Where do current RPA solutions fall flat?
-
Dynamic Information / Volatile Data – RPA struggles with unpredictable data and constant rule updates. For example, freight costs or tax rules that change frequently can break an RPA workflow.
-
Human Judgment & Contextual Decisions – RPA cannot handle subjective or context-driven choices. Some components of a workflow require sophisticated, human-like evaluation like product substitutions or pricing exceptions.
-
System Fragility / Technical Complexity – RPA breaks easily due to UI changes and system instability. Legacy ERP screens with pop-ups or dynamic paths can disrupt RPA workflows.
RPA vendors have issued a slew of announcements promising advances with AI and how they lead the way with agentic process automation that will integrate in-house and legacy apps into the modern enterprise application infrastructure.
Do these promises deliver what practitioners need? Or is it yet another attempt to overlay a new shiny object on their battered and bruised legacy technology stack? Hot patching AI to an existing platform simply does not work, very much like lifting and shifting legacy appliances to the cloud did not work. You can’t simply transform a rigid rule- based automation system by gluing generative AI on top of it.
The Reality of Most Agentic AI Solutions
AI is dominating the operations discussion and legacy automation vendors are eager to capitalize on the trend and tout the transformative potential of their solutions. However, if you dig below the surface it becomes apparent that the delta between promise and reality is wide. Technological innovation is weak and the cost is high.
For example, many vendors are simply creating AI-aided legacy script generation. This is not innovation, but rather a hot patch for a failed solution of the past. In fact, RPA is using the power of generative AI to generate more RPA scripts. Nothing has substantially changed in the underlying technology. While it is possible to model basically anything in RPA scripts, they get incredibly large and complex to manage. No enterprise operations team has the time or resources to maintain systems that are so complex and error-prone.
None of the legacy RPA solutions are offering automation of complex processes that RPA was not able to automate previously. Nor are they delivering advanced reasoning and interaction skills which enables non-technical staff to easily create their own workflows and process improvements. One of the most telling facts in this scenario is documented difficulty in deploying and maintaining these systems over time. In fact, most vendors require mandatory implementation partners and service partners because of this difficulty. Isn’t that contrary to the promise of AI to simplify processes and alleviate work from the user? Or are we simply shifting the cost from labor to professional services vendors to enable and continue the illusion of automation?
The Possibilities of True Agentic AI
At Opnova, we approach agentic AI from a clean slate, unencumbered by an existing technology stack. We are solving problems that were previously impossible by automating applications and processes like order management and employee management with human-operator-like reasoning capabilities. Our mission is to eliminate rework, those repetitive mundane processes that slow productivity and ultimately impact the bottom line.
We’ve talked to CIOs and operations professionals at dozens of companies, and they all agree that rework is the silent enemy of business transformation initiatives. The power of agentic AI to address the rework problem is limitless. We want to move beyond contextualized search with half-truths as answers and become a partner. Agentic AI can demonstrate immediate improvement in a range of enterprise operations including two of the most difficult to completely automate, Order Management (OM) and Employee Lifecycle Management (ELM).
Historically, the ROI of RPA for automating OM and ELM has not been realized. Both of these legacy applications have dynamic components, including pricing, shipping rates, and policies that script-based technology cannot solve reliably. RPA is brittle when working with these applications, meaning workflows can break easily when dealing with complexities in the data set. Ultimately, teams must spend time they don’t have to babysit bots to address snags generated by the failings of their RPA technology, exactly the outcome they were deploying automation to avoid. Once again, instead of solving the problem, they are shifting it from one problem owner to another.
A truly agentic process can overcome many common edge cases that humans deal with and overcome everyday such as application layout changes or a list loading slower than usual or perhaps the app crashed while loading or thrown a typically benign software exception that needs to be approved to disappear or an overly zealous updater hijacking the screen, the list of daily disturbances is a long one yet human operators deal with them with ease. Well, so do the AI agents now.
Questions to Ask Before You Buy
Agentic AI has the potential to deliver unprecedented advances in enterprise IT operations, security, and more. To ensure success, make sure your organization is a smart consumer in this new frontier. Ask the following questions before committing to a solution or vendor.
- Does the solution offer multiple tool-use capabilities from computer-use to API calling?
Computer-use means operating applications and accessing data like a human operator would. This is essentially the ability to observe and learn from operators and mimic their activity using the computer interface. For example opening a PDF attachment that is received by the Accounts Payable inbox, extracting the data and entering into the ERP with the contextualized information.
- Can you reliably create workflow automations, or are they limited to contextualized search and answers only?
Contextualized search capability does not mean actual problem solving. Actual problem solving typically requires at a minimum a data extraction step followed by a complex set of navigation between applications while entering the said data all the while having a list of reasoning steps in memory for example which shipper to use for a given customer. Is DHL the best option for EU based customers and USPS for North West USA?
- Can you purchase without hefty service contracts or are you being force fed an implementation partner?
It is a well-known industry fact that technology that is difficult to install and/or manage will not be used. If a vendor mandates an implementation partner as part of the contract, ask yourself why this is necessary. Similarly, if you need to purchase expensive service contracts as part of the deal, question whether your team will ever be able to manage the software on their own. Agentic AI developed and implemented correctly, should simplify tasks not create new, onerous overhead.
- Does the solution work across legacy and modern apps?
The true power of agentic AI in operations is the ability to connect all your applications and eliminate rework across the organization. If a solution cannot close the automation gaps between legacy and new applications, you will be adding another layer to your technology stack that does not advance your operational capabilities.
Agentic AI is a brave new world with the potential to truly transform business operations. The power is in the ability to replicate human actions and reasoning to modernize workflows and connect the impossible to connect. Do your homework and eliminate rework to advance your organization.
Cheers,
Sinan Eren
Co-Founder/CEO
