It's easy to get caught up in the AI hype, but excitement can stop us from seeing the practical steps needed to make AI truly work. At Appian, we recognize that AI is at its most powerful within a process. Before you get to embedding AI in process, however, you must determine if AI is what you need. It is important to stop focusing on the technology itself—the "what"—and start focusing on the actual problem it solves—the "why." This strategic shift from thinking about technology first to thinking about value first is critical to protecting your investment. Our validation criteria help you avoid the common mistake of “using AI just for the sake of using AI." 

• Hand work is about getting mechanical tasks done efficiently, such as moving data or repetitive clicking. This is what robotic process automation (RPA) is for.
• Head work is about the mental effort involved in areas like analyzing data, making judgments, combining information, and figuring things out. This is where AI should be used. 

This leads to our main point:

AI should solve a mental challenge, not just a simple task. If a task doesn't require thinking, don't use AI for it.

This rule makes sure that you only apply AI where it gives you a real advantage. It stops you from using expensive, complex technology for easy problems, and most importantly, it protects your investment by focusing on the areas that have the biggest impact.

• Gate 1: Cognitive burden test. This is the "Why?"
  It asks the core question: Does this task truly involve a thinking step?
• Gate 2: Technical fit. This is the "How?"
  It makes sure you're using the right kind of AI for the problem, specifically asking if the task is about words or numbers.
• Gate 3: Value integrity check. This is the "Should we?"
  It’s a crucial check on whether you'll get a good return on investment (ROI) and what the risks are.

Only the projects that successfully pass through all three gates should be considered a high-value AI implementation.

The cognitive burden test is the first and most critical filter. Its purpose is to clearly answer: Does this task really need artificial intelligence, or can a simpler automation tool handle it? This gate forces us to separate tasks that just need to be faster from those that truly require thought.

To help with this, we use the 2-second rule. If a human can do the task in under two seconds without much effort, trying to force AI into it will probably just add complexity, not speed or value. Remember, AI is for taking the heavy thinking off people’s plates, not for automating a few easy clicks.

To validate a project, ask these two practical questions:

• Does the user have to figure out, combine, or judge information to complete the task?

• Can you name the specific person whose brain will be unburdened by this automation?

Gate 2 is all about using the right tool for the right job—it's a technical check. Appian AI is highly specialized for language, context, and process rather than statistical forecasting or complex math. The simplest way to think about this is the "words vs. numbers rule":

Sum it up this way: For math, use Appian expression rules. For context around the math, use AI. This approach gives you the best of both worlds: the 100% mathematical accuracy of the decision engine and the human-like context and reporting of AI skills.

Gate 3 is the final reality check for any AI project. This is where we ask the hard business questions. The key principle here is: Just because we can, doesn't mean we should.

To pass this gate, a project must meet these three standards to ensure a positive return on investment (ROI):

• Cost vs. reward. Is the effort worth the benefit? We have to figure out if the time and money spent setting up the AI will be less than the time users will save over a reasonable period.

• Risk mitigation (human-in-the-loop). For any business-grade AI, this is a non-negotiable safety check. You must always have a person check the AI's output to prevent errors, ensure accountability, and maintain compliance. 

• Process value. Does the AI genuinely make the process better or are we in danger of slowing down a simple process with expensive tech? The goal is to improve the work, not just add a layer of technology for show.

The core change in strategy during the discovery phase is simple but powerful: don't ask what you want AI to do. Instead, ask about the mental strain. Listen for phrases that signal heavy thinking. When you hear:

• "...it takes forever to read these notes," you've found a summarization use case.
• "...I look at five places to understand what's happening," you've found a synthesis use case.
• "...I need a 'gut feeling' to do this," you've found a task that relies on judgment—a perfect fit for AI.

You can also use the following questions to actively find these opportunities:

• Where in this process does a user feel like they are doing 'homework'—summarizing or searching—before they can make a decision?

• Show me the exact point in the process where you have to stop and think for more than a minute. What information are you trying to connect?

This approach turns the conversation from a tech demo into a session focused on real business solutions.

• Words vs. numbers. Use Appian's powerful decision engine for math to guarantee 100% accuracy, and save AI for language, context, and summarizing. Don't use it to predict future numbers.
• Input dictates output. This is a basic truth of AI. If the data you put in is messy, irrelevant, or incomplete, the AI's output will be a problem, not a help. Quality in, quality out.
• Be a guardian of value. Your role is to protect the client's ROI. Recommending a simple, reliable decision rule over a flashy AI skill when it's the right choice is the fastest way to build trust and position yourself as a true partner.
• Human-in-the-loop. This is a mandatory rule for enterprise AI. You must always ensure there is expert verification to manage risk and provide a critical safety net against errors.

The only question left is: What piece of heavy thinking will you solve first?