UiPath Document Understanding Reviews

In Italy, one of the most prevalent use cases involves automating the processing of invoicing cycles. The issue we aimed to address through the integration of this solution is essentially the manual input of data into systems by humans and the need for checks and balances between invoicing and other physical documents. Our organization is in the manufacturing realm. We primarily use Document Understanding to process invoices, specifically a common document in Italy known as the BDT. Regarding the document format, it includes structural elements like tables, checkboxes, and headers. Some documents may feature large tables, and the header contains essential information that needs to be extracted. In terms of volume, for a medium-sized or small company, we handle approximately ten thousand of these documents annually.

The advantage stems from the seamless integration of this solution with the UiPath platform. If a customer already has the standard, robust UiPath platform operating within their systems, adding these smaller modules is all that's required to enable Document Understanding. It functions as an integrated ecosystem.

It facilitated the automation of our data entry processes.

Approximately twenty to thirty percent of our customer's documents undergo full automation in processing.

In our scenario, Document Understanding operates independently as a standalone module, not integrated with any other systems. The robots, however, interact with the systems.

The average processing time, before and after automating with Document Understanding, improved in speed for a minute.

Human errors have been reduced by seventy percent.

Document Understanding has contributed to freeing up approximately seventy percent of people's time for other projects.

I believe the most valuable feature is the prebuilt algorithm for extracting information from foreign invoices. This efficient algorithm eliminates the need to create one from scratch.

It has the capacity to manage diverse document formats, including handwriting and signatures.

Leveraging artificial intelligence or machine learning capabilities is beneficial. These technologies excel in field identification tasks, even when adjustments such as moving or rotating the identified fields may be necessary. The primary benefit of artificial intelligence lies in its ability to handle various layouts.

Around 20 to 30 percent of cases necessitate human validation for Document Understanding outputs. The human validation process typically takes less than one minute per document.

There is room for improvement in handwriting processes. It should enhance the user interface for constructing extraction logic. It is not as user-friendly as other parts of the platform. An additional feature that could be considered is the integration with generative AI. The deployment process should be more user-friendly and streamlined. Scalability capabilities should be improved, as well.

I have been using it for two years now.

It offers good stability. The need for maintenance decreases with the highest level of stability.

Scalability is limited as it relies on the document layout. Integrating generative AI could potentially address this aspect. Moving an algorithm to another project without making significant changes can be quite challenging.

Our experience with its technical support is quite satisfactory. I would rate it nine out of ten.

Positive

The deployment process is not as straightforward as a seamless deployment, such as with App Studio. The number of people required for a project depends on its nature. Typically, one or two individuals are sufficient for most deployment cases.
Maintenance requirements vary depending on the projects. The team size can range from one person to five, six, or seven people. The deployment of this solution required one month.

I believe a six-month payback period is reasonable for the time-to-value. A shorter duration would be more favorable for customers.

I find the pricing to be somewhat on the higher side. User decisions are impacted by the pricing structure.

Overall, I would rate it nine out of ten.

We use UiPath Document Understanding for two purposes: extracting information from medical certificates issued by a prominent university in Singapore and processing invoices for a client in the logistics industry within their ERP systems.

We implemented UiPath Document Understanding to significantly reduce the substantial mailout effort. Approximately 20 full-time employees were previously dedicated to these processes, but after implementation, we were able to halve the number of full-time employees required.

We are capturing the header line items, which include the account number, invoice number, invoice date, and the line items: quantity, line item description, unit price, taxes, item number, and ZIP codes. This is a sales sector document. The medical certificate is an untested document, and we need to capture specific dates, the doctor's medical certificate number, and the student's name. We also need to check whether a checkbox is checked. There are no handwritten documents to extract.

Around 80 percent of our documents are processed 100 percent automatically.

Before implementing Document Understanding, the average time per invoice for manual processing, including invoice scanning and data extraction, was 15 minutes. Following automation, the processing time has been reduced to six minutes, with the specific duration varying based on the number of features on each invoice.

Document Understanding has helped reduce human error by 70 to 80 percent.

Document Understanding has reduced staff time by nearly 50 percent.

The best feature is pre-labeling, as it eliminates the need to manually label each data point. This saves a significant amount of time and effort. Additionally, the generated text is integrated seamlessly into the tool, making it easy to use. The documentation is also very clear and concise, making it easy to get started with the tool.

Over the past few years, I have observed that the invoice model consistently improves with each new UiPath release. There is still room for enhancement in capturing line items from invoices. This is one of the areas where I believe we can achieve near-perfect data capture. Unfortunately, the current accuracy rate for capturing line items is between 50 and 60 percent. This necessitates manual two-way matching, which is time-consuming and inefficient. I believe UiPath Document Understanding can still improve in this area, but overall, it is moving in the right direction.

Despite advancements in artificial intelligence and machine learning, there are lingering concerns about data privacy and security. These concerns can have a significant impact on users, particularly in terms of geographic restrictions and data policies.

The accuracy level we receive does not justify the price, as many competitors are offering much lower prices.

I have been using UiPath Document Understanding for three years.

While the stability is improving, it still needs to be enhanced in terms of model learning.

UiPath Document Understanding is scalable, but there is an aspect of training that requires attention. The model should be trained with a specific type of invoice to ensure optimal accuracy. For instance, if the invoices are in multiple languages and formats, the post-model training results may not be as effective as compared to training the model with invoices in a single language or two languages at most.

The technical support is good.

Positive

In the past, we used IQ Bot from Automation Anywhere; however, its output fell short of UiPath Document Understanding's capabilities. This discrepancy stems from the sheer size of the model we currently employ in the collection. UiPath Document Understanding's effectiveness is attributable to this factor. Additionally, UiPath offers superior analytical reporting capabilities, whereas Automation Anywhere falters in this regard. With Automation Anywhere, we were required to create multiple models, whereas UiPath allows us to utilize a single model for a collection of invoices with similar structures.

The deployment itself was straightforward. However, the deployment of the automation may have been more complex. In terms of the Document Understanding skills required for deployment, the process is straightforward. It doesn't require a lot of effort and can be completed in a day or two. For an experienced or certified individual, the deployment can likely be completed within a few hours.

To complete the deployment, a team of three people is required to work together.

The return on investment is seen within the first year of using the solution. 

UiPath Document Understanding is priced high compared to its competition.

I would rate UiPath Document Understanding nine out of ten.

Our clients experience time to value after approximately four months of usage because, initially, it takes some time to become familiar with the models and begin to see results.

The number of people we have using the solution is specific to the AP team or data finance team. Currently, we have two teams working on the solution.

Document Understanding requires ongoing maintenance in the form of model retraining. In the event of any encryptions, we may need to provide validation to the user. Additionally, we need to ensure that our models are regularly retrained.

Organizations need to carefully evaluate the scope and requirements of their Document Understanding initiatives. While existing Document Understanding models have demonstrated capabilities in specific invoice formats, it is crucial to test their performance across a broader range of invoice types. I recommend conducting a pilot test using a sample of 20 diverse but similar invoices to assess the models' accuracy and applicability.

Our primary clients are in the pharmaceutical and hospitality sectors. We recently developed a process using UiPath Document Understanding called 'Medicaid automation' to automatically download invoices and structured data from legacy systems. We then built an ETL pipeline to further process this information. Additionally, we have experience automating contract downloads and parsing data from contracts, even for structured data sources.

Automating processes using structured data is straightforward. However, in many cases, we need to involve human workers because data extraction is not very accurate. Therefore, we need a solution to integrate human input and structured data into the automation pipeline to minimize manual intervention. Additionally, when accuracy requirements are very high, we can also set up a user interface. Conversely, for less stringent accuracy requirements, we can create a fully automated pipeline. This is the core idea behind using UiPath Document Understanding. We aim to automate processes for functions like finance, resource management, and revenue management.

I work primarily in the pharmaceutical and hospitality industries. Within these industries, specific domains have different usage requirements. For example, in the pharmaceutical industry, I work with finance teams, and their focus on unstructured data includes tasks like invoice processing. Revenue management teams might leverage unstructured data for contract management, extracting key details for further use. Both finance and revenue management teams should consider how generative AI technology can streamline their workflows. In my experience, I've implemented an agent capable of extracting data from compliance documents and providing structured responses to users. Other use cases involved HR-related document queries and automated responses. Within the hospitality sector, I've worked on customer success and revenue management projects. On the customer success side, unstructured data related to loyalty programs could be analyzed for insights. We also explored automating email generation and streamlining tasks related to standard operating procedures. Revenue management in hospitality often involves contract automation. For a large hospitality company, I worked on a project to extract data from B2B contracts stored in Salesforce, pushing that information directly into their financial system. It's important to note that while I used unstructured documents as a foundation for these projects, not all of them specifically employed UiPath.

Using UiPath Document Understanding, we have successfully processed invoice documents and contracts. We are now expanding to handle various additional contract types based on specific use cases. This could involve rebate management, B2B interactions, or other scenarios. Additionally, we can handle other document types, such as per-case order documents and various SOP documents (compliance and operational). Finally, we have also explored applying Document Understanding to marketing materials related to sales rep automation, where product information can be leveraged to generate responses.

We use UiPath Document Understanding for many formats. The format of documents depends on their type. Invoices and purchase orders, for example, are considered semi-structured. This means they contain a combination of elements, such as tables, key-value pairs, and line items, but these elements can exist in different templates and with some variation between vendors. Contracts, on the other hand, are largely unstructured. While they may contain structured elements like tables, they also often include running text and information that is difficult to categorize in a predefined format.

We can fully automate the process for 95 percent of the documents. The more high-risk financial documents may need human intervention.

AI capabilities significantly reduce development effort for handling encrypted data while simultaneously increasing its overall scope. This allows me to achieve what was previously impossible with conventional APIs, even in advanced tools like UiPath. While UiPath also utilizes a broad model for data extraction, they are now expanding towards generative AI. Consequently, we benefit from improved extraction quality and the ability to extract data in the desired structure, all with minimal development effort thanks to AI.

When human validation is required, it takes one to two minutes for a five-page document.

Previously, reviewing a difficult document like a contract could take around 30 minutes, while an easier document like an invoice took 10-15 minutes. After automation, processing invoices got significantly faster, taking less than half a minute. This is because the complexity of invoices is generally lower compared to contracts. For contracts, automation was reduced to around three minutes. In simpler cases, the processing time could even be reduced to as low as one to 15 seconds.

The significant reduction in processing time leads to a notable decrease in human errors.

Our clients can see the time to value within the first three months.

The most valuable feature is key-value pair and table extraction. While we previously relied on UiPath and Amazon APIs, we've transitioned to generative AI for its superior performance on unstructured data. However, this shift presents a challenge: while UiPath and Amazon provided consistent output and value, generative AI outputs can vary significantly across different documents. This means we still need logic-based parsing for tables, even though they often share similar formats.

The UiPath APIs lack reliable table parsing.

The accuracy of document extraction depends on the document's original format. For rich text documents, the accuracy is generally good. However, scanned documents like PDFs or images present a challenge and often yield lower accuracy. Another challenge arises when dealing with multiple documents in a single image. This scenario is common in invoice automation, where a single image might contain several invoices. Furthermore, processing files containing multiple document types, such as multiple invoices in one file, can be problematic. Currently, the system assumes each uploaded file represents a single document or invoice, which is not always the case. To address these challenges, I propose enhancing UiPath Document Understanding to analyze the entire document, not just individual pages. This would allow the system to identify individual invoices within a multi-page document and assign extracted data to the corresponding invoice.

I would like custom key value integration instead of generic key values for extraction.

The cost of UiPath Document Understanding has room for improvement.

I have been using UiPath Document Understanding and other IDP products/APIs for four years.

UiPath Document Understanding is generally considered a stable product. If we encounter issues when using it in the context of a complex backend process, the problem is likely not with UiPath itself but rather with the specific process design and the components involved in its development.

The high cost of adding bots hinders our ability to scale UiPath Document Understanding. 

The deployment takes around five days for my team to complete.

UiPath Document Understanding carries a premium price tag, but its current technological capabilities may not yet fully justify the cost.

I would rate UiPath Document Understanding five out of ten.

UiPath Document Understanding requires significant ongoing maintenance, especially when it integrates with screens or utilizes user interface automation. This is because changes to the website structure are highly likely to cause these integrations to break. Backend automation, on the other hand, typically requires less ongoing maintenance. However, it is still recommended to dedicate resources to monitor the solution approximately 50 percent of the time. This proactive approach helps ensure uninterrupted business processes even after a proper initial development phase.

For automating cloud-native platforms, scripting often proves to be a more suitable approach compared to tools like UiPath. However, when dealing with legacy systems, UiPath might offer a more effective solution.

We use Document Understanding to process invoices, purchase orders, and addresses. It extracts data from a scanned structured document and converts that in a structured manner to a spreadsheet. Predominantly, we use Document Understanding for payroll, procurement, invoice processing, and also in the finance department. Document Understanding has multiple models for extracting data from receipts. Departments have different use cases, but it's mostly used on the finance side to extract invoice data. 

The volume of documents varies from customer to customer. When everyone starts using the product, they typically process between 10,000 to 20,000 in the first year. Once you've achieved a stable environment, you might reach around 500,000 pages in the second or third year. It depends on the project and the customer's budget because pricing is based on the number of pages. 

We are not talking about 100 percent data automation end to end. If our customers work with hundreds of vendors, they deal with various templates. If a new vendor comes in, there is a possibility that the model may not identify that particular document. It's also possible that the upload quality isn't that great because of a bad scan, so there is always a channel for manual processing to handle exceptions. 

When you implement Document Understanding, we may start with 40 percent automated and 60 percent manual. As it progresses and matures, the percentage gradually improves. We may eventually achieve 80 percent fully automated processing with 10 percent manual so that exceptions can be handled with the help of human intervention.

Traditionally, the operations team has done many of these activities manually. A human takes information from the document and enters it into the system. There are many challenges inherent in performing these tasks manually. One is human error. Also, a department might receive documents in the middle of the night, and no one is around to process them. Document Understanding enables round-the-clock support and automatic processing

The implementation is fast compared to other solutions.  Documentation Understanding is more flexible because it has the artificial intelligence to understand new formats when they come in. It may read the information automatically. 

The amount of human validation depends on the type of input document. For example, let's say we are extracting data from a passport. We had to extract data from the passport. The solution can properly scan the documents. There are 192 countries with different passports. The bots are already trained with all the different types of passports. 

However, if the solution encounters a new format for receipts, invoices, etc., it may not identify it properly. During COVID, we had to process PCR tests from different diagnostic centers with different formats, so we created a model to figure out whether the person had negative results, but if a different format came in from a new diagnostics center, we might not have enough data to train the model. 

It will scan correctly without human intervention if it's a well-established document type, but if there isn't enough training for the model, a human needs to come into the picture. Also, if the data input is not properly scanned because of its model input and all those things, and the system cannot understand it, then human-in-the-loop comes in. 

The time needed for a human to validate a document depends on the number of fields and whether the file is a PDF form, invoice, etc. If you only need to validate the invoice number, you can complete that in one or two seconds, but it will take more time to validate all the line items in every field.  

Document Understanding has reduced our processing time by around 70 percent. In some cases, it may be 90 percent. It obviously takes more time for an employee to process a document with three or four pages and pull the data from various places. Using a solution with an OCR component like Document Understanding is much faster. It frees up employee time because we're not using resources to punch in data manually. We can use those employees to do other things that require more human intelligence.

The solution has reduced human error because somebody previously opened this document manually and typed whatever they saw on the screen. Now, what is happening is the data extraction is happening systematically. If things look fine and the confidence score is high, it inserts the data into the system. If the confidence score is low, it shows the screen to the user and asks them to correct it. Instead of merely typing the information, the user verifies what the solution has done. It's easily a 30 to 40 percent error reduction, and the operational efficiency is drastically increasing. 

Invoice processing is the most valuable feature. Most of my customers use Document Understanding for invoice processing. That's one of the most common use cases. Typically, each customer starts their RPA journey with the finance department because that's the area where you can see the most benefit. 

It can extract checkboxes, signatures, and printed documents. The extraction and conversion of printed letters is perfect. Document Understanding can also process handwriting and signatures using a machine learning model on the backend. UiPath's product team is constantly training this model continuously. Every two weeks, they are training it with a new set of data, so the model is constantly becoming more mature. I've seen a tremendous improvement since 2021.

The solution's machine learning model gives it the flexibility to accommodate documents with varying structures. Before document understanding came along, data extraction was done using template-based extraction tools. They created a machine-learning model that can be retrained for any number of templates. If you are actually not using machine learning, you will not explicitly identify fields like "Bill To," "Ship To" etc. You have to tell it the location where you want to find data. 

UiPath has already trained its machine-learning model, which has seen these types of invoices and trained the solution. You're building a better solution that requires less effort to implement because the product does a lot of that work for you. The deployment time is faster. It's more intelligent than conventional coding, which is just listing a set of rules. Everybody needs flexibility. It's not enough to have a solution to handle documents in a particular format. Whatever you do, it should have the intelligence to understand data in a semi-structured format even though things are returning in a different manner than the one that came before. 

Document Understanding's handwriting comprehension is improving, but it's still not as good as printed documents. Machine learning models, in general, are becoming mature, but it's still not to a point where I will give it five stars. I may give it a two or three. It is still not advanced enough to identify whatever handwritten content you give to it. It can process handwriting, but you need a human to validate it. With more training, it will become more automated. It will be better by 2025, but it is still not mature enough

Similarly, there is still room for improvement in reading printed documents. Ideally, if you have a model, Document Understanding should be able to extract every field from there. That's what customers expect. 

We have used Document Understanding for about six months.

I rate Document Understanding seven out of ten for stability. It has some room for improvement. 

I rate Document Understanding seven out of ten for scalability,

I rate UiPath support four out of 10. Their support has degraded badly. Presently, they are mainly focused on closing tickets. They have trouble communicating with our business users and end up closing the ticket because they don't understand what the issue is. It's a problem because the customer will lose interest in the product if they are not getting technical support. 

Neutral

UiPath can be deployed on the cloud or on-prem. The infrastructure costs of hosting it on-prem are high. We have done many cloud deployments, but I would say it's not that easy. Normally, we subscribe to the SaaS version of UiPath and configure it for the customer. UiPath has a cloud instance, which is a SaaS offering. I believe Document Understanding is hosted in Azure, but the customer can opt for AWS, Google, etc. There are no restrictions if customers want to put it on their private cloud.

An on-prem installation takes about two or three weeks depending on the complexity of the environment. Cloud installation is plug-and-play, so you can get it up and running in a day. They need to issue the purchase order for it, and we get the licenses. Once the customer has the license, they can log into the UiPath Cloud portal, and it will be activated. Within five days, they can start using Document Understanding. After that, you need to build the automations for your use case. The development time frame depends on the use case. It requires maintenance because you must train the model continuously as new templates come in.

The price is high, so it will take you about a year and a half or two years before you break even. 

Document Understanding's pricing is reasonable for developed markets because manual entry will be unable to match the cost of automatically processing one page. However, you can get labor for much cheaper in developing markets like India. It's not easy to sell Document Understanding in markets where you can get workers who will do this type of activity cheaply.  

I rate UiPath Document Understanding seven out of ten. It's an add-on for UiPath, so it isn't a standalone solution. If you already have a license for another third-party solution for RPA, you should consider whether it's beneficial to switch. 

The primary use case revolves around processing invoices. In Israel, where the solution is region-oriented, the invoices typically involve multiple languages within a single document and may also include various currencies. The capability of handling such diverse linguistic and currency elements is a notable strength of UiPath Document Understanding in this context. Through its implementation, our goal was to minimize manual tasks significantly and reduce the time required for invoice processing.

Up to this point, Document Understanding has been applied primarily to automate invoice processing in our implementations. For the customers for whom we have implemented it, the emphasis has predominantly been on invoice processing. This is because, within the customer's value chain, these processes are perceived to deliver the most significant value.

In terms of the types and volumes of documents processed with Document Understanding, the volumes are measured per page rather than per invoice. We typically handle a range of 50,000 to 100,000 pages. It's important to note that invoices, which occasionally consist of more than two or three pages, are encompassed within these volume metrics.

Typically, the document format comprises a header, a table, and often a summary, along with occasional total figures. This basic structure is effectively handled by Document Understanding, excelling in processing both headers and tables seamlessly.

Approximately seventy to eighty percent of our customers' organizational documents undergo complete and automatic processing.

The benefits are straightforward– it eliminates the need for physical forms on the table. This simplicity instills a high level of confidence in the model, and I foresee a promising future for it. It stands out as an excellent solution for companies, particularly those dealing with a substantial volume of invoices and vendors from diverse sources.

It has liberated time for other projects. Previously, we needed three to four people for validating invoices. Now, we have scaled down to one part-time person, who, for the most part, is engaged in other responsibilities. Invoicing tasks occupy only around five percent of their work time, handled intermittently.

The most valuable aspect is the AI training model, which distinguishes itself by offering a more transparent and controllable approach compared to other products on the market. Unlike some alternatives, this model allows precise retraining of machine learning instances. It provides visibility into the training process, enabling control and the option to retrain multiple times as necessary. In contrast to comparable products, this transparency and control contribute to enhancing the precision of the training model.

Forms AI performs admirably, posing as a strong competitor to Microsoft's PowerApps and other similar products in the market. It is straightforward and versatile, yet there is room for enhancement in certain design features that could improve user experience.

Document Understanding seamlessly integrates with other systems and applications within the environment it operates. Its integration capabilities extend beyond RPA modules, ensuring smooth and trouble-free connections with various components.

Human validation is required for Document Understanding at the beginning of Document automation journey, constituting around thirty percent of the overall process, while the tool handles the remaining seventy percent and document straight through processing improver further with model retraining. Notably, the retraining feature is a crucial and valuable aspect of the platform. This feature allows for retraining based on the validation actions performed by human validators. This is particularly significant because it enables refinement of the model in cases where documents are validated with low confidence. Some of the platforms lack the capability to provide confidence levels for field and data recognition, making this retraining feature a valuable asset for businesses seeking precision and efficiency in document processing. The human validation process for each document typically takes only a couple of seconds. The validation requirements are easily identifiable, allowing you to point to the specific area. Typically, pointing to it triggers a quick refocus of recognition to a different part, making the validation process efficient and straightforward.

The average handle time before implementing Document Understanding was approximately between three to five minutes, but after automation, it has significantly reduced to less than a minute, possibly even just a couple of seconds. This improvement covers the entire process, including validation, data exchange, mailing approvals, and more, all seamlessly happening in the background. Beyond the time savings, the automation also substantially reduces rework caused by human errors, enhancing the overall efficiency and accuracy of the process. As per the customer, errors do occur at times, and the associated risk is considerably high. However, the implementation of Document Understanding effectively mitigates this risk, eliminating the potential for errors.

I wish to have more pre-trained modules available in various languages. For instance, while Document Understanding currently supports Hebrew for Israel, I would appreciate the addition of pre-trained modules specifically tailored for different Hebrew-related forms. This enhancement could prove to be quite beneficial.

I have been working with it for three months.

The system is highly stable, especially since it operates on the cloud. We haven't encountered any disruptions or issues.

When discussing Document Understanding and RPA processes, it's essential to highlight that it's a scalable solution on the cloud. The scalability it offers is truly exceptional, making it arguably the best in the market.

The technical support is outstanding. In Israel, we have a local UiPath office, and they are incredibly helpful. Their responsiveness is remarkable, and if there's ever a need for assistance, they promptly provide valuable support. I would rate it nine out of ten.

Positive

The initial setup falls in the middle ground – not overly complex but not entirely straightforward either. It requires an understanding of how to retrain the model and fine-tune both the OCR and the application.

Deployment time is a matter of minutes. The deployment process is straightforward as it involves a cloud solution. You order the environment, set up both the robotic and Document Understanding environments, and start working. It's a simple and quick process. Typically, the deployment involves one representative from our team and relevant subject matter experts from the customer's side. These experts are individuals directly engaged in the process, and often a reinsurance manager, functioning as a project manager, is crucial from the customer's side. It is imperative to have a subject matter expert from the customer's side because our team usually lacks visibility into their business processes and requirements.

Maintenance typically involves one person responsible for document validation. The specifics may vary based on the document type; for instance, if it's invoices, it's generally handled by a single person specializing in invoice processing. While I would assume similar patterns for other platforms, variations might occur with different document types, requiring different subject matter experts for each form. However, from the technical side, it usually entails the responsibility of one person.

In terms of Return on Investment, while we haven't quantified it precisely, the notable reduction in personnel from three or four full-time roles to one person handling the task part-time signifies a significant cost avoidance. Instead of letting people go, the approach involves reallocating them to other tasks, essentially avoiding around ninety-five percent of the previous budget dedicated to this particular process. The benefits in terms of cost-effectiveness and time efficiency are substantial. In the context of time to value, I'd estimate around two months to establish a production process, yielding impressive results ranging from seventy to eighty percent.

I think this timeframe needs to be considered with the multitude of invoices and vendors involved. We're dealing with processing invoices from over two thousand different vendors, spanning two different languages, including instances where both languages are mixed within a single invoice. The complexity is heightened by the inclusion of both right-to-left and left-to-right languages. Despite these intricate challenges, achieving the high complexity production process within two months is not only sufficient but also a commendable outcome.

For those interested, I would recommend undergoing a POC to truly experience and be pleasantly surprised by the outcomes within a couple of days. In an overall comparison with other solutions in the local market, I would confidently rate this as a robust nine out of ten.

I use the solution in different ways. There are different ways to deal with the documents and extract data using OCRs, et cetera. I focus on document understanding with machine learning capabilities baked into it. 

I have worked on different types of documents. Both structured and unstructured documents as well as purchase orders, invoices, and time sheets as well. Depending on the type of the document or form, the solution would be different. If we have a structured document to retrieve the known information, then that would be easy. We would just use regular methods and simple extraction methods to get the data. However, if the documents are unstructured and the formats of those documents are different, that's where we would be using understanding and machine learning capabilities.

I've used it in healthcare, finance, and investments.

We're dealing with multiple vendors right now. When we deal with multiple vendors, each vendor has different structures for documents, and some of them provide data within papers while some of them provide just data as paragraphs. So for each of those types of documents, we have to extend the data based on the need.

How much the process is automated depends on the use case. It depends on the scenarios of the different types of formats of users. Sometimes there are different functionalities that we have to use within UiPath. If we're using Action Center, we probably would be able to automate almost all documents and that's where we would need users' input to validate the right information. In any case, we would be able to automate the majority of the documents since the bot would be able to expand the data. In such cases, the process might be longer. That's where we have to spend more time. However, if the documents are structured, it would be very easy for us to identify the data in those documents and then build the workflows.

The ability of UiPath to handle variant document formats, including handwriting is decent. Extracting the data is fine. However, it depends on what solution you implement and how much time you are ready to spend to implement that solution. If we have a plan to involve the Action Center within the solution, then that's where we would need a few inputs from the users to make sure that the automation is working fine, and that's where you would be able to achieve the majority of the success rate. However, if it is something that we just want to automate and we don't want to involve humans in it, then that's where we might result in a few exceptions as the data might not be right. That's where we would face some challenges. 

The machine learning capabilities have been quite fine. We've been able to digitize and classify documents and use them in our processes. That said, when compared to AI fabric, that's where we need to spend more time creating our own packages for it and then deploying those packages into the workflows. We need to make sure that we have a handle on all the documents.

I have found that 70% to 80% human validation is needed if we are trying to deal with sensitive data or if we are trying to deal with some confidential data. In those cases, we need to make sure that all the data is right. So as long as the document is structured and is well defined, and well-formatted, we might leave it 100% to automation. If any of these details are confidential or if any of these details require evaluation, then we will need user interaction. 

The validation process can be pretty quick. A code document doesn't take much time. It depends on the data. In any use case, I need to extract more than ten or 12 fields. If we're dealing with that number of fields, I'd estimate we need between two to 22 minutes.

The Average Handle Time, the AHT, depends on the cases. If there's no human involvement, then it would definitely take less time. If there's human involvement, the product could at least reduce the effort. The human involvement may drop from 20 to 30 steps down to one and they are just needed for validation. That scenario shows UiPath as a great time saver. Initially, we used to take 15 to 20 minutes to work on a document. Now, with automation, it might only take two to three minutes. It saves 65% to 75% in terms of time. 

There's still a chance of human error or tasks taking a few minutes as users would need to input some data into the document in the Action Center. That said, there is definitely a reduction and less of a chance of human errors over there. 

It's helped us free up time for other staff projects. That's the intention of implementing automation. Users can reduce their time on tedious tasks and focus on more important business needs. 

Document understanding works fine, however, it depends on what information you are providing it with. If the data is right, the data is good, however, in cases where the data is not right, it gets a bit difficult.

They could work on the digitizing and classification of documents. That would play a major role in document understanding since that's where we need to make sure that bots are able to extract data from multiple formats or multiple structures of the documents. The better they get at data extraction, the better we can automate. 

I've used the solution for two and a half years. 

I have worked on different use cases. There was one use case where I just worked on a similar type of document that had data entries for more than 300 to 400 users. I have worked with more than 400 to 500 documents of different formats and different sources. That's where I had to use machine learning packages. Right now, we are working on documents, with set formats and different structures, however, the volume of the documents is about 400 to 500.

They have a good technical team that supports the needs of the developers. 

Positive

It could actually take one year if you consider the effort which we had put into building that solution as well. We have a few developers working on it. We wouldn't see a return on investment within eight to ten months as we would just be starting with building the processes. 

We have witnessed some ROI. For example, it reduces review work from 60% to 70%. That's where the full-time employees would definitely save their time and can then focus on much more important business needs, which could help them get more projects or increase revenue as well. That's where you would see the most ROI. 

It definitely reduces the hours of work. The bots have the potential to also cover offline hours. 

While it depends on the use cases, the document understanding is good. I'd rate it eight out of ten.

Our client has PDF invoices and we use the solution to extract the details from them. We are using it in finance and health care. We have about 16 templates that we process now. The data is in semi-structured format and we mostly process things like signatures and tables. Out of the 16 templates, about 12 are completely processed automatically.

It has helped us automate finance statements and invoice billings.

Another benefit is that it has mostly helped reduce human error. We have a criteria of 75 percent matching. Out of 10 PDFs we have been getting eight PDFs with at least 75 percent matches. It has also helped free up staff time.

The taxonomy and Validation Station are among the most helpful features for us. If anything is extracted incorrectly, we can manually extract it there.

And we have included the AI Center for our customers to interact with PDFs to be extracted. Based on the approval or rejection feature, our customer can determine which kinds of PDFs they can automate.

I also like the table extraction feature. UiPath is very good with structured data.

Handwriting is more complex. We have not been able to get handwritten signatures correctly extracted in different languages. Our customer is in Dubai, and the solution cannot accurately process signatures in the local language. But it is a great tool for handling structured and semi-structured formats.

Another of the disadvantages is that we cannot include another tool. For example, with ABBYY extraction, we can integrate the process with any other product. We can integrate Document Understanding using JSON templates, but it is a bit of a complex model to extract the data from the JSON.

There is also room for improvement in long-running table extraction. If a table continues for more than 10 pages, in some cases, we have observed that it only extracts six or seven pages and skips the last pages.

I have been using UiPath for about 10 years.

Overall, the product is stable.

In our case, the use of Document Understanding is restricted to a particular group of users, around six or seven people.

The technical support from UiPath has been pretty good in the last year. It has been a very good experience. 

We used Azure DevOps for the deployment and we faced some issues regarding the deployment with UiPath and Orchestrator. We had a very good response from the UiPath technical team.

There is some room for them to improve the speed of the response because we often used to get late responses. But the resolutions are good.

Positive

We were using ABBYY, but it is more like a developer's tool with everything a developer needs for extracting fields. But we can train and retrain Document Understanding. In that way, I feel it's a better tool.

The pricing is reasonable.

As for additional costs, the solution is based on OCR, and sometimes the OCR cap is exceeded. It's not a major cost. Per month, we will have two or three scenarios like that. With ABBYY, once the cap was reached, we had to wait until the next day to use it again.

We did not evaluate other solutions. Using Document Understanding was a requirement from the client's side.

In terms of human validation for Document Understanding output, we have a limit of 75 percent correct scenarios. If it is below 75 percent, the user will be notified.

The solution doesn't require any maintenance unless the client requires more fields to be extracted. Only then are there changes that I need to make.

My advice is that if you are starting to learn about Document Understanding, you need to learn more about the taxonomy and what fields you are extracting. You need to have clarity on which position you are extracting, as it mostly depends on the position.

I use the tool for a couple of my client projects. My clients receive physical mail and may need to scan data to run processes like automation on it. Another use case is document classification. The solution helps with processes like classification, data extraction, and automation. 

UiPath removes manual processes and reduces human dependency. It takes a lot of effort to go through each physical mail or email, extract the data and transfer it to Excel. However, the solution automates the process and works 24/7. 

It gives a complete package to process and understand documents. The valuable features include taxonomy modification, classification, workstation, etc. There are out-of-the-box features like ML models which you can custom build.  

We have saved time with UiPath Document Understanding. We have seen a 50 percent improvement in scanning processes. Compared to humans, the tool runs 24/7. The human error rate has also been reduced. Our human error rate is five percent now compared to the previous 15 percent. 

UiPath Document Understanding has also freed up our staff who now spend more time on critical tasks. 

There is room for improvement in UiPath Document Understanding's pricing. It is expensive for small clients. Currently, there is a big gap between the basic package and the 200,000 packages. There is no package in the middle for small agencies. 

I have been working with the solution for more than five years. I started to work on the product when it was still under development. 

I have not encountered any performance issues. 

The exact number of documents processed per client varies. However, it ranges between 1000-3000 per week. The documents processed are very less. We process 10-15 documents daily. 

UiPath Document Understanding's support is always ready and helpful. 

Positive

UiPath Document Understanding was easy to implement and put into production. The timeline can change when you create your ML model. 

We have seen ROI with UiPath Document Understanding. 

The document format is mostly PDF and can be structured or semi-structured documents. We have not dealt with handwritten documents. Our real-time use case is for structured documents like emails and invoices. Most of the client documents go through without any errors. However, there is a five percent failure rate that needs to be considered since the document may contain unexpected data. 90 percent of documents go through it. 

The solution handles signature-based documents. We are still working on that prototype. We faced issues with seals. It differs from department to department and state to state.

The tool's AL and ML features work fine for us. We leverage these features for driving licenses. AL and ML keep a check on document generation. UiPath Document Understanding has come up with an API-based document understanding model which we will leverage soon. 

We implement human validation when we use anything new so that everything works as expected. 

I would rate UiPath Document Understanding an eight out of ten.