People are excited about AI and hyperautomation, and for a good reason. AI has the potential to automate complex enterprise tasks that involve human thinking and related behavior.
AI-driven hyperautomation for the enterprise looks like the current state of self-driving cars. We’ve got Teslas that can drive people to places on demand, and Waymo is roaming the streets of San Francisco and Phoenix with no driver at all! It’s a good start, but more must be done before considering ourselves in a fully autonomous driving world.
Challenges include incomplete data map versions, differing and changing road conditions, driving cultures, obstructions and many other variables. The system is also not functioning across every road, city, and location, nor is it in the larger, congested cities. And it still requires human oversight in all cases.
It’s the same for enterprise automation. Some automation exists, but to have effective hyperautomation in the enterprise, there are a lot of things that must happen first. Specifically: a “learning phase” to ensure automation can adapt to the challenges of the enterprise. This includes thousands of processes across every type of system, each with nuanced policies and different teams with embedded knowledge of how tasks are completed.
By using AI to learn business processes carefully and applying the right kind of learning rigor, it is possible to accelerate complex enterprise processes using hyperautomation.
A Poster Child: Customer Support
Customer Support is a great example of a people-intensive enterprise process that could benefit from AI-driven hyperautomation. Deloitte reports that 80% of contact centers are actively engaging in some stage of AI deployment.
The Customer Support/Service world changed 18 months ago with the advent of GenAI. Chatbots are now radically more effective at solving problems and are cheaper to run and implement than ever before. As a result, all the incumbent customer service platform providers— Salesforce, Zendesk, ServiceNow, and so on—are adding GenAI to their core platform functionalityOpens a new window . Their bots will be exponentially more useful and powerful because they’re based on the data in those systems and can learn from them.
However–what about all the things that cannot be deflected? The ones that still require an agent! For non-deflected Customer Support, the hyperautomation opportunity is bigger. Each customer transaction is a one-off and high risk by definition—because it wasn’t simple enough to automate!
For example, a customer support agent dealing with a product-shipment problem needs to navigate various systems—internal and external “stacks” and tools (e.g., ServiceNow, Salesforce, SAP, Oracle ERP, shipping tools, and homegrown apps)—and make decisions that depend on a lot of context. An automated fulfillment process may be the same in the US and Germany, with one (critical) exception: a choice of different local fulfillment partners.
Similar high-volume, high-stakes functions requiring cognitive power include claims processing, healthcare revenue operations, vendor onboarding and many more back-office functions.
Making Processes Self-Driving: Build a Learning Machine
By using AI to watch and learn from agents’ actual workflows at scale, it’s possible to efficiently create and train models specific to agents’ environments, enabling them to anticipate and respond accordingly.
By anchoring the AI model in problems humans solve, the model will continuously learn from real-life workflows instead of generative, morphing models that originated from statistical suggestions versus logic. This helps you get to the optimal state.
Briefly, there are three musts for this kind of new “learning machine”:
1. Work at depth
The deeper you can perform your workflow analysis, the better you can define individual workflows. All workflows are not created equal, even if they are running the same process. High-value step- and time-saving opportunities may be buried in a single workflow or in an obscure combination of steps.
2. Listen to your data
By looking at processes at an in-depth, individual workflow level, you can identify subtle differences in execution that can help you determine the optimal operating state for modeling. Optimize based on actual data and logic—don’t assume anything.
3. Carefully train and listen to your model
Models will be most powerful if you train them with many different users in different scenarios. Unlike with RPA, there’s no one-size-fits-all approach. Just like you would have many different cars driving the roads and mapping it out in creating our self-driving car above, you want many different agents training the model to ensure things are right and accurate.
For example, consider two agents working in a fulfillment operation. One agent executes the process notably faster than most other agents in getting to a resolution. Another agent works much more slowly, using more steps and systems in a much longer workflow.
It’s tempting to think that the fast agent is automatically “right” and declare his workflow optimal for your AI model. On deeper analysis, however, the fast agent reveals many reopened cases on the back end (for errors in how he resolved them). In contrast, the “slower” second agent had a consistent 100% resolution rate.
Or, you might have two “identical” agents working side-by-side doing fulfillment. However, one might have access to additional systems (since she’s Tier 1) over her cube-mate, who’s Tier 2. Their workflow may have overlap, but understanding the nuances is critical to properly automating the flow. Does the automation layer need extra access to this system? Why does only Tier 2 have access, and should that flow aspect be reconsidered?
Deflection and Beyond
There’s no question that AI will enable more business functions to be deflected from humans to bots and other, more intelligent, autonomous technology. Thus, expect more deflection—and perhaps more successful deflection—with GenAI and its successors.
The next big AI win will be in making “self-driving” processes for long-winded transactions that involve multiple systems and many physical steps for live agents. These processes must keep up with increasingly hyperautomated businesses to meet customer, financial, regulatory, and board expectations. AI-driven learning “machines” based on workflow analysis and other perspectives can help close that enterprise application gap sooner rather than later.
Guest post written by Boaz Hecht Co-Founder And CEO, 8Flow.Ai
8Flow.Ai will be joining us at Customer Response Summit (CRS) in Tucson, March 12-15, 2024. Learn more about CRS Tucson here.
