According to Appian, AI must and can do much better. Many companies have now completed pilots and implementations, but the projects have a high failure rate. The solution? An approach in which AI does not stand alone, but is linked to data and processes. This allows companies to automate back-office processes with concrete results.
The figures don’t lie, Appian explained during our visit to the company in London. A MIT study last year showed that 95 percent of generative AI implementations fail. They generate little to no business value. There can be various reasons for this. Think of poor integration, little focus on solving real problems, and unrealistic expectations.
As far as Appian CEO Matt Calkins is concerned, this is an absurd percentage for something that is seen as the breakthrough technology of this generation. The core of the problem? AI is often considered a standalone technology. “People thought AI could work on its own,” explains Calkins. “But that’s not the case. AI needs other technologies to deliver real value.”
Calkins, therefore, advocates “serious AI,” in which the technology is used primarily for essential business functions. The approach combines three elements: AI, data, and process. According to the company, this triangle forms the basis for successful implementations.
Data and process as the foundation
It comes as no surprise that data is the driving force behind AI. After all, a model is only as good as the data it is trained on. An AI agent needs access to all relevant information to draw conclusions or learn from mistakes. If your data is not in order, your entire AI project will collapse and become one of many failed projects.
As a process automation player, however, Appian also very emphatically links processes to successful AI. Because AI, Calkins argues, is only as good as the work you give it. According to him, the process is the missing link that connects AI to meaningful work. These are tasks that involve multiple employees and steps. A process actually gives AI structure and measurability, precisely the components that are often missing.
Serious AI does not ultimately take place directly in the areas one might expect. For example, it can be useful in high-volume back-office processes, where AI adds critical new skills to existing teams. Think of procurement, case management, contract management, or compliance. Calkins acknowledges that this may be boring AI, especially when compared to the industry’s AI promises. “But the results are definitely not boring,” says Calkins.
Tip: Appian raises the bar in battle for process automation crown
Applicability of serious AI
Appian has invested years in multiple tools designed to enable serious AI. Data fabric and process modeling are fundamental within the concept of serious AI. The first of these two, data fabric, enables linking AI to data across the organization. Process modeling, in turn, connects AI to large collaborations between people and digital employees.
An example of how Appian enables serious AI in practice is the platform’s DocCenter feature. DocCenter processes the influx of documents large organizations must deal with. Registrations, upgrades, compliance documents, complaints, and questions come in all kinds of formats. Digitally, on paper, as voice recordings, or handwritten. The AI reads the documents, determines their meaning, and extracts essential information. That information is written to the appropriate databases. The AI also directs the work to the right person. For urgent or emotional messages, the system adds extra urgency.
To achieve reliable output, Appian sends the same document to multiple AIs to verify the result. If they agree, action is taken. If there is a difference, a human must check it. This way, a success rate of up to 99 percent can be achieved. Process integration is also important here, not just reading a document. A document is a stimulus that requires action. DocCenter, therefore, integrates with all follow-up actions.
Appian shares a customer example of DocCenter with us. A large Australian insurer used to perform commercial acceptance assessments manually. With DocCenter, the process went from four days to 90 minutes, with 99 percent accuracy. The savings amount to more than a million dollars per month, according to Appian. Meanwhile, 70 percent of all customers reportedly use AI in production this way.
The outcome is leading
Appian will also be focusing extra attention on agentic AI in the coming period. Appian 25.4 will mark the end of the long beta program for Agent Studio. This tool, announced in April, allows developers to build AI agents that orchestrate entire business processes. The agent has access to all tools, records, rules, and documents. It independently determines the best path to the desired result. They also adapt to changing circumstances.
According to Calkins, the functionality differs from other AI agents in three ways. Agents have access to all data in the organization via the data fabric. They take action by initiating processes. And they learn thanks to extensive tracking of everything that happens on the platform.
With this agentic AI step, Appian, like other artificial intelligence companies, clearly understands that the technology only has value when it is embedded in business-critical processes with clear objectives and metrics. It is not just about technology, but also about business outcomes. In the coming years, Appian expects more and more organizations to follow this path from experimenting with AI to structurally embedding it in the processes that really matter.