Workflow: the governance engine for AI implementation
This is a guest blogpost by Don Schuerman, CTO at Pegasystems.
As AI fever continues to run hot, overwhelmed enterprises may be tempted to select the shiniest “solution” without fully assessing whether it’ll properly address their goals or pain points, which could affect their processes and eventually their return on investment (ROI).
This is confirmed by the recent MIT study, which revealed that 95% of enterprise AI investments fail to deliver. Although business and technology leaders might find it disappointing and frustrating to see zero return on their initiatives, this highlights a crucial lesson: investments shouldn’t be AI-driven, they should be outcome-driven. The goal shouldn’t be to just deploy AI, but rather use AI, often in conjunction with other tools, to increase efficiency, streamline experiences, and drive business innovation.
The AI rush to the head
There is a lot of excitement in adopting new technology like AI, but there’s also a lot of noise in the marketplace about the different tools available, which can create confusion. For most enterprises, large language models (LLMs) alone are not the solution because using an AI tool as a universal solution to every challenge will only lead to disappointment. The value for the enterprise comes when LLMs can be effectively integrated into workflows, data, and business context.
Integrating modern AI tools into existing technology stacks isn’t easy. Many enterprises have hundreds of technologies scattered across the business, including old legacy systems and new tools they’re adopting. Deployed at design-time and integrated into runtime workflows, AI can provide a path towards getting all those siloed systems connected to work more meaningfully.
AI at design time
A great way to take a business-first approach and set the organisation on the path towards AI transformation, is to start by looking at specific workflows. By identifying the places where the business could deliver more efficiently or effectively, leaders can focus AI efforts on making processes more efficient, compliant and easier for employees to deliver better customer experiences.
Take for example, a bank issuing a loan. There are certain steps that employees need to follow due to internal policies, but also government regulations. The workflow was likely designed around the assumption that most of the work would still be done manually. It can be hard for subject matter experts to even see the possibility of what the better process could be.
By using AI as a design-time partner, enterprises can rethink these workflows. Available tools can consume existing documentation, screenshots, even videos of old systems – and instantly recommend a new, digital workflow that applies AI agents to automate historically manual tasks. It can design self-service experiences for potential loan customers, agents that automatically review documentation, even business rules calculate risk in a transparent and predictable way. Taking this design time approach with AI not only accelerates time to market, it allows you deliver a fundamentally better loan process.
Consistent experiences at runtime
A modern workflow isn’t just a series of forms and tasks; it orchestrates how agents automate historically manual work like document review or researching a property’s history. It dynamically pulls in the data it needs from other systems without manual intervention from the users. What is more workflows can act as the governance engine for all those AI agents.
Imagine an AI agent that is aware of all the workflows available in your business. When a user or customer requests an action, the agent doesn’t need to reason the next steps on its own. After all, this is where the risks of hallucination and inconsistency loom largest. A bank wouldn’t want an agent figuring in real-time how to process and whether to approve each new loan request.
But if that agent could find a validated “Loan Processing” workflow. It could then ask the user or customer for any information that is needed, walking them through the required workflow steps. If a decision is needed that requires predictability and consistency, the workflow can provide the business rules that an agent can follow. It also can ensure how other agents are invoked for specific tasks, like validating a customer’s identification from a driver’s licence. The enterprise gets a faster and more efficient process through AI, the customer gets a better experience and a quicker result with all this delivered predictably and consistently.
2026 should mark the shift from “AI for AI’s sake” to smart application of AI and other technologies to deliver better business outcomes. Workflow serves as the governance engine that helps enterprises integrate AI more thoughtfully into the business adding value and competitive advantage.