Creating bots is easy — scaling them is another matter

The field of Robotic Process Automation (RPA) has seen a major boom thanks to the use of AI tools that make it easier to streamline the development of software robots. At Transform 2019 this week, experts weighed in on what will be required to take RPA from a simple point solution to a robust digital factory. The goal is not so much to replace humans, but to find better ways to complement human workflows.

Telecom giant CenturyLink discovered that scaling and managing a bot workforce required a thoughtful approach. Brian Bond, consumer vice president at CenturyLink, said things started changing when they got up to around 100 bots. “After that, a lot of the initial bot developers were doing maintenance on existing bots. Something could change in Salesforce or another tool, and you have to maintain that,” he said.

RPA bots excel at cutting and pasting data across multiple applications. This is particularly important at CenturyLink, which, due to growth via acquisitions, had a mishmash of different applications for similar processes. “Call centers agents and field technicians would end up doing a lot of swivel chair copy and pasting activities,” Bond said.

RPA allows CenturyLink to automate many processes that span different apps quickly. This gives the IT team time to be more strategic in more significant digital transformation activities that might involve replacing disparate application across the company.

Improving the approval process

The initial bot efforts were easy to implement. In the early days, Bond’s team was able to spin up 50 bots in 60 days. However, they started running into various management problems as the number of bots grew. “This got us to think about the things we need to do from a management perspective,” Bond said. This included better resource planning and implementing a stage-gate process for bot creation and change. A stage-gate strategy divides a larger process into a series of approval points so that subject matter experts, finance, security teams, and management could all review new bots before they are unleashed into the organization.

Initially they were trying to manage this process through SharePoint, which got complicated when they were trying to build 25-30 bots in parallel. So, Bond’s team built their own proprietary system for managing the bot stage-gates. Now the appropriate expert is automatically notified when a bot requires approval. These experts can also explore the status of all bots under consideration from a comprehensive dashboard.

Bond said it was also important to communicate financial and performance feedback about the bots so that managers could make better decisions. This involved coming up with a set of scoring and logging standards that measured things like how many times a bot ran in a month, and the amount of time and money they saved over humans doing the same tasks. This has helped Bond’s team to gain funding back from other parts of the company.

Keeping the right talent in-house

CenturyLink also had to rethink its outsourcing strategy. In the early stage of the RPA initiative, it seemed easier to outsource many aspects of bot development, such as process analytics and solutions architecture. This allowed them to quickly build up the first set of bots.

But some aspects of these jobs involved a steep learning curve. It could take four to six months for an expert to get up to speed on the nuances of CenturyLink’s bot ecosystem. Then the contract would end, the expert would leave, and a new expert would have to start learning all over again. “It was critical to bring some of those roles in house,” Bond said.

This is especially true if you want to scale successfully. The first wave of RPA tools has focused on proving utility at a small scale, said PD Singh, vice president of AI at UiPath, an RPA vendor. Now RPA tools need to incorporate new capabilities for bot management as part of an integrated system that works in conjunction with ERP and CRM. “You need to build it integrated with other functional pieces in the organization,” Singh said.

For example, process analytics capabilities are required to figure out how much time and money companies are spending on a particular process. This can help prioritize which processes should be automated first to provide the most value. After a bot has been created, managers can see how much money or time it saves in practice. These same tools could also determine if there were other benefits, like improving the customer experience by reducing problem resolution times.

Automatically building bots

Going forward, Singh expects to see AI help to reduce the effort of creating new bots. UiPath plans to release a new feature in August that can automatically generate a basic bot file by observing a subject matter expert over time. The resulting file could be used by bot development experts to create a robust, production-ready bot in a fraction of the normal development time.

Down the road, UiPath will also be adding a feature that makes it possible to observe front line workers to identify common processes. For example, it might observe that 500 agents execute the same processes throughout the day. “Once you get those insights, you can do more and more there,” Singh said.

UiPath is also working on new capabilities to weave visual and document understanding into RPA apps. Visual understanding will make it easier to interpret image data as part of a process, such as identifying items as part of an automated checkout system like Amazon Go. Document understanding will make it easier to move data from documents such as shipping logs into the right fields in an ERP database. The company is also working on an AI Fabric that promises to make it easier to weave third-party AI components into RPA workflows.