Deep Data Insight: Using Artificial Intelligence to improve workflows

Deep Data Insight: Using Artificial Intelligence to improve workflows

 

Introduction

Deep Data Insight works strategically with a client based in California who specialise in finding commercial properties for their clients in the healthcare sector. Their client works on a large-scale; they have around 50,000 properties within their portfolio at any time and work on both a purchase and lease basis. The client is a successful and growing business; they are adding hundreds of new properties to their portfolio every month.

Deep Data Insight has created an Artificial Intelligence factory that produces bespoke and licensable AI solutions for organizations from almost any sector.

One of their modular platforms, EDDIE, uses Optical Character Recognition and Intelligent Character Recognition technologies to provide huge ROI for companies that have medium to large amounts of data to process.

 

Challenge

In short, DDI’s client faced a workflow challenge. Their existing processes were manual, with information scattered over a number of files and locations. They were experiencing a lack of efficiency and occasional mistakes. And since their client base covers numerous and disparate locations, the task of bringing this all into one place was vast.

In workflow terms, the challenge was a combination of digitizing documents, checking for errors and duplication, consolidating information from a variety of sources and finally extracting important information from less important information documents. The data was in handwritten, typed and picture format. 

 

Solution

Deep Data Insight deployed their EDDIE platform to provide solutions to four specific areas:

  1. Address matching

At any one time, the client holds a huge repository of building addresses which are uploaded onto a series of spreadsheets by a team of researchers. Different researchers can potentially enter the same address multiple times, so these require cross-referencing and master indexing to ensure duplicates are removed and there is one unique, accurate entry per building. 

To complicate matters further, the client is also in charge of a list of around 3,000 physicians that will move between institutions. In fact, the value of the properties can depend on the number and seniority of physicians within a building, so, the master index is a live and moving thing! 

EDDIE solves this problem by pulling data from the client’s servers, processing and cleansing the data before sending it back in a master index to the client. A master index can be retained by Deep Data Insight if required.

Obviously, security is of the highest priority, so the Deep Data systems are completely secure, as is the process of transferring data.

The technologies involved in this solution are exact matching, parsing, deduplication and master indexing.

Since EDDIE is phenomenally quick, the ROI is impressive for the client. EDDIE can run an entire cycle for around 300,000 buildings within 5-10 minutes.

 

          2. Offering Memorandums

A critical part of what Deep Data Insight’s client does is to produce marketing brochures for their properties. These are around 15-20 pages in length and cover all aspects of the building, including environment, utilities, condition etc. They are all typed by a person in the first instance. Since this client works on such a large scale, hundreds of brochures are being produced each month. 

The challenge comes from the fact that all of the information needs to be accurate and de-duplicated. The process improvement comes from extracting the important data which will be in amongst less important, marketing information. The client’s database needs to include key information every time for every property – square footage, location, rental terms etc. Without EDDIE, this is a time consuming process that is prone to errors.

EDDIE uses AI to go through the whole brochure to pull out the 20 or so pieces of information that are going to be valuable. This was previously done by staff members by reading each document, which had been time consuming.  It will also go through an address matcher since the same brochure might come in from multiple sources.

Once OCR and ICR have been used at the start of the process, the technology being used is string matching and a deep learning language model; if this cannot be applied, then EDDIE will trend the model by Question and Answer.

Since the EDDIE can work away in the background, the client will start processing the brochures overnight, so that at the start of the working day, everything is ready, saving at least one person’s salary.

 

          3. Transferring ‘Underwriting files’ onto a central database

The client produces thousands of underwriting files every month. These are pre-filled excel sheets, manually completed. They will vary in their make-up; sometimes cells are merged…some contain pictures.

DDI’s client needs all of these files moving from their archive onto a central database, so that they are more accessible in the future and can be searched globally. Naturally, not all the information contained is required…the salient information needs identifying, extracting and digitizing.

This is what EDDIE does, using Logic models. EDDIE will process each sheet including all tabs completely within sixty seconds. This provides enormous ROI to a process that would have taken hours for a human to complete.

 

          4. Lease files

Every Lease File that the client accesses is a complicated legal document of up to 200 pages in length. They contain a massive amount of superfluous information for the client’s purposes and are therefore impossible to search quickly. In fact, within each full document there are around 22 fields that actually need extracting. Imagine having to review the whole document for that one piece of critical information, for example what are the building’s boundaries. Even though they are legal documents, there are still many different styles.

EDDIE processes the entire document, and extracts only the salient information using string matching technology and a deep learning language model. If data in a field cannot be directly matched, then EDDIE will trend the model by Question and Answer plus logic and a rule-based method. This is particularly useful if a specific piece of information is needed that isn’t standard… …for example who is responsible for maintaining the roof?

The tech being applied here combines language transformation (A BERT plus GPT2 or 3 language model) as well as OCR and ICR at the start of the process.

ROI is incredible! Processing time is 2-3 minutes to search around 100 pages to extract all 22 fields.

 

Results

It is fair to say that using Deep Data Insight’s EDDIE platform, the client is able to bring huge workflow benefits to their large-scale operation by using Artificial Intelligence to replace human interaction. Not only have their processes been super-charged, but they are making use of  what would previously have been downtime and reducing errors along the way. They are gaining a competitive advantage over other organizations, and all in a completely secure, locked-down environment.

Jeewa Perera is the CEO of Deep Data Insight, and comments:

“EDDIE is a brilliant and flexible platform, and even though the client are only scraping the surface of its capability, they have established long-term benefits through a real partnership with Deep Data Insight”