While there’s strong growth robotic process automation (RPA) adoption across digital services providers (DSPs), and tangible value delivered, we also see hurdles preventing large-scale enterprise-level adoption.
RPA is a low-code, low-cost option for enterprises to automate high-volume manual processes, delivering cost, efficiency, accuracy and transparency. But, often, after automating the quick wins and the straightforward use cases (e.g. order provisioning, assurance, pre-bill checks) many firms struggle to find further RPA use cases that can continue to deliver the 40-60% savings that the quick wins did. Analysts have called this out as well; Forrester reported in a recent article that “RPA is not a sustainable market in its current form”.
Other reasons include the eventual sunset of fragmented legacy systems, improved self-service interfaces powered by maturing natural language processing (NLP) capabilities, better data integration driven by increasingly sophisticated cloud platforms etc.
In a recent survey by HfS, 72% of telecom buyers responded that intelligent automation that includes RPA were critical to operations (along with autonomics, cognitive automation and analytics). What remains a problem is how to identify the right intelligent automation use cases.
As firms conquer their quick-win use cases, while the need for RPA seemingly diminishes, there are actually things in the way of helping it deliver value. The decision-making criteria for operational activities may not be well documented, consistent or even have an explainable basis preventing it from handling levels of non-algorithmic decision-making or from processing semi-structured data.
We have identified four emerging technologies that can be added to the RPA hopper, to help build out and deliver high-value use cases across the business:
Omnichannel integration for upstream integration
Increasing cost pressures mean that the number of agents continues to be a bottleneck in providing immediate customer service, with a Forrester report highlighting: “Firms are underserving opportunities to enhance customer experience.” Can RPA help?
While RPA has been used at contact centers for front-office automation, there’s still quite a bit of manual customer interaction needed for routine requests. Many RPA processes also need to be coordinated with back-office functions. Additionally, the front-office bots may not work in digital channels preferred by millennial customers such as text messaging, Facebook messenger etc.
However, firms can use RPA get quick solutions to routine requests so that agent bandwidth can be reprioritized to non-routine, higher-value tasks.
Chatbot technologies have been maturing over the last few years and can seamlessly handle traditional customer interaction channels such as voice and website in addition to emerging digital channels such as text messaging, social media and chat. We believe that using RPA as the glue for these chatbots to integrate with the upstream and downstream system can dramatically increase the variety in nature of customer-pleasing use cases.
A recent use case we had implemented demonstrates this – traditionally, when a home internet subscriber faces problems with their modem and thinks their device is faulty, they would either bring it to the nearest store or call a support center. In many instances, it would not have been possible to troubleshoot the problem in real time.
We have implemented a chatbot that can engage with the customer over a variety of channels including voice or text messaging. The chatbot takes customer inputs, and upon understanding the customer’s intent, triggers RPA bots to diagnose the problem, apply fixes, or send the diagnostic report to the service provider for action, depending on the situation. Thereby, a better customer experience comes from preventing any unwanted store visits – nearly 50% of current store visits do not have a faulty device.
Semi-structured data processing
RPA is typically restricted to use cases that depend on the rule-based processing of structured data. However, most processes that a typical operations team would need to handle contain semi-structured data such as invoices, purchase orders, contracts and email notifications. Forrester refers to this capability as “stage 2” or “enhanced digitization”. It considerably increases both the scope of use cases that could be automated and the potential ROI.
Consider any large enterprise – the number of vendor contracts, purchase orders (POs) and invoices received against those POs would make it prohibitively expensive to manually check every single invoice, match it to a PO, and check the respective contract to see if all the requisite terms and conditions have been complied with. Automating this scenario with RPA is not feasible due to the variable nature of formats of invoices and POs.
But, some of the leading RPA product vendors have made considerable strides in handling and process semi-structured data using a learning model. The bot will need to be trained on several possible variations of input data formats via an operations-friendly workflow-driven user interface. The bot will attempt to identify key data elements for any new format, and if it’s unable to do so within a preconfigured accuracy constraint, it will require a human operator to teach the new format. Over the time, accuracy rates will continue to increase as the bot learns new formats. In this scenario, it would be possible to target almost 100% of invoice processing via automation.
Integrate with smart decisions based on custom machine learning models
Machine learning, a subset of artificial intelligence adds intelligence to RPA bots. Machine learning models capture human knowledge as model parameters and then enable the RPA bots to take decisions like humans. It elevates RPA to “intelligent” RPA meaning it can address a larger variety of use cases. The multiple scenarios such as support from experts, handling of unstructured data, reviewing of responses, etc.
As an example, we added a machine learning model to enhance bot capabilities in the service provisioning and order entry space for a tier 1 service provider in the US. The existing process could not be automated entirely with RPA bots. Traditionally, an RPA bot would be paused till an expert is present to provide a decision where human judgement is needed. We created a machine learning model that could learn and eventually remove the need for human judgment-based decision making.
With help from the process expert team and machine learning experts, we collected decision metrics for the previous few months and analyzed them to build a text-based intelligent-decision-engine. As the response data was unstructured, we applied NLP and natural language understanding (NLU) techniques to data analysis and model building. We trained the prediction model with earlier responses dating a back few months, and then connected to the RPA bot. Now, when the RPA bot retrieves response data from the commissioning server, it calls the machine learning model via an API and proceeds in line with the model’s output. We removed the dependency on human judgment-driven decision-making, gaining an initial decision accuracy greater than 93%, something that will constantly improve with retraining.
RPA can be engineered to be the glue for not just legacy applications, but also to enable selective, pain-free adoption of emerging technologies. Good design while implementing RPA can help set up a platform to build future predictive analytics and machine learning models. Many leading RPA product vendors provide the capability to build rich audit logs and process metrics. You can’t get this level of visibility with a manually executed process and you can use the data to gain previously unconceived of capabilities. However, building out these rich data stores are design decisions that must be made consciously and mindfully to maximize RPA’s value.