18 Jul How to Get an 80% ROI over Your OCR Implementation
Posted at 04:07h in Blog
Any time we consider working with a new client to help with their automation efforts, there is one key thing that they always want to know:
What Is The Return On Investment Of Implementing OCR Automation In My Company?
When customers are building a business case to invest in purchasing or expanding an OCR initiative, determining their organization return on investment (ROI) is essential to securing funding.
Calculating ROI is an exercise that helps to identify and quantify the business value you gain from investing in OCR, but many organizations struggle to find an accurate methodology for calculating ROI.
This guide aims to provide a clear and tested framework to help your organization make informed decisions when investing in OCR and reap its benefits for many years to come! Before getting into the math, let's identify the critical components of the equation.
Conventional OCR providers have available two different engagement models, namely Perpetual Licencing and As A Service Licenses.
This last model has become more and more common as companies change their buying patterns to reduce entry costs.
A perpetual software license is a type of license that authorizes an organization to use the OCR technology indefinitely after a single fee.
Along with a perpetual software license, the vendor typically provides a technical support period of one to three years. During this initial period, the vendor also provides frequent software updates. However, updates may or may not be provided for free in perpetuity. Most of the time, a perpetual license requires the use of the software on-premises.
In comparison, software as a service (SaaS) is a software distribution model in which a vendor hosts or provides applications with a monthly or annually ongoing payment. The vendor typically provides support and updates included within the service fee with no extra cost to the customer.
Most of the time, SaaS licenses are available for cloud-based services. Regardless of the chosen engagement model, license fees depend on many variables related to the volume and intensity of use.
These Variables May Be Different From Provider To Provider, But They Typically Are Related To The Following Factors:
- The number of pages processed.
- The number of CPUs/Servers used.
- The number of revision stations.
- Plug-ins for different languages, table extraction, template building or API.
Choosing the right combination between all these variables is not often simple or straightforward. For example, understanding the number of CPUs needed strongly depends on the volume of documents processed and the pace of submissions, thus peaks have a significant impact on this variable.
The same applies to the number of revision stations, which strongly depend on the overall accuracy and pace of document submission.
To ensure that the license is optimized, a significant forecast effort is required that may not be possible to perform most of the time.
Thus, customers often rely on a vendor’s forecast that, as you might imagine, is done without the knowledge of the customer's operation and business reality.
Most OCR technologies don't work out of the box; they require a setup investment on the design and implementation of templates and extraction rules.
Templates are the base of most OCR technologies. A template defines a set of locations over a page where information can be retrieved for a particular document layout.
At least one template configuration for every single layout variation will be required.
Most technologies already have a set of tools to support the development of templates. Nevertheless, our benchmark points to an investment of 2 to 6 hours per template configuration.
So, imagine that you are processing invoices from 3,000 suppliers. That will require an investment of 6,000 to 18,000 hours in implementation services to have the system up and running, assuming, of course, that your document inbound is relatively structured in nature.
Extraction Rules / Machine Learning
In recent years, more and more OCR technologies are reducing their dependency on templates and using more generic data extraction methodologies.
These methodologies often leverage domain-specific extraction rules and the use of machine learning. Based on our benchmark, it is not clear whether this approach brings significant advantages over the traditional template approach since they often require the creation of a clean document data set to train the extraction engine.
This is a particularly daunting task since most data sets are polluted with errors that strongly impact the system performance when used as the basis of training.
Moreover, this approach will significantly decrease the visibility and control over what is happening on the extraction engine, often leaving you with a black box that is very hard to audit.
Our benchmark points to a cost between -10% and +20% over traditional template approaches.
OCR technologies act as an intermediary layer between systems; thus, to fully automate any process, the OCR technology must be fully deployed and integrated with existing systems.
Depending on the target systems, the integration effort might be simple or very complex. Therefore, you should consider the following:
Is your OCR technology already providing out of the box connectors to your target systems?
Is your OCR technology available via an API or integrated with third-party integrations services such as Zappier?
Are you creating highly couple integration on third-party systems, increasing switching costs?
Our benchmark points to an advantage for systems with clear and open API since they enable a high degree of transparency, reducing the dependency on third-party systems.
Often the availability of an open API allows internal teams to take ownership over the integration and take advantage of data capture across a wide range of inbound channels, such as web pages and mobile.
Your inbound changes over time, requiring adaptations of the data capture configurations, either by adding new templates or new extraction rules.
You will need continuous investment in monitoring system performance and in the development of adaptations to ensure a better accuracy level.
This cost may be high if you are using data capture across different inbound channels since information is often less structured and standard.
You should also account for updates and IT management costs to ensure uptime and security.
Our benchmark points to a clear advantage for SaaS platforms since they include software updates within their fee and cover most IT-related costs.
In most processes, the quality of data is critical. No customer wants to integrate incomplete or incorrect data.
Most OCR systems provide confidence levels alongside with the extracted data. These confidence levels are often based on validation rules or even historical data, acting as a proxy for the quality of extraction.
Nevertheless, these confidence levels are far from being an accurate SLA. Human-based validation is still needed, even if the system is predicting with 100% accuracy. This is perhaps the less obvious challenge behind OCR that is often ignored by both the customer and OCR vendors.
Customers often kick off an OCR initiative aiming at fully automating processes and end up still supporting a validation operation based on human intervention with a high ongoing cost.
This is mainly because OCR vendors don’t provide a clear and accurate SLA, therefore pushing the operational risk to their customers, even though that cost is often superior to the license investment.
Our benchmark points to the need for human intervention for every 15,400 documents per month. Thus, you should expect that, on average, a human can review between 80 to 100 documents per hour.
How To Calculate ROI?
With all these price drivers in mind, you would probably still find it daunting to understand the ROI behind your OCR implementation.
This is mainly due to the fact that most variables that impact cost are not easily predictable.
Nevertheless, for illustrating the ROI calculation and the several price drivers, we provide a conceptual case study for a popular use case that is accounts payable.
Let's assume the following scenario:
- 20,000 invoices processed per month (with an average of 3 pages per invoice).
- 1,000 suppliers (with different layouts).
The manual processing of this amount of documents would require between 12 and 14 FTEs with a cost that can range between $50,000 to $65,000 per year (assuming standard data entry rates).
Let's have a look at the ROI once you have implemented an OCR for the same process.
Here are some more assumptions (using SaaS as a benchmark):
- Price per document – $0.05 / doc
- Integration fees – $5,000.
- Template development and configuration – $30,000
- API connector – $250 / mo
- Human based revision – $8,500 / mo
Assuming a 4 years simulation, your final cost per document is $0.52 per document, thus providing a significant ROI over manual input.
Can We Still Increase The ROI?
At DocDigitizer, we designed our service to overcome the limitations of common OCR technologies, mainly by providing a simple and transparent engagement model focused on maximizing ROI and reducing our customer's risks.
To do this, we decrease the number of price drivers using the following approach:
- DocDigitizer charges a fee-based solely on the number of documents, independent of format, layout, or the number of pages.
- We don't require setup fees, configuration processes, or the development of templates. You can start using our service right-way with no entry cost.
- We ensure an accurate SLA, validating all data extracted. Removing the need for having human-based validation on the client-side.
- Our API is available for free.
Taking into consideration the previous example and assuming a 4-year simulation, your final cost per document is only $0.13 using DocDigitizer.
DocDigitizer can provide you with up to 80% ROI over traditional OCR engines, decrease your entry costs, speed up the implementation, and make your automation more agile and scalable.
Start now your digital transformation!