{"id":284,"date":"2020-09-24T22:25:26","date_gmt":"2020-09-24T22:25:26","guid":{"rendered":"https:\/\/machine-learning.webcloning.com\/2020\/09\/24\/how-chatham-financial-increased-process-capacity-by-400-with-indico-ipa\/"},"modified":"2020-09-24T22:25:26","modified_gmt":"2020-09-24T22:25:26","slug":"how-chatham-financial-increased-process-capacity-by-400-with-indico-ipa","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2020\/09\/24\/how-chatham-financial-increased-process-capacity-by-400-with-indico-ipa\/","title":{"rendered":"How Chatham Financial Increased Process Capacity by 400% with Indico IPA"},"content":{"rendered":"
\n
\n

\"\"\n<\/div>\n

Chatham Financial delivers risk management advice and technology to more than 3,000 organizations around the globe. In doing so, it has to process tens of thousands of complex, unstructured documents each year, a fact that had it looking for ways to automate at least some of that chore.\u00a0<\/p>\n

The company found what it needed in Indico\u2019s Intelligent Process Automation (IPA) platform. As an initial use case Indico enabled Chatham to automate one step of a common document review process, saving 15 min. of processing time for each document. That reduced the cost to perform the transaction by 75% while increasing process capacity by 4 times.\u00a0<\/p>\n

\u201cRather than a human-driven assembly line with everyone sitting at the conveyor belt with their hammer or chisel, Indico allows us to reimagine Chatham\u2019s offering as a pipeline, where a PDF comes in and key components are automatically extracted before it comes out the other end,\u201d says Dr. Christopher Wells, Chief Data Scientist with Chatham Financial.\u00a0<\/p>\n

In what is typically a \u201csix eyes\u201d process, meaning three people have to review the document, Indico takes away one step, or set of eyes, Wells says. And Chatham is just getting started.\u00a0<\/p>\n

Chatham\u2019s AI journey\u00a0<\/strong><\/h4>\n

When Chatham began its artificial intelligence journey in 2018, the initial idea was to build its own process automation application. \u201cA few PhD types and I downloaded TensorFlow and started building tools and labeling documents,\u201d Wells says. He borrowed an in-house user experience (UX) expert in hopes of making it usable by business people.\u00a0<\/p>\n

The company had already been putting data into data lakes and warehouses to make the data more accessible and flexible. \u201cBut trying to build an entire natural language processing engine from scratch was not feasible, given the time and resources we needed to spend,\u201d he says. The job is all the more difficult because of the unstructured nature of most of Chatham\u2019s documents, meaning a templated or robotic process automation approach would not suffice.\u00a0<\/p>\n

Christopher first encountered Indico at a start up fundraising event in Philadelphia, where he had a chance to meet with Indico\u2019s founder and CTO, Slater Victoroff.\u00a0 Slater described a breakthrough approach using Transfer Learning to \u201cunderstand\u201d documents without tedious and complex rules or templates.\u00a0 Slater further explained that this could be accomplished with as few as 200 samples for training the Indico \u201cDocBots\u201d.\u00a0\u00a0<\/p>\n

Before long Chatham brought Indico in for a demo, during which the group was immediately impressed with Indico\u2019s user interface, which they deemed crucial to the effort.\u00a0<\/p>\n

\u201cThat\u2019s really where you make contact between the data, your intellectual property, and the machine learning,\u201d he said. \u201cIf the subject matter experts can\u2019t or don\u2019t want to work with labeling the data, or if it\u2019s too hard to do, the venture is going to fail.\u201d<\/p>\n

After a thorough evaluation process, including comparing Indico to what Chatham could build itself, Wells and his team decided to bring in Indico for a proof of concept (POC) test.\u00a0<\/p>\n

Initial process automation results\u00a0<\/strong><\/h4>\n

The POC use case, which is now in full production, addressed an interest rate cap confirmation process. It involved a centralized SQL Server that acts as the source of truth for much of Chatham\u2019s processes as well as a blob store that holds PDFs of confirmations and other unstructured content. Chatham provided the 200 samples for training, and Indico was able to build a highly accurate model in just a week.\u00a0 The low training data requirement combined with the fact that Indico IPA was able to produce such impressive results in such a short time had Chatham convinced.<\/p>\n

Wells\u2019 team built a middle layer using Jupyter, an open source workflow engine, and Excel, with Indico sitting in between them as the natural language processing (NLP) layer. When a confirmation comes in, it\u2019s uploaded to the document store. That triggers a process whereby the Jupyter-based application pulls the document down and feeds it to Indico, which \u201creads\u201d the document. It is able to identify and pull out key terms \u2013\u00a0a step previously performed by a human \u2013 and enter them into the Excel spreadsheet.\u00a0<\/p>\n

The output is then fed to Compass, a custom application Chatham built years ago, which uses a script to put together an email to a subject matter expert (SME), who reviews the finished document. But, thanks to the IPA application, instead of having to go through the document line by line to find key terms and compare them to what the customer is reporting, the expert merely looks at the email. It is formatted with green or red checkboxes that indicate whether the document is good to go or requires further review.\u00a0<\/p>\n

For each document, the automated process saves 15 min. of what would otherwise be human processing time, resulting in the 75% cost savings and 400% increase in process capacity \u2013 meaning it takes far fewer people to do the same job.<\/p>\n

The results were so dramatic that Chatham was quickly able to deal with a backlog of more than 1,000 documents that had built up over the years.\u00a0<\/p>\n

\u201cOnce we rolled out our first version of this model, that backlog got cleared in one day, one 24-hour period,\u201d Well says. \u201cThat was the first time we were without a backlog of documents to process in I think two decades, which was a huge victory. Bottles were popped on that day, for sure.\u201d\u00a0<\/p>\n

Fast forward to today and Chatham has five IPA use cases in production and five more expected by year-end \u2013\u00a0no mean feat in 18 months for a mature organization with well-defined processes for adopting new tools, Wells notes.\u00a0<\/p>\n

\u2018Beneficiaries of our own success\u2019<\/strong><\/h4>\n

Chatham now has more than 50 internal Indico users, a testament to Indico\u2019s ability to turn business people into \u201ccitizen data scientists.\u201d\u00a0<\/p>\n

In fact, using Indico has improved the dynamic between Chatham business users and data scientists. Whereas at first SMEs questioned the ability of an automated tool to do their jobs, that has morphed now that the company has seen success after success with the tool. Key to that is the fact that business people are using the Indico tool right alongside the data scientists, \u201cnot just throwing their data over the wall to the data science team,\u201d Wells says.\u00a0<\/p>\n

\u201cThat collaborative piece, having one platform that we\u2019re all working on together, has really helped,\u201d he said.\u00a0<\/p>\n

With a number of successful projects under their belt, Wells\u2019 team now has projects flocking to the door.\u00a0 \u201cWe\u2019ve <\/strong>been beneficiaries of our own success,\u201d he says. \u201cWe have ongoing projects with every business vertical, and most practice area teams \u2013 accounting, transaction document process and client onboarding.\u201d<\/p>\n

Through it all, he notes, his data science team has added only one employee.\u00a0<\/p>\n

\u201cWe were able to handle building out all the infrastructure and the use cases, help business owners make the case to executive team leadership, deliver on model training and all the data pipelining necessary to connect our systems to Indico systems while only growing our headcount by one, which I think is a huge success,\u201d Wells says.\u00a0<\/strong><\/p>\n

The future of IPA at Chatham<\/strong><\/h4>\n

Looking ahead, he says Indico\u2019s IPA platform creates options for Chatham to further streamline its processes.\u00a0<\/p>\n

For example, the initial document review process the company automated is but one step in a multi-step process, he notes. All documents of that type undergo review by three different employees along the way. The company has successfully automated the middle step, taking away one set of eyes, but the door is now open to automate processes upstream, where the client uploads the document, as well as downstream, where final checks are conducted.\u00a0<\/p>\n

Wells is optimistic. \u201cWe\u2019ve pushed Indico hard on both our use of the models and types of things we\u2019re doing. We\u2019ve trained hundreds of models across many use cases, production and otherwise,\u201d he says. \u201cWe\u2019ve really stretched it to its limits, and it\u2019s held up.\u201d<\/p>\n

Watch the full video case study below:<\/p>\n

<\/figure>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

https:\/\/indico.io\/blog\/how-chatham-financial-increased-process-capacity-by-400-with-indico-ipa\/<\/p>\n","protected":false},"author":0,"featured_media":285,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/284"}],"collection":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/comments?post=284"}],"version-history":[{"count":0,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/284\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media\/285"}],"wp:attachment":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media?parent=284"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/categories?post=284"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/tags?post=284"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}