It seems that some of the luster might be wearing off Robotic Process Automation (RPA). It’s to be expected; every new technology goes through a hype cycle where it’s initially going to solve every problem… then reality sets in and adopters begin to see what is real and what is hype. A recent survey by Pegasystems revealed that it takes an average of 18 months to deploy bots into production and only 39 percent are deployed on schedule. Once deployed, 87% of bot users experience some form of bot breakage or failure. If you hired a team of humans who delivered similar results, I’m not sure they’d have a job for long.
So, what’s going on here? Why do these RPA projects take so long to deploy and become ‘fragile’ in production? The answer is simple. Automated processes lack the agility of human intelligence that allows them to react to new or unexpected situations and course correct to keep the process on track. The ‘unhappy path’ as we call it in the tech world. Engineering teams have been dealing with managing failure paths since, well, since the invention of the wheel. It is often the hardest part of creating any technical solution. The permutations of possible failures for a process of any complexity starts to quickly grow exponentially. It is almost impossible to ‘program’ how a bot handles these paths up front, even for an experienced engineering team. Asking your business staff to develop resilient bots using self-serve tools is almost impossible.
RPA vendors are hoping ML will solve the fragile bot problem. It won’t.
RPA vendors and luminaries are just muddying the waters by declaring that machine learning (ML) is the answer. Unfortunately even the best data science teams aren’t going to develop machine learning algorithms that will be able to react well the first time a critical file is missing for end of day processing. Machine learning is great at being shown thousands of historic scenarios and finding predictable patterns in the data, it’s not so great at reacting to brand new situations. That still requires human intelligence and logic.
The good news is that automation doesn’t mean people go away. We may become more efficient, and hopefully most of us will free up to solve more interesting problems, but people still need to be available to guide an automated process. The trick is plugging humans into the process in a way that empowers them to quickly react to process exceptions and to keep the business on track. More importantly, people need to be given the tools to quickly model how they react to problems so they don’t keep happening.
No-code AI to the rescue: designing models that emulate human logic can give your bot a ‘brain’.
Business teams already know how to work with data, handle exceptions and create new solutions that prevent problems from reoccurring. They do it manually every day by creating ‘models’ using tools like spreadsheets and other ad-hoc technologies. There is something very different about working with an interactive modeling tool like a spreadsheet vs. using a ‘process designer’ technology like RPA. In the modeling tool, business users can touch and feel the data, change calculations, see anticipated results and adjust logic as needed to deliver the desired outcome. In a process automation tool, the user would need to anticipate all of those possible outcomes up front and ‘code’ the bot to handle it.
The compromise is simple. You need both process automation and logic modeling tools. The operations teams need RPA bots that can automate the redundant and costly manual work that historically added little value, but the business still needs their modeling tools where they can deal with new information and modify how they react to it on the fly. More importantly, their models need to ‘evolve’ based on past experiences and begin to handle situations they’ve seen before automatically. This is where AI starts to add value – machine learning that can emulate the intelligence of prior human responses without intervention. Without the first step of creating models to make better decisions, the AI tools will have no one to learn from in order to improve decisions going forward. This paradox is one of the biggest misconceptions when people talk about AI. We need to do a better job of capturing our human intelligence before machines can effectively learn from us.
Emerging Intelligent Automation technologies can integrate human intelligence, AI, and automation.
At this point you might think that fixing the RPA fragility problem may require armies of staff working in spreadsheets to create models that inform bots. Clearly that’s not a workable solution. Manual spreadsheet work is often what you were trying to replace with your RPA initiative in the first place. Spreadsheets have been a powerful tool for the business for decades, but they’re really not cut out for digital transformation.
At the same time, we need to give the business modeling tools that allow them to inform their processes. If it’s not using spreadsheets, then what is the answer? Is it Python? Realistically, despite the ‘everyone will code’ mantra, writing code to model business decisions isn’t practical. There is too big of a learning curve and writing Python code simply isn’t as efficient as working with grids of data and declarative expressions like users do in spreadsheets.
A promising solution is coming from an emerging sector of technologies called Intelligent Automation. IA tools are like the next generation of ‘super spreadsheet’ that is designed to connect to the modern automated enterprise while still giving business users the power they need to interactively model business logic. These platforms include powerful data modeling tools, a business language, business rules and the ability to leverage machine learning directly in an integrated tool. More importantly, these IA models can be easily deployed as APIs that can integrate with RPA or other production systems in real-time.
The next generation of the digital enterprise isn’t as ‘robotic’ as many predict.
The next generation of the digital enterprise isn’t as ‘robotic’ as many predict. Good old-fashioned human intelligence will still be required for companies to be efficient and competitive in the foreseeable future. The biggest technology challenge we have around automation is not how to completely remove humans from the process, but rather to figure out how to intelligently plug them into processes of the future.
