The ROI of AI Agents for DMOs
Based on a bunch of conversations at DI Annual 2025 this month, it's clear that many DMO executives still equate generative AI primarily with writing and editing in ChatGPT. But as one COO of a large US DMO told me, "Surely there's more to all this than just learning about prompting."
So many ways to respond to that.
Meanwhile, there are also DMO leaders who established AI road maps for their organizations going back two years. For them, they're increasingly focused on developing AI-powered automated workflows and their more sophisticated cousins, AI agents. The purpose of these synthetic workflows is to automate tedious, time-consuming and repetitive tasks to free up staff to focus on more high-value work.
Think of it as an evolution from using AI to building AI. You begin by visually mapping what people do in your organization. Then you identify where AI can potentially take over all the mundane stuff better and faster.
I’ve been working with my colleagues at Matador Network to develop automated workflows for our Studio team. We have endless terabytes of client video created during the last 10 years. Now, with AI, we want to capitalize on those existing assets to create new video, but first we need to use AI to understand what we actually own.
We contracted Erick Bonilla, founder of Pocho.tech based in Uruguay, to build a series of new AI workflows in n8n. One of the automations is designed to cull through 10 years of video files in Dropbox and catalogue everything in Airtable. The AI then creates standardized file names, project IDs, links for raw edited videos and finished client hero/cutdown videos, and location and content tags.
The cost to develop the AI workflow was about $2,000 all-in, and it saved more than a month of work at a mid-level manager salary. The beauty of automations/agents is you can measure ROI annually fairly easily, which will help wake up your board if they're sleeping on AI. Moreover, once an automation is in place, the savings will continue to accrue in the future indefinitely for tasks still required, and there’s zero human error.
During a staff Zoom, I asked members of our Studio team if they felt these new agents signaled to them the beginning of the end of some people’s jobs. The response was unanimous. Nobody wants to do this type of repetitive grunt work.
I spoke with Bonilla for his take on the future of AI agents.
Greg Oates: Erick, what services does Pocho.tech provide?
Erick Bonilla: We identify opportunities within organizations of all sizes, where we believe implementing AI systems can dramatically improve the efficiency of operations. We can create workflows that process data, or we can create AI agents capable of reasoning, analyzing data, creating reports, and providing decision-makers with an enhanced way of digesting information.
I have been doing this for almost two years, working with people mostly in the Middle East, the US and Latin America. The Pocha company site is brand new as I try to expand in the world of AI.
GO: What’s the difference between automated workflows and agents?
EB: A workflow is like a mass production factory. It's a set of operations that work through specific logic to automate a repetitive process that a human has done before. This could be writing and reading emails, writing summaries, reformatting and transferring content through different tools and applications, etc.
An AI agent is a thinking environment that makes decisions for you. You can set a workflow for reading your emails. But then you can have an AI agent that summarizes your emails and tells you, if you have 10 emails, what those 10 emails are about in just one email, SMS text, WhatsApp message, Google doc or whatever.
GO: So an AI agent has more autonomy and decision-making capabilities than an automated workflow?
EB: Definitely. The AI agent is capable of being completely autonomous to make decisions within a given context. You give it a context with defined restrictions that narrows the scope of the decisions it can make.
GO: Can you provide any success stories where you worked with a client who had a specific challenge, and you created an automated workflow or agent that successfully delivered a solution?
EB: The most recent example is an EdTech company in Boston that provides a training and placement program for clean energy jobs. They've been operating for the past five years, but there’s been a lack of funds, employees and program applicants to scale beyond Massachusetts. So they decided to use AI and automation to build flows without having to invest a lot of money in hiring people.
What I did was streamline the application process for people interested in entering the educational program. We created a form and a workflow that takes the information from the application and puts it into an application tracking system. This system has workflows that can trigger emails, daily reports, weekly reports and invitations to interview.
Then I created an AI agent. Once the applicants are interviewed by a human, we take the transcript of that, analyze it using AI, and build a report for the decision maker so they can decide if this person is capable of entering the program or not. We also created a scoring system. We built it based on human criteria, which is subjective, and then we converted that into an objective dataset so we could classify people and score them based on numbers.
All of this is done by an AI agent that sets a score for the applicant at the end of the interview. Then we have a leaderboard where people are ranked, and the top 15 get into the education program and placement process.
GO: Were they happy with the end result?
EB: Oh yeah, very happy. Here's where the beauty and magic come in. Now the CEO can envision a way of expanding outside Massachusetts, and tell her team, "Hey, we're going to New York!" Where before, she thought she needed to hire more people in each new state to oversee the operation. And, because an AI agent works 24/7, she doesn't have to wait until 8am to get a report or transcription, or anything anymore. So now, she can think about how to scale her company more successfully without having to hire more people.
GO: What's the road map looking ahead? Where are you going with the company in terms of next steps?
EB: We're trying to create an outreach system because the purpose of the company is not only to train people but also to place them in clean tech job positions. The outreach system is going to help the team find more companies with open positions, so the company can place more people in those positions.
We’re building a system involving AI agents and workflows where we're going to take a picture of all the open positions and analyze and understand them using AI. We want to know, what are the requirements in terms of learning and knowledge capacity for people to get in these positions, so the company can build a better training program.
GO: What was the biggest challenge working with them?
EB: I think the most important one, and this is something anyone working on this will experience, is the resistance of some team members who don't trust the performance of AI enough. People who have low experience working with AI, and maybe only know ChatGPT, will look at this and say, "I don't believe this can work."
So that means there’s a lot of educational work. People need enough knowledge about what AI tools are capable of to suggest what can and can’t work. I think that's the main challenge. But once people see it working, they start to lean into AI. Suddenly, they become like, "This is great. Can you add this other metric? Can you make it do this other thing too? Can you make me one for my personal side hustle?"
It's interesting when you see people moving from resistance to acceptance.
GO: Working with you has firmly cemented something I've been preoccupied with this year. It's this: One of the most exciting things about AI, which is giving us more time back during the day, requires the most boring thing, which is having uniform file and data structure. As we've discovered together only too well, AI can't reach its full potential unless you have everything structured super cleanly.
The analogy I often use is that AI is like an 8-year-old savant. The person is brilliant but he or she often needs really consistent environments and clear guardrails, or things go sideways or just crash. Can you talk about what any organization needs to focus on when it comes to file and data structure?
EB: I think one of the most basic things, and I'm living this with a company I'm helping right now, is documenting files and folders exactly the same. You must set protocols for everything when you’re building a folder system to store content and data.
That's one of the biggest challenges for companies trying to implement AI automation with poorly standardized systems. When we didn’t have AI a few years ago, we didn't think of a machine reading our documents or files. We never thought that could be possible, so we never really focused on creating really good file structure.
The biggest challenge for companies, coming back to Matador, is how to organize the field of play so that AI doesn't struggle. We easily lose the excitement of implementing AI in an organization when an AI is trying to navigate a maze of poorly organized files.
For that, you need to build SOPs that tell everyone on the team, “We're going to be naming files this way now so that a 5-year-old can read the titles and understand what's in every one of them. Because if a 5-year-old can read this set of folders and understand what's inside all of them, then AI can do that too. And if AI can find and read everything, then AI can do some truly amazing things and save companies a lot of money.
GO: So how are organizations like ours, which have been in operation for a long time, supposed to go back decades and reorganize all our files?
EB: We can run a workflow that can do that for you. That's where the ROI comes in. You can make a one-time investment building a workflow that's going to completely standardize all your filing and foldering. But if you do that, you have to know that the ROI might not show up right away. It's going to start paying off later in terms of the time, money, stress and everything it’s going to give back.
A CEO must understand that the return on that investment is going to be ongoing through the years, not in the next quarter necessarily. If you have that mentality, then your company is going to become AI-first.
GO: You’ve told me before that AI is not a tool, it's a culture. Can you explain that?
EB: Right now, the biggest obstacle for AI expanding in many companies are the CEOs. I understand the challenges because they fear a lot of reputational risk. You have a board and customers that are going to be judging you. If you don't have a very good AI road map, then it could turn into a failure, and the fingers are going to be pointing at you. I understand that completely.
But AI should be a culture within companies, and that starts top down because the CEO has to embrace trial and error. AI is going to fail at least during the first time you start with any automation. It's like when you're building a prompt. We were talking the other day, and you said using AI is not about writing the perfect first prompt. It's that fifth or sixth prompt that's going to get you to the right answer. You're going to be iterating with automation, too. That's something CEOs have to transmit throughout the organization.
GO: I think for many tourism organizations, we haven’t even begun talking about the big picture value of AI. From an internal perspective, for example, agents offer a whole new level for optimizing operations with AI.
EB: Yes, exactly. When we're in a company and we're facing challenges, the first intuitive thinking is, "How can I solve this? I'm going to Google it.” Now we're transitioning to, "How can I solve this? I'm going to ask ChatGPT.” This is great, but that's not AI-first. That's changing one tool for another. You're still browsing. It's literally the same thing. You're not really using AI.
AI-first means, "I do this thing that's consuming 10 hours a week. I do the same thing every time. I spend two hours every day doing the same four steps. How can I use AI to automate this?" That's AI-first because you're using AI to stop wasting time. I think that's the culture we should be striving for.
If a CEO injects the whole company with that mindset, then the employees are going to eventually use AI to solve their issues more independently and autonomously. At the end of the day, what AI really does is gives us more time and more of our lives back, and we have not yet understood that because it's new. But it's true, AI is the way for humans to have more life.
GO: I want that.
EB: We all do.
GO: I have this belief that eventually, in any organization, all meetings will be recorded and deposited into a central knowledge base where an AI can summarize everything and connect all the key takeaways. I might have a meeting and someone on the other side of the world has a meeting, and the AI will know we're talking about the same thing. It will see and surface how there's potential opportunity to connect our efforts. It's like one plus one equals three, or more. Do you think that's where we're going and that it will benefit companies?
EB: Definitely. It's the most common sense thing to do, but I know it's also counterintuitive because there are many people who would not like to have all conversations recorded.
The fidelity, the accuracy of the information, and it all being in one place and uniformly structured, is everything. You can have thousands of conversations and tell the AI, "Go here and search for this specific idea in all the transcripts where we have talked about that." You just need to build the SOPs. Then a human can quality control it and say, "This is right. This isn't right. Maybe we can tweak this, or build on that."
GO: I’ve experienced this firsthand where people are vehemently against this idea of all company meetings being recorded. For many it portends that clichéd dystopic future we see in science fiction, where everything's controlled and you have no freedom, right?
EB: I can see it too. We must remember that these tools can be turned off. If we let the technology control us, then that's on us. I think at the end of the day, it's about how do you balance the risk and rewards? This is something CEOs are used to. Do you prefer to live in blindness and have an uncompetitive company because you decided not to take that risk with automation? Or do you take the risk and have everything recorded, even though some people are going to say it’s Orwellian in a way, and even though ultimately it's going to improve everybody's life?
We have to find that balance, and that’s the CEO’s role to determine what’s best overall for the company, customers and employees in the future.
GO: Any last thoughts?
EB: I think we are not quite there yet to know for sure the amount of ROI that is going to come from agents and automations. Definitely it is going to be positive ROI. For that, I'm completely sure. What I'm not sure is how do we measure ROI based on the anticipated lifetime value of the process? So much is changing so quickly and it's difficult to see the future.
As the tools get better, then the workflow gets better too, so the ROI increases, right? Except, we don't really have models for that. For CEOs, I think, they have to try to envision how they can measure the value of automations now, and how that might evolve exponentially in the future.