, ScaleHub acquired FasterAP Inc., a crowdsourcing API firm with a product called CrowdCigar. Today, the company has approximately 40 employees. We estimate that the company has revenues of around $7 million and that it is growing quickly and is profitable.
On the surface, ScaleHub provides a cognitive capture software system that is designed to process large volumes of documents, videos, and images. This software service is native cloud-based and targeted at business process outsourcers and shared service centers. However, a look beneath the covers reveals an innovative and powerful differentiator: the use of crowdsourcing. As the name suggests, crowdsourcing leverages large numbers of people to undertake work tasks. Crowdsourcing platforms such as Microworkers and Amazon’s Mechanical Turk distribute microtasks across many workers and collate and manage the work undertaken.
Cognitive capture products that leverage machine learning (ML)-based systems such as NLP, NER, and OCR are increasingly available from a practical standpoint. But even the most advanced ML- or AI-based cognitive capture system will struggle to reach complete accuracy. This is particularly so when the source documents vary in form or quality as in, for example, handwritten documents. Though deep learning (a form of AI) claims that over time it will be able to reach these high levels of accuracy, the cost, resources, and environmental impact of such systems are far from optimal.
With ScaleHub, documents (PDFs, TIFFs, etc.) are captured, processed in a regular fashion, and then imported to the ScaleHub portal, where they are analyzed (see Figure 1).
Where required, the documents are then automatically broken into anonymous snippets, much like a jigsaw puzzle. The snippets are distributed to crowd workers who key in or annotate the information they see in the snippet. For example, a worker might key in “this is a letter ‘t’ rather than the letter ‘J’” to verify or complete an OCR operation. Each piece of information is sent to two crowd workers to ensure they agree that the result is correct. If they disagree, it is sent to a third, more experienced, worker to break the tie.
The results from the human analysis are automatically reassembled to complete the process. What we have here is a form of HITL (Human in the Loop) where crowd workers augment and substantially improve the accuracy of AI-based document capture (to near 100%) on a vast and granular scale.
The same techniques can be used to initially train a system for a customer; however, in this case it means revealing the entire form so that the positions of the form fields can be reliably plotted for future automatic extraction. To ensure that this can happen at speed while guaranteeing the same level of data protection, ScaleHub has an automatic anonymizer which is applied to these tasks. Thus, crowd workers can receive an entire form with the form fields and associated legends to be plotted visible, but the data hidden.
ScaleHub’s Computer Vision image processing service operates similarly in terms of task distribution and verification; it generates and manages workloads such as object detection (identification of objects within a 2D image), segmentation (identification of objects of a specified type within a 2D image that may contain many object types), and polygon identification (identification of an object type with a multipoint box within an image). Applications here include processing and verifying insurance evidence (e.g., dashcam footage), identifying buildings from satellite or drone images, and building high-quality training and test sets for ML models.
In theory, any BPO could run something similar, as crowdsourcing platforms are freely available. But where ScaleHub truly sets itself apart is that it has built a product it calls an “Elastic Crowd Balancer.” This orchestration service manages, monitors, and optimizes the crowd and capture activities across platforms, ensuring accuracy first and foremost, plus 24/7 availability, and – in theory at least – near unlimited scalability. Today, ScaleHub orchestrates approximately 15,000 crowd workers around the clock, has access to 2.3 million workers, and supports 160 languages and multiple regional locations.
Though crowd working has been around for some years, it is still a novel and underused resource. However, driven to some extent by the pandemic, crowdsourcing may well begin to play a strategic role in outsourcing decisions.
To assuage any concerns a new organization may have in taking this route, ScaleHub has ensured that its services comply with GDPR, HIPAA, and ISO 27001. The company also allows for data sovereignty to be managed by running data centers in the US, EU, Japan, and Australia while restricting all data administration through ScaleHub itself by running virtual private clouds and encrypting data both during transfer and while on the servers.
Just as importantly, ScaleHub ensures that all crowd workers get paid above minimum wage in their country of residence, and it has built complex payment calculations to ensure microtasks are paid equitably into the system and accounted for in its service pricing. Furthermore, as of today ScaleHub is the only company we know of in the market that promises 99.x% data accuracy in its service level agreement (SLA).
There is no question that ScaleHub checks the innovation box with a bold marker. The use of crowd platforms to augment AI-based cognitive capture makes a great deal of sense; although AI certainly improves the speed of capture processing and accuracy rates, it has its limitations. The focus in the industry right now is on the potential of deep learning, and while deep learning has indeed been a game-changer with the entry of Microsoft, Google, and Amazon into the market, its limitations are also becoming more apparent.
Layering crowd-based microtasking into the equation can take those accuracy levels and scalability to another level. This approach would undoubtedly be more widely used in the industry if not for the sheer complexity of managing microtasks and their associated workers. ScaleHub has an orchestration system to do just that, and as the firm emerges into the mainstream it will gain a lot of attention, and rightly so.
Advice to Buyers
Though growing fast and profitably, ScaleHub is still a relatively small company. Therefore, this is an ideal time for BPOs and shared service centers to take a look and see if there is a potential opportunity to trial or work in conjunction with ScaleHub. Though the firm is relatively small, its product is designed to operate at scale. Moreover, it is the only company in the market with such a crowd orchestration product layer (to the best of our knowledge). Leveraging ScaleHub to augment extensive volume capture activities in banking, insurance, and healthcare is, at the very least, an exciting possibility to explore.
- First to market
- Crowd orchestration for capture
- Grow internationally
- Become a prime choice for BPO organizations
- Partnering with BPOs
- Partnering with shared service centers
- Accuracy rates >99%
- Processing >2B documents per year