This is the 3rd and final part of a blog on Intelligent Process Automation (IPA), written by our CEO and Chief Automation Officer Dr Steve Sheppard.
Part 1 focussed on setting the right expectation for automation across the organisation, finding good opportunities and then prioritising. Part 2 covered intelligent design, implementation, delivery, optimisation and maintenance. This part 3 covers the use of artificial intelligence and related technologies within process automation.
As a reminder, Intelligent Process Automation brings together the fields of Robotic Process Automation (RPA) and Artificial Intelligence (AI), encouraging the use of AI technologies such as natural language processing and machine learning within process automation. ‘Hyper-automation’ (Gartner) and ‘Digital Process Automation’ (Forrester) are other similar terms used to describe the combination of RPA with other technologies, such as AI, to extend the scope and scale of automation that can be achieved.
Many organisations have endeavoured to use RPA within their business to automate processes but with mixed levels of success. A significant number have not achieved the scale of benefit expected, or the RPA industry promotes as being achievable. This blog series highlights some of the key challenges that occur and provides a few pointers on how to overcome them.
Note, RPA isn’t the only technology for process automation, it can be achieved through BPM platforms, low-code platforms, traditional middleware and other technology solutions. Although there is a focus on intelligent RPA within these blogs, many of the principals covered are also appropriate to these other technologies.
If you have any questions, feedback or want to understand more about how Combined Intelligence can help you on your intelligent process automation journey then please get in touch via firstname.lastname@example.org.
Artificial Intelligence in Intelligent Process Automation
Blogs 1 and 2 focussed on how to perform process automation intelligently. This blog focusses on how to utilise artificial intelligence within process automation.
Artificial Intelligence (AI) is a broad topic, there are many different types of AI and an ever-increasing list of their applications. To simplify, we can group these into four main AI pillars:
- Learning – Machine Learning (and Deep Learning)
- Understanding – Expert Systems, Knowledge bases, Rules engines and Graphs
- Reasoning – Planning, Scheduling and Optimisation
- Communication – Natural Language Processing, Speech, Vision, Hearing
There are many opportunities to utilise each of these capabilities within an intelligent automation solution. In fact, many automation platforms are utilising AI within their platforms already, for example, in how they use computer vision to interact with Citrix based applications, in how intelligent environment monitoring tools can dynamically manage the environment and react to issues, and in enhanced functionality to intelligently handle unstructured data (more on this below).
These platform providers are also beginning to deliver cloud hosted AI toolkits to enable the custom use of AI (for example, machine learning) within the platform itself (not without licensing implications though!) and to allow automations to call out to external AI capabilities e.g. from Microsoft, Google, Amazon etc when needed.
Below I go through a set of example opportunities for combining AI with automation. This is by no means an exhaustive list, but it endeavours to highlight a diverse range of differing examples some of which are automation utilising AI and others are AI solutions that include automation.
Communication and Understanding (Part 1)
The way people interact with technology has evolved and will continue to evolve rapidly. We are moving beyond keyboards and mice, through touchscreens to voice interaction and whatever the future holds. Conversational AI approaches such as Natural Language Processing are enabling technology to interpret the spoken word and speech generation is enabling these technologies to respond in a human-like way.
AI capabilities such as Natural Language Understanding (NLU) can understand what is being said (to an ever-increasing extent) and in turn provide reasonable (in many cases) responses. This ability, demonstrated by virtual assistants such as Siri, Alexa and Hey Google, is rapidly evolving and appearing to provide more and more intelligent responses (note, most experts wouldn’t consider this currently to be real intelligence but I won’t get into that debate now!).
Chatbots are increasingly being used to automate text-based interactions. Often these have been ‘trained’ to be able to deliver subject area information or to ask an appropriate series of questions to guide the user (often a customer) to the information they need or to take input from the user in order to kick off an action.
A common theme with these communication channels is the use of automation to dynamically provide information and to process data that has been gathered. Much of this needs to be delivered in real time, requiring data level or API integration rather than, for example, UI automation. Post processing, once a complete set of data has been gathered, that often does not need to be real time can be done through various automation options including, for example, unattended robots.
Natural Language Understanding can also be used when processing free form text, for example from documents or emails. This is increasingly allowing sentiment (for example detecting a complaint), key data and meaning to be extracted from free form text which has traditionally been a blocker for automation (i.e. requiring a person to interpret). Automation is still better at processing structured data but advances in NLP/U are enabling better automated handling of unstructured data. Although it’s still better to get, for example, customers to fill in a structured webform rather than receiving unstructured emails.
Intelligent Character Recognition and Computer Vision combined with an understanding of context is even enhancing how technology can process printed/handwritten text, images and video. This capability can also be taught to do a better job with people initially fixing exceptions and the system learning what to do next time it sees this type of exception. AI based Computer Vision actually forms a core capability within some RPA platforms as it’s used to extract information and guide interaction with Citrix type environments where user interfaces are running remotely and only accessible to the UI automation via images/video rather than locally.
A key principal to consider, from an automation perspective, is how to design and implement automations that can process data irrespective of the channel or method it was received. Channel specific workflows can transform inputs into a common structure that are then processed consistently by a shared workflow leading to consistent outcomes. A common failing with multi-channel implementations is to end up with different end to end process automations for each channel, increasing the implementation time, creating additional maintenance demands and potentially leading to undesirable channel dependent variations in outcomes. Ultimately automation should support the principal of omni-channel customer service i.e. the same experience and outcomes irrespective of channel combined with the ability for customer engagement to occur consistently across multiple channels even for the same service request.
Understanding (Part 2) and Learning
The previous section looked at understanding in relation to communication. AI based understanding can also play an increasing part within other stages of end to end business processes.
Automation has traditionally encapsulated simple business rules through workflow logic. AI provides the ability to encapsulate more complex and fuzzy rules. Options also exist to use the history of human decisions for the system to learn to make its own decisions. Care must be taken with these approaches though as poorly implemented solutions can generate unexpected results when presented with input data not seen before. There is also the issue around what is called Explainable AI, i.e. the need in many business scenarios to be able to explain how an AI component has made a decision. Traditional machine learning solutions have been black boxes with little understanding of what’s happening within the box to explain why certain outcomes are produced. Machine learning solutions can also suffer from bias unless care is taken.
Increasingly there is the potential for machine learning (including deep learning) to be able to replicate the decisions (which may be subjective) that people have traditionally made. These machines also have the potential to factor in more data than a person would be able to, to be more consistent and not impacted by human limitations (tiredness, distractions etc). But they are typically only as good as the information they have been provided, how they have been created and subject to ethical issues as well as bias and a need for decisions to be explainable.
Reasoning, covering AI based planning, scheduling and optimisation is a more complex form of AI typically used in scenarios such as controlling autonomous vehicles. It involves AI instructing a series of actions that produced a desired outcome.
AI is being used within sectors such as telecommunications, finance and IT for monitoring purposes. For example, AI can be used to monitor the performance of a telecommunications network, detecting anomalies and then responding to these by automatically reconfiguring the network to cope with the anomaly. AI is also being used to help predict failures and to be proactive rather than reactive, initiating automations or maintenance to solve the predicted problem before it occurs. This approach can potentially be applied to monitor the performance of the digital/automation workforce. Large automation deployments can end up with 100s of worker running 100s of different processes. Being able to intelligently optimise performance and react to issues could help optimise operational costs and minimise costs due to failures.
Analytics and AI can be used for other forms of prediction, for example, decided on next best action when interacting with a customer or as part of a decision making processes which uses a wide set of data to decide what to do next to achieve the best long term outcome.
These uses of AI can assist within automated processes but also as the initiator of an automated processes.
Using AI with Automation
Many of the automation platform vendors have invested in embedding AI capabilities within their platform. This ranges from core capabilities such as computer vision (supporting UI automation), through extension capabilities such as ICR and text understanding, to the ability to utilise bespoke machine learning solutions within automations. Some platform providers are making their own machine learning training and operational environments available through their cloud platforms. Others are providing APIs to allow automations to interact with external environments provided by organisations such as Microsoft, Amazon and Google (amongst others).
There are pros and cons of each approach (get in touch if you want to discuss). I would also highlight the need for the right skills, developing AI solutions is a different skill to implementing automations, so having a team with this diversity of talent will become increasingly more important (we can of course help!). Both also have their challenges with regards to ongoing operation and maintenance, change, evolution and re-training will be needed. All of which should be considered from the start in the design and implementation of the intelligent process automation environment, approach and implementation.
You can read more about these challenges and how to overcome them in our blog on he benefits and pitfalls of intelligent process automation which you’ll find here.
That’s the end though of this introductory three part blog on intelligent process automation. I hope you found it interesting. If you want to understand more about any of the aspects covered by this blog or want to provide feedback please contact me via email@example.com. Please also subscribe below to our newsletter so that you can receive our latest news, blogs and articles direct to your inbox. Or alternatively follow us on LinkedIn.
Good luck on your automation journey! While on this topic and before you go elsewhere, you might want to read our thoughts about the Automation Journey and how we can help your organisation accelerate through this journey.
Thanks for reading.