Key Takeaways
Automating business processes has long been constrained by a communication gap: computers demand unambiguous, structured languages like Python, C++, and Java, which lack the back-and-forth flexibility humans use to clarify intent. The post cites a Formstack study of 280 companies finding managers spend roughly 8 hours a week on manual tasks, underscoring how much work still resists conventional automation. Because real-world processes are full of ambiguity and exceptions, rigid code struggles to capture them. The proposed answer is Natural Language Processing, letting people describe processes in plain language so machines can interpret and execute them the way a colleague would, without forcing business experts to think like programmers. The takeaway is that bridging the gap between human and machine language, rather than writing ever more code, is the path to automating the messy, exception-heavy work that fills most people’s days. See the Kognitos platform.
Despite 50 years of widespread computer usage, computer languages lack a toddler’s ability to use a back-and-forth approach to effective communication. And while computers have changed our lives in a great many ways, their requirement to use unambiguous structured language limits their effectiveness at automating business processes. A Formstack study of 280 companies showed that managers spend an average of 8 hours a week on manual tasks.
Computer Language Limitations in Automating Business Processes
Computers speak to computers just fine. Highly structured computer languages such as Python, C++, and Java enable seamless handoff of information between machines.
Humans can similarly easily communicate with other humans. This is true even when using computers as intermediaries. Email threads, for example, allow dialogue and negotiation. If you send an email requesting someone to do something that they don’t fully understand or agree with, they’ll email back with questions.
The trouble emerges when humans try to communicate with machines. All computer languages are geared toward unambiguous instructions that the computer interprets in only one rigid way. Close to 9,000 computer languages have been invented thus far and not one comes naturally to humans.
Every company utilizes business processes, and they want to improve efficiency by automating them. Robotic Process Automation (RPA) has emerged as the most common approach, and the market is on fire. According to a ResearchandMarkets report, the RPA market is expected to reach nearly $26B by 2027. A 2016 McKinsey Digital report says that RPA enables a first year ROI ranging between 30 to 200 percent.
But RPA’s dependency upon rigid computer languages limits its effectiveness when business processes are complex or evolving. As an example, even a small insurance business will process tens of thousands of medical insurance claims in a year. When humans do the work, they can easily apply the required business rules while processing the claims.When RPA-enabled computers do the same work, they typically fail when they encounter ambiguities such as missing information or incorrect entries, or novel scenarios that have not been anticipated.
RPA customers attempt to prevent these incidents by first bringing in consultants to discover and optimize their business processes before automating them. A January 2022 ITWeb article, Finding the sweet spot of human-centric RPA, says: “While RPA software can radically improve business processes, building an RPA capability requires significant investment of time, money and people.” Naturally, as the investment increases, the ROI declines.
This lack of easy machine to user communication creates additional costs over time as business processes change – which most inevitably do. Computers are unable to ask the questions required for them to learn how to keep up with changing requirements. Companies must bring in developers to adjust the automation to the new business process requirements. It’s no wonder that an EY survey found that 30% to 50% of RPA projects fail.
Electronic forms such as those used by tax agencies, or the Department of Motor Vehicles are meticulously programmed to allow only certain responses. When filling out a form, if I have a legitimate response unanticipated by the developers, I’m typically not able to submit my input.
As another example, consider the Multiple Listing Service (MLS) for realtors. A realtor needs to manually input data such as the property age, the lot size, etc. over time. If the MLS communicated like people, it could collect information directly from the seller, leaving the realtor more time to sell.
The Natural Language Processing (NLP) Solution
As a toddler begins to learn a few words, effective communication leaps exponentially even if the words aren’t completely correct. The ability for the communication process to handle ambiguity in turn enables the toddler to learn much more quickly.
The same holds true when we use natural language to allow machines to be more like humans. Many developers frown on natural language because it’s ambiguous. But it’s the ambiguity that makes NLP both more accessible and easier to use. The fact that the machine will disambiguate the instructions at run time results in a more robust automation which does not break on changes in the environment.
Natural automation through natural language processing is the antidote to RPA limitations. The system discovers any missing or incomprehensible information and reaches out to the appropriate person for answers, just as a human does, in response to natural instructions. Once the information is received, the system puts it in the right spot in the system of record and goes on to the next step in the business process.
Natural language acts as the common language between the human and the machine allowing both sides to negotiate the path forward when the machine is stuck, and also enabling the human to teach the machine without first getting trained in a computer language. Kognitos uses natural language instruction and exception handling to vastly simplify and reduce the cost of business process automation.
My daughter grew up to be a nuclear scientist, and she utilizes Python in her work. Despite her many years of developer training and experience, I look forward to entering a new era where she’ll again be able to speak with computers as naturally as she did as a toddler when requesting I get her a cookie.
How to Get Started Automating Business Processes
- Choose a well-defined, repetitive process for the first automation. A good first automation has clear inputs and outputs, well-defined decision rules, and occurs more than 50 times per month. Invoice approval routing, expense report validation, and status update notifications are commonly successful first automations.
- Document every step and decision rule in the current manual process. Walk through the process with the person who runs it today. Document every step, every decision, and every exception. Rules that cannot be documented cannot be automated. Processes where 'it depends' is the primary decision rule need clarification before automation.
- Configure the automation in a sandbox environment. Build the automation in a sandboxed environment using real but non-production data. Configure each step, test each decision rule, and set up exception routing before any production data is touched.
- Test with real historical transactions before go-live. Run 100 real historical transactions through the sandbox automation. Compare outputs to what the manual process produced. Any discrepancy requires rule investigation. A 100% match on historical data gives confidence before go-live.
- Deploy to production with monitoring and a go-back plan. Go live with the automation enabled but with a clear plan for reverting to manual if a critical error is discovered. Monitor the first week closely. Once the touchless rate is stable and exception handling is working correctly, remove the manual fallback.