The Missing Skill Behind AI Automation

Everybody wants to build AI systems that do the work for them. Very few people want to build the skill required to make those systems useful.

That skill is process development.

It isn’t a particularly exciting phrase. It doesn’t sound like the kind of thing you’d put on a conference stage with dramatic music behind it. But if you can’t define what needs to happen, what inputs are required, what decisions must be made, and what a good output looks like, AI has very little to work with.

AI isn’t a mind reader. Neither are the agents, automations, or software systems built around it.

The people who get the most leverage from AI won’t simply be the ones who know the most tools. They’ll be the ones who can turn messy work into clear, repeatable processes. I’ve found a simple framework that makes this much easier:

• Analyze what happens now.
• Optimize the process for a better outcome.
• Standardize the new way of working.
• Mechanize the parts worth automating.

The order matters. Most failed AI projects skip straight to the last step.

AI automation starts with process development

When people talk about AI automation, they often begin with the tool. They ask which agent to use, which platform can connect their apps, or how to get a model to complete a task.

Those questions matter eventually. They’re just not the first questions.

Before AI can automate a process, someone needs to define the process. That means identifying the trigger, the sequence of actions, the decisions along the way, the handoffs, and the desired result.

This can apply to a small marketing task, like turning a research brief into a first draft. It can also apply to an entire department that takes in customer information, coordinates several teams, and produces a service or product.

A business is, in many ways, a collection of processes. It takes in money, labor, information, and materials, then turns them into value for a customer. If the processes are unclear, the business depends on tribal knowledge and individual heroics. That’s a difficult foundation for automation.

I learned this after spending an almost unreasonable amount of time looking for business processes. I was searching for them everywhere, as if I could catch every useful process and catalog it. Eventually, I realized that most processes are specific to a company and its circumstances.

So I changed the question. Instead of hunting for the perfect process, I started looking for the process behind process development: how do you create a process that actually works?

I asked marketers, business leaders, and people with MBAs. I had a hard time finding a satisfying answer until I spoke with Tom Muccio, a former Procter & Gamble executive who helped lead the transformation of the P&G and Walmart relationship so the companies could operate more like one organization.

I asked him directly: “What’s your process for developing processes?”

His answer was immediate: analyze, optimize, standardize, and mechanize.

That framework has stuck with me because it maps so well to the way we need to approach AI.

1. Analyze the process you actually have

The first step is to document what currently happens. Not what you wish happened. Not what the process supposedly looks like in a slide deck. What actually happens on a normal day.

Start with the entry point. What triggers the process?

Then work through it step by step:

• What information or materials come in?
• Who performs the first action?
• What decisions need to be made?
• Where are the handoffs?
• What checklists, tools, or systems are used?
• What exceptions regularly appear?
• What output is produced?
• How do you know the output is good enough?

For many processes, a one-page list is enough to begin. You don’t need a sophisticated diagramming application. Pen and paper can work. The important thing is to get the invisible steps out of people’s heads.

This is where teams often discover that “one task” is actually fifteen smaller actions. A content workflow might include choosing a topic, checking search intent, gathering source material, outlining, drafting, editing, adding examples, formatting, publishing, and reviewing performance. If you simply tell an AI agent to “create content,” you’ve skipped the part where the work is defined.

Analysis also exposes the parts people have quietly added over time. Maybe three people review the same item because nobody knows who owns the final decision. Maybe a team copies information between four systems. Maybe a step exists only because an old tool used to require it.

If you don’t understand those details, you can build a beautiful new AI process that fails on its first day. The system may remove a step that looked unnecessary but was protecting quality. Or it may produce an output that technically follows the instructions but doesn’t meet the customer’s real needs.

Before asking AI to improve a process, give it an honest description of the current one. You can learn more about choosing the right starting point in my guide to the marketing workflows worth automating first.

2. Optimize the process for a better outcome

Once the current process is visible, you can start imagining a better version.

This is the fun stage. You’re no longer constrained by “that’s how we’ve always done it.” You can ask what should happen, what could be removed, and what new capabilities AI makes possible.

Give AI the documented process and use it as a brainstorming partner. Ask it to identify:

• Repetitive steps that could be automated.
• Steps that could be combined or eliminated.
• Information that should be collected earlier.
• Decisions that need clearer criteria.
• Potential failure points and exceptions.
• Places where a human review would improve the result.
• New ways to improve speed, cost, quality, or customer experience.

Notice that optimization isn’t the same as automation. You may find that the best improvement is a change to the human part of the process.

Suppose AI can summarize customer feedback. That might free a strategist to spend more time interpreting the patterns and deciding what the company should do next. The goal isn’t necessarily to remove the strategist. It may be to move that person toward the work where judgment has more value.

Likewise, automating a low-value step may make it possible to add a high-value human interaction elsewhere. If a system handles routine qualification, a sales representative might be able to spend more time with serious prospects. If AI speeds up research, a writer might spend more time developing a distinctive point of view.

Optimization asks a broader question than “Can AI do this?” It asks, “What would make this process meaningfully better?”

That distinction matters. Otherwise, you end up preserving a bad process and simply moving it into a more expensive, more complicated tool.

3. Standardize the new process

After you’ve designed a better process, write it down before you automate it.

This step feels slower than jumping straight into an AI builder, but it prevents a great deal of confusion. If the new process exists only in a meeting, each person will remember a slightly different version of what was decided.

One person thinks the AI drafts the email and sends it automatically. Another thinks a manager reviews every message. A third person assumes the system should route unusual cases to customer success. Those interpretations will eventually show up as inconsistent outputs, missed handoffs, and frustrated teammates.

Standardization creates a shared operating agreement. It should clarify:

• The trigger that starts the process.
• The inputs required to begin.
• The sequence of steps.
• The rules for making decisions.
• The person or system responsible at each stage.
• The points where human judgment is required.
• The acceptable output and quality criteria.
• The actions to take when something goes wrong.

You don’t need to write the first draft from scratch. If the process was designed in a group meeting, you can run the notes or transcript through AI and ask it to create a draft standard operating procedure. Then have the people who perform the work review it.

AI is useful here, but it shouldn’t be the final authority. The people closest to the process will spot missing context, unrealistic assumptions, and exceptions that a model can’t infer from a clean summary.

For a small process, standardization might take an afternoon. For a process that crosses teams, vendors, or departments, it may require several meetings, training, and a phased rollout. People don’t automatically adopt a new process because it’s logically better. They need to understand what changed, why it changed, and how they’re expected to work with it.

If your AI outputs are inconsistent even when your prompts seem reasonable, the problem may be upstream. The missing step for more consistent AI results is often a clearer process and better-defined inputs.

4. Mechanize what’s worth automating

Only after the process has been analyzed, optimized, and standardized should you mechanize it.

Mechanization is the automation stage. This is where you decide which parts should be handled by software, AI, agents, or integrations.

Some steps will be deterministic. They can follow explicit rules: take this input, check this condition, update that system, and notify this person.

Other steps will be flexible or creative. AI may generate options, classify information, summarize context, or make a recommendation. A person may still approve the result, resolve an unusual case, or provide the final judgment.

That distinction is useful because not every part of a process should be treated the same way. Some work is ideal for full automation. Some is best handled through human-AI collaboration. Some should remain human because it depends on empathy, accountability, taste, or a decision with serious consequences.

The right question isn’t “How do I automate everything?” It’s “Which parts can be mechanized without damaging the outcome?”

Even with today’s increasingly capable AI systems, automation takes time and resources. You may need to build integrations, test edge cases, monitor performance, and pay for model usage. If you automate the wrong process, you haven’t removed waste. You’ve made waste run faster.

That’s why I’m cautious about starting with the tool. Tools are easy to buy. Clear processes are harder to create, but they’re where the durable advantage comes from.

What process development changes for marketers

Without process development, AI tends to function as a task delivery system. You ask it to write a draft, summarize a document, generate ideas, or classify a list. Those tasks can still save time, but the leverage remains limited.

With a well-defined process, you can delegate a deeper level of work. An AI system can manage a sequence of related tasks, carry context from one step to the next, and know when to pause for human review.

That changes the role of the marketer. Instead of simply producing more individual outputs, you can design systems that handle substantial portions of a workflow. You’re still responsible for direction, quality, and judgment, but you no longer have to personally push every step forward.

This is also why process development belongs alongside your uniquely human strengths. My Human Edge Framework for marketers focuses on the judgment, empathy, taste, and original thinking that AI shouldn’t replace. Process development helps you protect those strengths by moving repetitive coordination and execution into systems.

AI can create leverage, but only when someone has thought carefully about the work being leveraged.

A practical checklist for your next AI system

Pick one workflow you currently perform manually and walk through these questions:

1. Analyze: What triggers the workflow, and what happens from beginning to end?
2. Optimize: What could be removed, improved, accelerated, or redesigned?
3. Standardize: Can another person read the process and understand exactly what to do?
4. Mechanize: Which steps are worth automating, and where should a human remain involved?

Then test the smallest useful version. Don’t try to automate an entire department at once. Start with a process you understand well, has a clear output, and contains enough repetition to justify the work.

Document what happens, inspect the result, and revise the process based on real use. Automation isn’t a one-time installation. It’s an operating system that needs feedback.

The central lesson is simple: don’t begin with mechanize. Begin with analyze.

When you can define the work, you can improve it. When you can standardize it, you can teach it. And when you can teach it clearly, you can decide what AI should do, what people should do, and how the two should work together.

In this episode, Dan Sanchez emphasizes the critical skill of process development in the age of AI, outlining a structured approach to creating effective processes. He introduces a four-step framework—Analyze, Optimize, Standardize, and Mechanize—that enables marketers to leverage AI effectively and enhance productivity. Takeaways Process development is essential Follow the four-step framework Chapters 00:00 The Importance of Process Development 03:03 Mastering the Process: Analyze, Optimize, Standardize, Mechanize

episode final v1: Everybody wants to build AI systems to do the work for them, but nobody actually wants to build the skill that you need in order to build and automate all these systems that all these agents can do. Because let me tell you, the AI is actually good enough. The software around it, the harnesses, the intelligence, it's all there to automate a ton of work. But there's one massive missing skill that's holding people back. From being able to create the systems that can do so much of the work for them. Not all of them, but a lot of it. And in this podcast episode, I want to talk about this one skill: process development. Now, welcome back to the AI-driven marketer. I'm Dan Sanchez, my friends, call me Danchez. And I know this is not a sexy term at all, process development. But think about it. If you can't define the process, if you can't discover like what you actually want, the outcome of the thing and the inputs you need to create the thing, and some of the steps broadly in order to create said thing, then you're out of luck. Of course you can't automate the processes because the AIs can't. Guess what you want, you know, they can't they aren't mind readers. You actually have to be able to define a few bits and pieces of the process for them to be able to serve you well. All these autonomous agents and systems can't happen unless we can actually define what we want to have happen. And of course, some of it is defining it so that AI can fully automate it, some of it is defining it so that AI can collaborate with you on it, and some of it's defining it to find out what are the few pieces that only humans can do in it. But it's a wonderful skill, a boring skill, but a very powerful skill in this age of AI. And those who actually take the time to master it are gonna be the ones outrunning everybody else with AI. And it's not even particularly hard. It just takes a little bit of time, usually pen and paper and some thinking and the framework that I'm about to hand to you. Because believe it or not, I used to be a a like like people search for freaking Pokemon. I was searching for business processes at one time, like you gotta catch them all. And I was obsessed with them. I was searching for them everywhere. And then it dawned on me that not all business processes are created equal and much of much of them are like tailored to specific situations for specific businesses. And I started a different hunt of a different journey of a kind. And I started hunting for the process behind process development. How do you create the perfect process to run different parts of your company? I asked a lot of really smart marketers, different business people, different people with MBAs, and I actually had a hard time finding the answer. And I searched for over a year until I actually ran into a former PG executive. And this wasn't any executive. This was Tom Muccio. This was a guy who was the not the brains, but really the the the I'd say the leader behind the transformation of PG and Walmart coming together. To behave like one company. Think about how hard that is. Think about the process muscle that guy has. So I finally got a chance to ask him, Tom, what's your process for developing processes? And he looked at me and without batting an eye, he's like, yeah. It's analyze, optimize, standardize, and mechanize. And I was like, whoa. He then went on to explain kind of like this the step-by-step process he takes in order to Make processes. Remember, this guy knows it and has lived it for most of his career. And now is like a high end consultant you can hire to help you do this kind of stuff. But I wanted to share it here with you because I find that this is the kind of stuff that you can use to help you build all these AI systems. Just like in a previous episode. The last episode I published on the show, I talked about how more of us are not just using tool, but building tools. But in order to build a tool, we need to be able to define what the tool does, what we need to give the tool and the process it needs to use in order to drive that outcome. But let me walk you through the advice that Tom gave me, and I'm going to be outlining extensively in a book coming soon, about how to use this process. Analyze, optimize, standardize, and mechannize. Let's start with analyze. Most of you already have processes that you're using. The things that you want to automate, you're probably doing manually. Some you know the steps and some you don't. They're like, they're, they're Things that you do all the time. You just consider like, I'm just doing the thing, but it's actually a bunch of little baby steps in between. It might take, it might just be you, it might be your whole team, it might be across corporations with vendors and all kinds of things. But the first step in improving a process or even just automating it is defining what's currently happening. Actually taking the time and with pen and paper just mapping out step by step, like, it starts it all starts when this sets when this happens. This is the beginning of the process. This triggers it. There's always an entry point to a process. And then we go step by step. Maybe there's a few decisions that need to be made. And then it gets handed off here. They run through this checklist and then it moves here. And then it runs through this process. And then we have these outputs on the other side. If you think about it, a whole business is nothing but a process. We take in money and either labor or some kind of raw materials and we output value. To the customer on the other side, whether it's services rendered or products, right? Like the whole business is just one series of processes. So being able to prove it is important. And the first step is analyze where you actually measure what are we currently doing? Because if you don't know what's currently happening, shoot, you could throw a whole new process in there with AI and come and dazzle people with AI, and you're like, this is the new AI process, but you never took the time to analyze what was going on now. That's when AI processes die. On day one, because you didn't realize there's a few critical things that you missed because you never analyzed to figure out what was going on currently and why it was going on. You got to do that first. But once you do know, once it doesn't even have to be complicated. It could be a step-by-step list, it could be a one-pager. That's it. Once you know, once you've analyzed, you can move to the second step. Optimize. this is the fun part. This is the part where you can bring AI into the conversation and kind of brainstorm what can happen, what can be improved, what's possible now with AI that wasn't. possible before. Ha ha. You can t ask AI and it'll tell you. Give it the process. Give it the part you analyzed and documented in order for it to help you find out what those pieces can be. And the cool part is it doesn't even have to be AI. You can optimize the human part too. Like, if we save this, if we can automate this part with AI, maybe we can improve this human element for the sake of our customers here or save money by doing a human evaluation here. Like it's there's all kinds like The world is your oyster. What do you want to have happen? What could possibly how could you put make this better? Those are the this is the fun stage because you get to dream up a new version of reality here. I always love this. The where the brainstorming happens, where people's ideas can come to fruition. So we optimize it, but then it can't just stop there. We have to standardize. Believe it or not, you have to do this step first before you move to mechannize, which is the automate. when you Bring in those AI agents. Hold off. We gotta actually document what we want to have happen. We gotta put it down on paper. Otherwise, the new thing that we all agreed needs to happen? What if everybody has a different interpretation of what that new thing is? What if that new thing that Susie has to do and Susie understands it differently than you do on her step of the new process? that's when writing it down helps. Luckily, it's easy to write it down. Just dictate out the process. Shoot. Take the brainstorming meeting where you all agreed on what needs to happen next. Run that transcript from that meeting through AI and have AI create the rough draft of what the new process is so everybody can read it, be on the same page. And sometimes it takes a little longer. If it's a big process requiring lots of people, it might take multiple meetings to iron out and take some training. It might take some onboarding. You might have to roll it out in phases for people to get used to it. Cause people hate change, right? Especially if you're bringing in some AI tools. You might have to reintroduce them to a new way of doing it and accept working with these things. And then you automate. Because trust me, nothing's more painful than automating something that never should have been automated in the first place. But you automate what the parts that are left. You have to standardize and get everybody on the same page and kind of build the systems and test it, make sure it works. But before you automate, you want to make sure it's worth automating. As easy as it's gotten with AI, it still takes a lot of time and tokens, sometimes some very expensive tokens. To build out the whole process, the new software, whatever you're doing with agents, because even by the time maybe you watch this a year from now, from the time I've published this, maybe agents are orchestrating even more. So I don't even want to specify what you can, but we know we're automating some of it very deterministically. We specify every single little detail and point and decision that can be made, but some of it's gonna be more creative and flexible. AI can do some have some more judgment calls in there. But that's the process. And I promise, if you don't start with Mechognize and you actually go through the process of analyze, optimize, standardize, and then get to the point where you can unleash AI to automate, speed up, and even improve a lot of these steps, this is where it really starts to become fun with AI. And I have a feeling that the people who are ahead of the game a year from now are the people who really master this idea of process development today. Because I promise, without this step, AI is just going to be a task delivery system. It's going to be one-to-one or a few tasks at a time. But if you can get the process down, you can delegate much more and much deeper level projects. And those are going to be the marketers that get promotions because they don't just have themselves, they have whole AI teams that can handle much more substantial workloads. So learn it today. It's analyze, optimize, standardize. Mechanize. And a new book is coming with more. So if you like what you're hearing and you think this would be useful for you, you can get a free copy of this book. Yes, because you're a listener to this podcast, I'm gonna give it away for free and then I'm going to charge for it. It's going to be free for a limited time and only to people that join the launch list. So go to danchez.com slash process. You can join the launch list and get notified when this thing comes out. You will have a free ebook and audio version of it. So Join the launch list, get the free book, and I'll be expanding on how to develop processes here.