Q1. Could you start by giving us a brief overview of your professional background, particularly focusing on your expertise in the industry?

In my two decades of experience, I have split my time between consulting on the vendor side, and leading enterprise platforms on the client side.

In consulting, I spent time offering enterprise solutions to a diverse range of clients from various industries, including oil and gas, telecom, government, manufacturing, and financial services. Meanwhile on the client side, I established large teams that were organized around core platforms and offering shared services to the lines of business. 

Throughout my career, I focused on unstructured data in all forms but particularly documents, within the context of digitalizing and automating processes to achieve straight through processing. 

My hands-on experience, implementing solutions from many vendors, as well as my experience researching the market to select the right solution to solve the type of business process problems we were facing, gave me the depth of knowledge and the wide lens to be able to paint a meaningful picture of the industry landscape. 

These days I have been advising startups and consulting firms on IDP strategies. 

I am happy to focus on Intelligent Document Processing today as it is one of my favourite areas of focus. 

Q2. Looking ahead, what transformative trends do you anticipate will most disrupt or enhance the enterprise IDP landscape, and how are organizations poised to adapt? 

IDP platforms evolved from using OCR to custom Machine Learning models to Large Language Models or more precisely Vision Language Models or a combination of all in a hybrid configuration. 

We will see the influence of the following trends in the next couple of years: 

• Open source LLMs and availability of AI models will further lower the entry barrier into the IDP market. More startups and new players will be relying solely on AI, skipping the OCR phase altogether
• Competition resulting in lower cost of LLM calls will make VLM even more appealing to the consumers
• Despite paperless initiatives and digital transformation efforts, reliance on paper in business processing is not going to go away anytime soon. A recent survey by AIIM found that paper remains prevalent, with 61% of IDP processes involving paper documents, and 48% expecting paper volumes to increase
• Legacy IDP vendors will continue to lose market share to new and more innovative players, including cloud vendors

 

Q3. How are industry-specific demands shaping the adoption and customization of IDP platforms? 

Increased demand for unusual use cases 

IDP platforms are architected to address various set of use cases and deployment options. Most of the legacy IDP vendors are suited for large volume high value use cases with low variation, and with human in the loop to validate documents and to ensure meeting the accuracy level. 

However, increased focus on front office processes and client experience priorities are creating the need to address more low volume with high variation but high value use cases. These use cases require the use of AI in addition to the traditional approaches. 

Highly regulated industries demand compliance-ready solutions 

Financial services as an example, don’t have the ability to compromise and accept the risk associated with adopting any solution that does not offer compliance with data privacy, security and data residency requirements. Hence, vendors who strive to sign lucrative contracts with multi-year commitment and high volume, invest heavily in making their IDP solutions check all the boxes. 

Low code and interoperability with other automation tools 

In my recent experience, while addressing an automation problem, we used the analogy of hands and eyes. For the RPA bot to process the transaction, it needs to “read” the data on the submitted form. IDP was the “eyes” while RPA offered the “hands”. Therefore, we will see increased focus on making IDP software able to interact easily with downstream systems, either agents, APIs or other means of passing the data to achieve the desired outcome. 

Q4. How are advances in AI and low-code/no-code capabilities changing IDP deployment strategies? 

Just over two years ago, I managed a program aimed at replacing a very well-established legacy IDP platform that has all the bells and whistles of enterprise, high volume, and centralized platform. The key challenges were speed of delivery (or lack of), high cost of solution development, and inability to cover wider spectrum of use cases. 

This was only a few months before the launch of OpenAI’s ChatGPT and the fury of events that followed, which enabled using LLMs to improve speed and accuracy of document classification and text extraction. 

This forced us to delay the project and adjust the vendor selection criteria and then added a step to review the vendors product roadmaps and their stance on using AI in the near future. 

No surprise, at the end of this race, the short-listed vendors were the ones who offered low code and speed of delivery in cloud deployment. The use of AI at this early stage was a major differentiator and showed the ability of some vendors to move strikingly fast and incorporate LLMs in their IDP solution. 

However, a couple of areas of caution that should be kept in mind when using LLMs in IDP solutions: 

• Overall cost per document increases, therefore, it should be used selectively for high value use cases, if prices are not brought down through negotiation
• Traditional ML approaches in IDP platforms enable setting up confidence level for classification and extraction, which is critical feature to determine how and when to present the document to a human to validate for quality assurance. This feature is not easily done when using LLMs which is a major drawback
• Model drift remains a concern, LLMs don’t promise stability of the accuracy levels over time

Q5. What types of advanced AI models have had the greatest impact on the scope and complexity of documents processed by your IDP platforms? 

The introduction of VLMs addresses use cases that are historically difficult for radiational IDP to resolve, such as processing documents with image and text combinations. A good example would be analyst research with graphs and charts, medical reports, or financial reports. 

Q6. What business outcomes have been most significantly impacted through your IDP initiatives? 

Efficiency savings remain the main driver for IDP initiatives, and this will continue to be the case. IDP help putting the data in the hands of knowledge workers to be able to process transactions, or in the hands of digital workers wherever appropriate. This significantly speeds up the process and reduce the cost of manual handling of documents in physical and digital formats. 

Improving the turnaround time improves customer satisfaction rates, and help complying with service level agreements, especially in processes that have strict cycle times and low tolerance to latency. 

IDP is enabling compound positive effects when combined with other automation technologies, such as digital workers (bots/agents) and low code / no code applications. 

Data analytics is another driver for IDP as it helps extracting data from imaged documents (unstructured data) which is an area that used to take a lot of efforts and cost a lot in the past. 

Q7. If you were an investor looking at companies within the space, what critical question would you pose to their senior management? 

I am going to address this question with focus on companies that are developing platforms that address wide spectrum of use cases and target the enterprise market. 

Here are my two cents: 

• IDP users need help realizing the benefit of automation, but this should not be on the expense of accuracy. If the wrong figure to be produced in a financial transaction, then the impact will be felt immediately in dollar figures. Focus on balancing automation and accuracy. There is always a trade off, and the difference lies in how much manual validation will be required to ensure meeting the accuracy targets. One way to improve progress on this front is to enable model confidence levels for data extraction which help improve data reliability and determine the human in the loop engagement at the right and adequate times
• Continue to invest in the user interface improvements. For high volume and high value transactions, human interaction with the platform remain critical and user productivity and overall experience make a lot of difference when selecting an IDP platform
• Continue to prioritize data privacy and security, especially in highly regulated industries. And tailor your approach to the individual country’s/region’s compliance requirements
• Last one, which not a lot of people talk about, make it easy to test and benchmark your IDP solution against the competition. Many enterprises don’t have the knowledge or the experience to run their own test; this is something I made sure to do instead of relying on vendor’s published marketing material