AI, RPA, and Document Automation

How do AI and RPA relate to document automation? And could they be useful to you? Martin Srubar November 7, 2019
Hand drawn image of a toolbox with the label DOCUMENT AUTOMATION, with AI and RPA floating out of it.

In a previous blog post we talked about the complicated relationship between Artificial Intelligence (AI) and document automation. We concluded that, at the moment, better results can be achieved through document automation that doesn’t have deep learning at its core.

Robotic Process Automation (RPA) is another recent buzzword for technologies that have been around for a while, and help automate tasks that people perform using computers. The relationship of RPA and document automation is also complicated. At one end of the spectrum, document automation can be considered to be a highly specialised RPA for creation of documents; others might argue there is no relationship at all.

However, AI and RPA can play an important role outside of the core document automation engine, particularly at the time of solution implementation.

We often find that one of the major barriers to adoption of document automation is the implementation effort and associated cost. The implementation teams that have good command of AI and RPA – may they be tools tailor made by the document automation provider or well-configured off-the-shelf products or services – will become much more productive and accurate than the ones that don’t.

The image below shows what can be involved in an implementation of document automation software by organisations who need to maintain hundreds or thousands of templates.

Hand-drawn illustration showing the steps involved in automation implementation: taking stock, analysis, defining, implementation, testing, and go-live.

Basic document automation implementation stages.


The above illustration focuses on the automation of documents itself, and doesn’t include any work related to integration with other systems, setting up of the infrastructure, training, and others.

Let’s consider an example implementation process. In the first scenario below, implementation doesn’t use any AI or RPA; in the second scenario, it does.

Hand-drawn illustration showing the steps involved in automation implementation: taking stock, analysis, defining, implementation, testing, and go-live. These steps are compared for two types of implementations, without AI and RPA help, and with AI and RPA. The implementation process using AI and RPA is significantly shorter - by 50%. The analysis, implementation, and testing stages have been reduced.

Comparison of implementation duration with and without AI and RPA.


As you can see, automation implementation with the help of AI and RPA is significantly quicker. But let’s have a look at where and how the improvements have been realised.

Hand-drawn label: pink box containing the words Take Stock.

There is no improvement here. In this step we gather all materials – templates, documents, process description, business logic description – and verify that they are relevant to the intended future state of the process.

Hand-drawn label: red box containing the word Analyse.

This is where, in this example, AI analysis helped save at least 50% of the time needed. Machines can “read” through hundreds or thousands of documents quickly, and assess commonalities and differences. The human business analyst can then go straight to the content instead of having to search for it or worrying that anything was missed.

If the source information is well structured, AI can also help improve the analysis of business logic, processes, and data requirements.

Hand-drawn label: orange box containing the word Define.

Defining how the content will be reused, what the template-related business rules are, and what data templates need is still knowledge and skill that only humans have. That is until we find a way to download human brains into AI machines. No time saving here.

Hand-drawn label: yellow box containing the word Implement.

Major time savings can be achieve through the use of RPA. The RPA helps implement what is defined in the previous stage into templates. There is some time needed to set up and test RPA itself, which is included in the implementation time. This means that the more templates you have, the more beneficial the use of RPA can be.

Hand-drawn label: turquoise box containing the word Test.

Here we show about 60% reduction in testing time. RPA can help speed up generation of test samples and can run error evaluation or compare the actual output with the expected output.

Note: You may notice that the testing stage is shown as subsequent to the actual implementation of automation. In real life, testing is often incremental, taking place as each template is automated. We separated it in this way to show the different impact of AI and RPA on each of the stages.

Hand-drawn label: blue box containing the words Go-Live!.

This is often an organisational change management process. AI and RPA usually are not used and therefore no time saving is realised at this stage.

When should I use AI and RPA?

While AI and RPA have improved a lot recently, there is always time and effort required to set them up for each implementation. This means that the use of AI and RPA makes for a good business case with larger volumes of templates. The tipping point at which AI and RPA bring significant benefits depends on the size of templates, and their automation density (do your templates have a lot of static text that never changes, or are they highly dynamic?).

Hand-drawn line graph that shows an upwards trend of benefits from AI and RPA as the number of templates increases. The graph is divided into three sections. Section 1, at up to 500 templates, the graph area is red and the line approaches the x axis labeled with 0: AI and RPA use are not recommended. Section 2, between 500 and 1000 templates, the graph area is yellow, and at this point you can consider using AI and RPA. Section 3, 1000 templates and above, the graph area is green, and it is recommended that you use AI and RPA for implementation.

Whether or not to use AI and RPA depends largely on how many templates you’re automating.

As a general rule, you don’t really need to worry about AI or RPA unless you have more than 500 templates to automate in your project. Note that we have also worked on projects including over 1000 templates, where the use of AI could not be justified. However, the more templates you’re automating, the bigger the benefit of AI and RPA.

What this boils down to is the expertise of your implementation team. AI and RPA are just tools in our toolbox. There is no need to use them just because AI and RPA are the buzzwords of the day. However, if used well, AI and RPA can turn impossibly resource-intensive projects into tremendous wins.

About the Author

Martin Srubar

Senior Technology Evangelist

Martin’s engineering background and his passion for great products whether in physical or software form are complemented by his understanding of the ActiveDocs application and how it meets the requirements and fits the architecture of the company’s clients. Martin continues to engage with potential and existing customers, adding market intelligence and customer feedback into the company’s ongoing product development strategy.

Do you have any comments or feedback?

Learn More

Hire agreements and leasing documentation automated at a large truck and bus manufacturer.

Read more

An energy retailer communicates with 1.5 million clients using ActiveDocs.

Read more

Learn about a major implementation thoughtfully executed in Australia's court system.

Read more