A Start-Up Vision for AI-as-a-Service
Almost two years since the launch of the Sydney start-up, IDM asked Sypht CEO Warren Billington to outline the growth and development of the company’s cloud-based AI-as-a-service platform for data capture.
Sypht has established a proprietary SaaS platform on the Amazon Cloud that uses Google Vision OCR, machine learning and natural language processing that instantly and accurately extracts and interprets valuable information from unstructured data in documents used in everyday business processes.
IDM: Warren, what has Sypht been focusing on for the past 12 months?
WB: We started very much as a team that evolved out of BCG Digital Ventures, driven by our parent company the BPAY Group, with a core engineering and data science group of 10. We’ve now almost doubled our staff and have increased the data science team to five people, which is a considerable investment from us, but, indicative of the demand we're seeing in market across a whole variety of different document types and use cases and as we continue to innovate around the product.
We've also become much more of a commercially oriented business, so building out our sales, customer success and marketing capabilities has also been a focus this year.
IDM: Where is Sypht being deployed in the market today?
WB: One of our early wins was at Australian SMB accounts specialist MYOB. We are working behind the scenes to process a huge range of invoices and other documents submitted online by MYOB users. We have now processed over 22 million documents at a median of five seconds per document with accuracy over 90%. Since we commenced working with MYOB in 2019, we've actually seen incremental gains as it relates to accuracy which is a testament to the self-learning capability we have built in to Sypht. The more data that we process over time the more we have been able to improve in performance for MYOB.
Another large customer is a major Californian energy utility that is using Sypht to clear a significant backlog of manual documentation and processing in the field. For them the challenge was to digitise inspection reports that thousands of contractors would have to complete. They need to be able to process that information quickly to ensure they are meeting obligations and requirements around compliance. We've been able to address that by reducing report processing time from 4 years and automate the entire process within 3-4 weeks.
We have also had success with one of the organisations within the NDIS space, My Plan Manager These organisations have to deal with a huge variety of documents submitted as part of NDIS claims processing. There is also strong demand around identity verification in government and in lending use cases, for example where multiple documents are required for income and identity verification.
Another area that is starting to look really quite promising is around property, with document types such as leases and contracts. The market is really starting to evolve across multiple verticals.
Sypht CEO Warren Billington
IDM: How is the Sypht platform evolving?
WB: Sypht is really focused around data capture but we're now seeing an opportunity to overlay value added services. Once you've captured data, then there is the opportunity to be able to plug Sypht directly into downstream or upstream systems which allows more end-to-end automation of workflow and straight-through processing through Sypht Connect.
Another new product that we are launching is called Sypht Signals. This enables organisations to set-up signals around key events and triggers for real-time detection and freely search structured data post-extraction to discover insights creates such as detecting fraud as it relates to invoices and claims processing. Another tool we have developed but are yet to commercialise is called Sypht Validate, which will help organisations review data in a UI where AI confidence levels can be set at thresholds to either confirm predictions as true or annotate the correct values so the AI continuously improves allowing organisations to augment their human workforce with intelligent automation. As we start to build more capabilities around new document types and new use cases, these will be developed as AI and Insight products that can then be commercialized more broadly into market. I think the big shift for us is repositioning Sypht from just a single product with an API to an “AI as a Service” platform with multiple products into market.
IDM: Is the Machine Learning technology that Sypht applies only of use to organisations that are processing millions of documents? What about the SMB sector with much lower volumes?
WB: One of the key value propositions that we believe is really differentiating Sypht is around our adaptiveness across many business units, industry use cases and document types for rapid scale. Our ability to build self-learning algorithms and model capability powered by Sypht’s global data network effect can be done from a very low sample of documents. So, when we work with a new customer, we need no more than a couple of hundred documents to then be able to build a model that delivers strong performance. And we believe that is really differentiating for us.
The whole notion that you need large volumes of data to get to those performance levels is not the case. We use a lean model development process and can build self-learning AI models within three to four weeks that allows us to rapidly mobilize a proof of concept with a minimal effort to deploy and integrate and then prove the value very quickly.
IDM: Has COVID-19 impacted this market in a significant way?
WB: It’s having a big impact. It's clear that organizations now want to move more quickly on digital transformation and there is a growing appetite to be able to introduce AI based automation or cloud based technologies. But people also want to be able to move quickly and be able to prove value and ROI.
Sypht can take a low sample of data and get a POC up and running within three to four weeks to prove value that then provides the business case for further investment and the roadmap for future use cases for unstructured data. There's very much a focus from businesses on short term projects that deliver immediate ROI, rather than committing to long term strategic consulting- driven transformation projects