When it comes to automating document processing, titles and deeds in particular present a vexing challenge. The reason is simple: these documents vary enormously depending on where they come from; the very definition of unstructured content.
In the U.S., titles and deeds for real estate properties are
handled at the county level. Within a given state, there may be anywhere from
three counties (Delaware) to 254 (Texas). In all, the 50 U.S. states comprise some
3,100 counties—an average of 62 per
Naturally, each county has its own forms for titles and
deeds, and they are not all alike – far from it. So, businesses that need to
process lots of these kinds of documents—real
estate, legal and mortgage firms, as well as other financial institutions, for
example—historically have been
forced to manually extract data from these forms and enter it into spreadsheets
and other financial systems. Clearly these firms could benefit from process
The trouble with templates
Document process automation promises a remedy, but not all automation systems are equal. OCR templating software, for example, relies on templates for automation. The templates indicate where relevant information is located in a document, such as a name, address, deed number and so on. With an accurate template, an OCR tool can then automate the process of extracting that data from a document and entering it into the destination system.
But when you’ve got more than 3,000 counties, each with its own
forms for titles as well as deeds – well now you’ve got more than 6,000
potential formats. Building templates in your process automation software for
each of them becomes an onerous exercise, if not a futile one.
Rigid rules and expert systems fall apart when faced with variability. You need a system that can actually understand language and context—like a human—to find the information you want. That’s where intelligent process automation comes in.
Intelligent automation addresses key challenges
Let’s say your company offers business, financial, legal or tax
services that require the routine processing of title and deed documents. You’ll
be facing challenges including:
- Dealing with documents from more than 3,000 counties, with variations in structure in terms of the position of data fields and surrounding context
- Many unstructured documents with data-field boundaries that are not well-defined
- The ability to auto-extract data from some fields but with no confidence score around the resulting accuracy
automation solutions can help address these challenges.
Intelligent process automation (IPA) software is fundamentally different from Rule-Based, Robotic Process Automation and other templated approaches because it can handle data even from unstructured documents. IPA uses optical character recognition (OCR), machine learning, and natural language processing (NLP) to enable it to understand the context in a given document and to identify the information you’re looking for.
process automation success
address this document process automation challenge, the first step is to take a
representative sample of the documents involved in the process – several dozen,
but nowhere near 3,000. The documents are used to train a deep learning model
to understand what relevant information looks like in a title or deed,
including the property owner’s name, address, purchase price, lot number and more.
the model is trained, there’s no need to create templates that tell the system
where each field is physically located in new documents that need to be
processed from each county. The model is smart enough to recognize a name,
purchase price and so forth, no matter where it may fall in the document. That
means the same model can easily adapt to documents with different layouts.
machine learning models can process a range of titles and deeds formats with
better than 95% accuracy. That’s a far greater success rate than any
templated-based solution could muster for any client with a massive document
base, and with far less effort upfront. In fact, the biggest problem we have when
convincing prospects to give intelligent automation a try is they’re gun-shy
after having spent lots of time conducting tests of templated approaches with
little to show for it.
We feel your pain, but rest assured—there’s no snake oil here. IPA offers a simple, effective way to automate the processing of titles, deeds and all sorts of other unstructured documents. To learn more, check out this white paper from experts at the Everest Group, “Unstructured Data Process Automation.”
Reference: Source link