AI without processes is fast and confident, but frequently wrong
Artificial Intelligence that acts with speed and confidence might seem like exactly what every company needs. But there is a detail most people overlook: when that AI is not anchored in solid processes, it makes mistakes — and it makes them fast, at scale.
This is not an exaggeration.
A survey by Harvard Business Review Analytic Services, commissioned by Appian, found that only 16% of organizations manage to extract high measurable value from their AI investments. That means the vast majority are paying a premium for something that still does not deliver on its promise. And worse: many of these companies do not even know exactly where the bottleneck is — whether it is in the technology itself, in the teams operating it, or in the processes that should be supporting all of it.
This was exactly the scenario at the center of AppianWorld 2026 in Orlando, where Appian went beyond polished slides and brought real-world cases showing how process automation and AI can work together in a way that makes sense, both for the business and for the people using it day to day. The event brought together technology leaders, operations managers, and digital transformation experts to discuss something that rarely makes headlines: the difference between an AI that impresses in a demo and one that actually works in production, under pressure, with real data and real consequences.
Appian CEO Matt Calkins cited the HBR report during the opening keynote and connected the data to his longstanding conviction: AI only becomes useful when it is structured. He highlighted a line from the report itself — the next phase of AI maturity depends on embedding AI into the core of how work gets done — and followed up with pragmatism: Appian has been doing this for years. AppianWorld was therefore the stage to back up that claim, with concrete announcements and practical examples.
The company has 25 years of history building systems for environments where mistakes carry serious consequences — banks, pharmaceutical companies, government agencies. And now it is positioning AI not as a magic solution, but as another layer within a framework that already works. The question worth asking is: does this proposition hold up off the keynote stage? That is what we are going to explore here. 👇
AI as a new process layer
To understand what Appian is trying to do, it helps to consider where the company comes from. Appian is not, by nature, an AI company. It is a process automation platform that has spent 25 years specializing in digitizing and automating complex business processes at large regulated organizations. Think of banks processing loan applications, pharmaceutical companies managing clinical trials, or governments issuing licenses. Processes where an error is not just costly but sometimes outright dangerous. That is the market where Appian has operated for decades and where it is now adding AI as a new layer.
Why is this combination so important? The answer lies in a fundamental characteristic of AI that is frequently underestimated. Wei Smith, Product Manager for AI Agents at Appian, was direct during a session at the event: an AI agent that does not know your processes is fast and confident, but frequently just plain wrong. Speed and scale do not help if the output is incorrect. They only make the problem worse.
Jake Rank, Senior Director of Product Management for AI at Appian, saw this play out in practice after the general availability of Agent Studio. Customers wanted to deploy agents for practically everything, which is understandable given that the market is clamoring for agents. But experience showed that a standard business rule or a simple integration just works better in many situations. Appian built that lesson into Composer, which uses entered requirements to advise when you need an agent and when you do not. The organizations extracting the most value from AI are not the ones deploying agents everywhere, Rank says. They are the ones that know when not to.
The real problem behind the numbers
When you look at that 16% managing to extract real value from Artificial Intelligence, the first reaction is to assume the problem is the technology. That the models are not good enough yet, that the data is bad, that there is not enough computing power. But the reality that emerged at AppianWorld 2026 points in a very different direction: the biggest obstacle is not technical, it is structural. Companies are trying to apply AI on top of fragmented, poorly documented processes full of exceptions that were never resolved. The result is predictable: the AI learns the problems along with the workflows and starts replicating them at industrial speed.
This is a point Appian has been hammering home repeatedly in recent years, and it gained even more momentum at this edition of the event. Automation without process is just chaos at high speed. It does not matter how sophisticated the language model is or how precise the pattern recognition system might be: if the workflow it is operating on was not designed with clarity, with well-defined responsibilities, and with adequate checkpoints, the AI will scale exactly what should have been fixed. And the consequences in sectors like healthcare, finance, and government can be far more serious than simple operational inefficiency.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The data from Harvard Business Review Analytic Services is not just a number for a presentation slide. It reveals an organizational maturity that is still under construction across much of the market. The companies in that 16% group do not necessarily have the most advanced AI or the biggest technology budget. What they have in common is a process discipline that predates AI adoption — a solid foundation on which intelligent automation can operate with predictability, auditability, and measurable results.
A participant in the process, not a substitute for it
Choosing when to use an agent is one thing, but the challenge is ensuring that agent performs reliably. During the keynote, Calkins explained how this works in practice: an AI agent receives a single, well-defined task, can only do what the process allows, and is always under human supervision. Appian sometimes even verifies the result by having two AI models perform the same task and comparing the outputs. A process is a reliability machine, Calkins said. It catches errors before they cause damage. That was already true for humans, and now it is true for AI. In this logic, AI does not replace the process — it is a new participant that depends on the process to function.
Siddharth Goyal, VP of Intelligent Automation at Xebia, an Appian implementation partner, confirms this in practice. Two years ago, his clients were lining up to organize their data, assuming better data was the key to better AI. That conversation has shifted, he says. Most of the value is created when organizations focus on the process, not the data. Think about the process you want the AI to operate within, and the rest falls into place naturally.
AI agents: what changed in the conversation
One of the topics that sparked the most heated discussions at AppianWorld 2026 was the role of AI agents within enterprise workflows. Contrary to what many people imagine, agent does not mean total autonomy. In the vision Appian presented throughout the event, a well-built agent is one that knows exactly how far it can go on its own and when it needs to escalate to a human. That distinction might seem simple, but it represents a significant shift in how the market is thinking about AI maturity.
For years, the automation conversation revolved around removing the human from the process. The more automated, the better. The less human intervention, the more efficient. But that logic starts to become problematic when you are dealing with decisions involving regulatory risk, significant financial impact, or contextual judgment that models still cannot exercise consistently. What Appian is advocating — and what the cases presented at the event support — is that the most effective agents are those designed to collaborate with humans at strategic points in the process, not to replace them indiscriminately.
This approach has a technical name that has gained recent popularity: human-in-the-loop. But in the AppianWorld context, it gained a practical layer that goes beyond the concept. The Appian platform lets teams configure precisely which decisions an agent can make autonomously, which need human review, and which should be blocked until a responsible party validates them. This level of granular control over agent behavior is exactly what separates a mature AI implementation from a technology bet without guardrails. And for heavily regulated sectors, this is not a differentiator — it is a requirement.
But there is a subtle risk that Mark Talbot, Director of Architecture and AI at Appian, highlighted candidly. Appian emphasizes the importance of the human in the loop, but what if that person stops actually looking? If someone always accepts the AI suggestion without thinking, human oversight becomes a facade, he said. Real oversight requires process knowledge. You have to maintain that. Appian takes this risk seriously. In Doc Center, the module that lets organizations process and review documents automatically, the platform tracks who consistently accepts all AI outputs and who actually makes corrections. Not as punishment, but as a signal: if nobody is correcting anything, the question is not whether the AI is good enough, but whether the human in the loop is still truly exercising oversight.
Don’t start with technology: the lessons from Pfizer and Carlyle
Pfizer demonstrated at the event what it means to put process before technology. The pharmaceutical company has worked with Appian for years to manage contracts with healthcare professionals across 143 countries. Every day, 75,000 employees work with the system, in an industry where getting it wrong is not an option. Anne Furey, VP of Meetings, External Engagements and Travel at Pfizer, was clear during the keynote: in our industry, compliance is not a choice, it is our license to operate.
Before the implementation, closing a contract took six to eight weeks, sometimes longer. Now that time has been reduced to less than 24 hours. But that gain did not come from deploying AI first. Her colleague Kathy Maltz, Senior Director of Digital and Technology, explained the rationale: you cannot automate complexity, you need to eliminate it first. So we did not start with technology, we started by listening. Only after Pfizer simplified the process, standardized steps, and eliminated unnecessary stages was AI added as an extra layer. The result is a system where AI processes documents, verifies data, and flags discrepancies, but an employee always makes the final decision.
A similar pattern is emerging at Carlyle, the investment firm managing more than $470 billion in assets, spread across thousands of legal entities and bank accounts in 27 countries. There, payment processes are not a back-office function — they are the core of how operations happen. Working with Appian, Carlyle built a global payment system where AI processes invoices and extracts data automatically, but humans provide the approval. The result: a 40% reduction in the time between invoice date and payment, measured across more than 14,000 payment requests and over $4.5 billion in transactions in the first three months. Shakira Fraser, Head of Financial Operations at Carlyle, was direct when she took the stage: this is not a roadmap, it is not a pilot, it is not a proof of concept. This is a production platform that is already delivering real results.
Mission-critical for NASA
Perhaps the most impressive example came from NASA. The U.S. space agency built a completely new contract management system — the NASA Contract Management System (NCMS) — with Appian in just nine months. For an organization with eleven complex system integrations, active across every NASA center and responsible for approximately 85% of the agency’s budget, that is exceptionally fast.
The catalyst, by the way, was not enthusiasm — it was necessity. NASA lost 135 people through a voluntary separation program and did not replace them, while the Artemis mission was accelerating. As the workload increased, staffing decreased. Before the implementation, contracting specialists worked with a patchwork of outdated and siloed systems. They were so frustrated that many preferred drafting their contracts in Word rather than in the official system. For the organization, that meant no centralized data, no overview, and no control over its own procurement process.
That changed with Appian, says Melanie Landers, who led the technical implementation as Enterprise Applications Program Lead. The system played a direct role in the Artemis II mission, with which NASA sent humans back toward the Moon for the first time since the Apollo missions. For that mission, 2,700 suppliers from 47 U.S. states collaborated: companies that provided parts, materials, and specialized services for building the Orion capsule and the launch vehicle. Every contract required went through NCMS. Allison Sandt, Acting Director of the E-Business Systems Office, who led the functional side of the project, summed it up: without procurement processes that work well, no rocket gets off the ground. 85% of NASA’s budget goes through procurement. We are an essential part of every successful mission.
Process as a foundation, not bureaucracy
There is a quiet bias in the tech world when it comes to process. A lot of people associate process management with bureaucracy, with endless flowcharts, with chain-of-approval bottlenecks that slow things down more than they help. But what AppianWorld 2026 made clear is that process, when well designed, is exactly the opposite of that. It is what allows automation and Artificial Intelligence to operate at speed without losing reliability. It is the difference between a car with no brakes going fast and a well-tuned car that can accelerate and brake with precision.
Appian built its reputation on this premise. In the company’s 25 years, the focus has always been on creating platforms that enable organizations to map, automate, and monitor complex processes in environments where the margin for error is minimal. When the AI wave hit with full force, the company did not need to reinvent its value proposition. It simply positioned AI as another capability within an ecosystem that was already designed to handle complexity with control. That is product maturity, and it is also strategic vision maturity.
But not every successful implementation has the scale of Pfizer or NASA. Mark Talbot described in a conversation with Techzine a telecom provider where his team built an agent that verifies incoming architectural drawings: do the document types match this type of project? It sounds simple, but it saves three days of wait time per request, he said. The agent has access to the knowledge base, works through the steps independently, and logs its reasoning. An employee can see exactly how the decision was reached afterward. That was followed by a second use case: identifying duplicate project submissions. Two small steps, both with demonstrable time savings. That is exactly what we mean by Serious AI, Talbot says. It is not spectacular, but it is reliable and measurable.
The technology is promising, but it is not complete yet
The customer cases are impressive, but they do not tell the whole story. Behind the successes is a reality that Appian itself does not hide: the technology is promising, but it is not complete yet. Agents do not self-improve automatically, governance among business users is an open question, and the path from pilot to production is longer than many organizations expect.
Talbot is straightforward about what Appian’s agents are at their core: wrappers around an LLM, just like comparable tools such as LangChain or LangGraph. What sets Appian apart is not the agent itself, but the process layer around it. Agent refinement is still mostly manual work today. His team manually feeds feedback data into an external model and then adjusts the agent’s prompts by hand. The next step is for the platform to do that on its own, Talbot says. All of an agent’s input, output, and reasoning is already logged in Appian. The data is there. Now the system needs to learn from it on its own, without anyone having to intervene.
Rank explains why this step is so important. Without a learning loop, an agent’s accuracy drops as soon as business conditions change, new scenarios emerge, or policies are updated. You would be constantly patching things. With a learning loop, that same change becomes input, he says. The agent recognizes the pattern and performs better next time. But there is a clear limit: an agent’s definition does not change automatically based on user feedback. Fundamental adjustments always go back to the development environment, with all associated checks in place. The developer decides.
The governance question around low-code development by business users is another point where Talbot is transparent. Appian positions its platform in part as an environment where even non-technical employees can build agents. But what does that mean for control and compliance? That is a challenge, Talbot says. I strongly believe in involving the business, because they understand the work we are trying to automate. But to be honest, I do not know if our current approach is scalable. His colleague Rank sees it differently. In his view, responsibility for compliance falls on the business anyway, not on IT. The closer we bring the tools to the business, the better they can effectively own that responsibility. The two perspectives are not mutually exclusive, but they show that Appian is still searching internally for the right answer to a question that is at least as urgent outside the company.
Success can increase complexity
Beyond the technical and governance challenges, Goyal identifies another risk from implementation experience: organizations systematically underestimate how long it takes for capabilities to actually deliver value. This is not a switch you flip. It is a multi-year journey, he says.
Eli Zogby, VP of Process and Platform Excellence at Canada Life, a Canadian insurer serving millions of customers, was remarkably honest on stage about how challenging that journey is in practice. If I explained it well up here on stage today, it might seem like we have everything under control, he said. But we do not. Canada Life has built successful use cases, shortened cycles, and standardized processes. But every time a project succeeds, a new danger emerges. Ironically, project success can actually increase organizational complexity, Zogby said. You solve a local problem, celebrate, and start over. But without shared standards, you are constantly reinventing the wheel. The hardest step, therefore, is not the first use case. The hardest is what comes after: scaling, standardizing, and achieving real adoption.
What separates companies that extract value from those that don’t
Going back to that 16% managing to extract measurable value from AI, it is worth understanding what they do differently. Based on what was presented at AppianWorld 2026 and what the survey data reveals, a few characteristics keep showing up.
The first is the existence of a clean and well-governed data layer before any AI initiative even begins. It sounds obvious, but in practice it is one of the biggest bottlenecks companies face. AI needs quality data to work well, and quality data is the result of well-structured processes.
The second characteristic is the presence of a clear automation strategy that goes beyond isolated use cases. The companies at the top are not automating random tasks. They are mapping entire journeys, identifying where AI can add the most value and where human intervention is still irreplaceable. This strategic clarity prevents wasting resources on automations that impress in a POC but do not scale to production with the same performance.
The third, and perhaps most important, is a culture of continuous measurement and iteration. Organizations that extract real value from AI do not treat implementation as a project with a beginning, middle, and end. They treat it as a living process that needs to be monitored, adjusted, and evolved constantly. This requires tools that provide real-time visibility into agent performance, into deviations in automated workflows, and into the measurable impact of every decision made by the system.
Market positioning and the layer around AI
The question that remains after all these honest observations is how Appian positions itself in a market where vendors like Microsoft, Salesforce, and ServiceNow see the same opportunities. CEO Calkins has a clear answer: he positions Appian not as a competitor to the big platform players, but as an indispensable complement. Not the one supplying the power, but the wiring that makes sure the light actually turns on.
AI is indeed going to change everything, Calkins said in an interview with Techzine. But it cannot do it alone. It needs a layer that makes it trustworthy in critical processes. We are that layer.
In October 2025, Gartner positioned Appian as a Leader in the newly defined BOAT segment — Business Orchestration and Automation Technologies — alongside Pegasystems and ServiceNow. But Gartner also noted that, as of that publication, Appian had been slower than some competitors in adopting AI agents, with Agent Studio still in beta at the time. Jonty Padia, Practice Director at analyst firm Everest Group, sees Appian as a company with a strong technology position but a cautious communication style. Appian prefers to show what is already working rather than what is coming next, Padia said. That builds trust, but in a market where competitors are launching ambitious roadmaps, it can mean their story does not resonate as loudly as it could.
CTO and co-founder Mike Beckley outlined in the closing keynote how this uncertainty will evolve over the coming years. His message was strategically relevant not just for Appian, but for anyone thinking about where the real value of AI lies. He presented an overview of ten AI benchmarks from recent years and pointed to a pattern frequently overlooked in the best-model debate: the models are converging. The distinction is no longer in which model you choose, Beckley said. Whether it is the latest commercial model or an open-source alternative, that matters less and less. What counts is what you build around it: the processes, the governance, the structure that enables the model to do its job.
Beckley also described specific changes in how companies will deploy AI over the coming years. The data fabric — the architectural layer that makes data from different systems available in real time without physically moving that data — is getting smarter. Today, the platform already knows where the data lives and how systems are connected. What is coming is a deeper understanding of the nature of those connections, so that an agent does not just find the right data but also understands what that data means in the context of a specific decision.
The Gartner numbers make clear what all of this means for the market. Today, only 5% of enterprises use a consolidated automation platform that brings together processes, agents, bots, and people. Gartner expects that number to reach 70% by 2030. The race to own that position has begun.
AppianWorld 2026 left a clear message hanging in the air: the race for Artificial Intelligence will not be won by whoever has the most advanced model or the biggest technology budget. It will be won by those who know how to integrate AI with mature processes, responsible automation, and well-calibrated agents. And those still trying to shortcut that journey will keep being part of the 84% that invest heavily and get little in return. 🤖
