

How to Ask Better Questions of Your Data, with Kevin Dugan
Directionally Correct: Data 2030 is an educational podcast for data and analytics professionals, hosted by Lucas Thelosen, co-founder and CEO of Orion by Gravity.
This episode was recorded live as a LinkedIn webinar. Lucas talks with Kevin Dugan, VP of Analytics & Marketing Intelligence at DMi Partners, about a 20-year career in digital marketing, how AI is changing attribution and client reporting, and why the analyst's job now centers on asking better questions.
What you will learn
- How combining affiliate platform data, Google Analytics, and click-path analysis gives a fuller picture of attribution
- How AI helps analysts find patterns in large marketing datasets in far less time
- Why AI adoption can dip before it improves, and how teams mix leadership direction with bottom-up experimentation
- Why prototypes come together quickly while maintenance and production quality stay difficult
Resource mentioned
"Your AI Marketing Transformation Will Get Harder Before It Gets Easier," Harvard Business Review: https://hbr.org/2026/10/your-ai-marketing-transformation-will-get-harder-before-it-gets-easier
Chapters
0:00 Welcome and how the host and guest met
2:30 Twenty years in digital marketing, from paid search to analytics leadership
5:44 Finding the right questions, and multiple lenses on attribution
9:49 Using AI to find where email engagement levels off
11:02 Turning call recordings into client health insights
13:07 Claude, Orion, and why AI adoption gets harder before it gets easier
17:24 How data teams and the analyst role are changing
19:32 Spreadsheets, semantic layers, and analytics power users
22:09 Who owns context, and leadership versus bottom-up adoption
24:36 Predictions, from the analyst role to marketing to AI agents
27:15 Starting out in 2006 versus now, and advice for newcomers
29:43 Staying engaged with AI, and why prototypes are easier than production
32:21 Closing thoughts
Transcript
Welcome and how the host and guest met (0:00)
Lucas Thelosen
This is interesting, to do this live stream. I was just talking to a guy who has been doing this for a long time. He's like 200 episodes in, and he was saying how he's now debating doing podcasts, because then you can do some editing, you know, to cut out some of the ums. But I don't know, it's nice to have an authentic conversation, just chat about life. So thanks for jumping on and doing this.
Kevin Dugan
Of course. Thanks for having me.
Lucas Thelosen
I guess we can just kick things off. Maybe a little background on our relationship here. We have known each other now for, has it been two years?
Kevin Dugan
I think so. It's probably Google Next.
Lucas Thelosen
Google Next.
Kevin Dugan
'25 or '24, I don't know. Yeah, it was that.
Lucas Thelosen
You were in line for the coffee, and we started chatting.
Kevin Dugan
John at BCOne, who I spoke to yesterday, actually, so we're still in their ecosphere, which is fun.
Lucas Thelosen
He's a great guy. Awesome. So it has been two years, and we talked back then. We were just starting out here with Orion. And then about a year ago, we started to engage. One of the things that really blew me away when you told me about your background is that you went to Harvard at the same time as a good friend of mine. So this was a great class coming out of it. And then you went to a company and you stayed there for over 16 years.
Kevin Dugan
20 now.
Lucas Thelosen
20 years. Congratulations.
Kevin Dugan
Yes.
Lucas Thelosen
That is amazing. It's so rare, especially for a young, ambitious guy, to stay somewhere for 20 years and have a formative impact on an organization. So I'm super excited to have this conversation with you today. Tell us about your journey and your thinking as you did this.
Twenty years in digital marketing, from paid search to analytics leadership (2:30)
Kevin Dugan
Yeah. It's definitely not typical to be somewhere so long, so maybe not that relatable, but I do feel like I've been lucky to wear many hats through my time at DMi. I started in the fall of 2006, and for a handful of years I was on our agency services team, as we called it. So I was running paid search. I kind of predated paid social. That's how long I've been doing this. We were on early Yahoo paid search and Google Ads. Eventually Facebook ads came along and I ran those, and I ran SEO and managed a couple of e-commerce websites on Shopify and Magento for clients through the years.
Eventually I came to manage that agency services division, and things were humming along. Then at the end of 2023, the rest of our leadership team and I had our annual two-day meeting, where we really get into the bones of the business and think about where we're headed. We made some pretty big shifts to our leadership structure and the roles we had. One thing we carved out was the need for a head of analytics, because we were a little scattered in how we were delivering reports, and it was kind of checking a box. We were sending things in Excel and PowerPoint, and we had folks in Power BI. We were all over the place in how we were delivering insights and data, and I think we were behind the eight ball. So we decided, what does it take to become best in the industry at analytics? We need somebody focused on this. That was always a passion of mine. I was always kind of the Excel nerd here in the company, and I stepped into that role.
We rolled the agency services into what we now call media services, which sits alongside our affiliate division, which is one of our biggest services in terms of total clients and revenue. That was a nice fit for the services I was managing, and I was able to run analytics full time. By stepping into that role, I was able to really focus on centralizing and extending what I had been doing on my small team across the whole organization. That was a really exciting change, and it felt like a career shift at that point.
Then within about six months, we settled on Looker as our next-gen solution to centralize reporting, take it to the next level, and get everybody on the same platform. Now we have Looker running all of our reports across the whole organization, and through a series of connections, that's how we came to work with you guys, through the Looker consultants we were working with. So now the next phase of what I'm doing is, I have the same analytics and market intelligence role, but as everybody in this role is experiencing, it's a whole lot of AI adoption and acceleration. So it's been a fun journey, but that's the longish version of how I got here.
Finding the right questions, and multiple lenses on attribution (5:44)
Lucas Thelosen
Yeah, that sets us up really well for what's next, the future and everything. But before we dive into the future, in the present and the past, one of the things that's always really hard for someone working in analytics, and I was head of analytics, I ran analytics teams in various companies, is knowing what good analytics looks like. What are some interesting findings people have found that would help me make a good case internally that we should invest in analytics? It really does make a difference, because there aren't that many stories out there. I talk to professors who teach analytics, and all they do is say, "Here's how you use Python, and this is how you do a regression." They don't necessarily tell you, "This is best-in-class analytics. This is how Uber became successful, or what Amazon did for their supply chain." So one of the questions I always have to ask is, what has been something impactful, something you did or found, that might be a good takeaway for someone listening?
Kevin Dugan
Yeah, it's a great question. A lot of what we're working with now in analytics is not necessarily how to get to answers, but what questions we should be asking. Getting to the answer is almost the easy part. Now it's what I can do with this data that will reveal an interesting insight, either for DMi or for our clients. Almost all the work we're doing is client-facing in some capacity.
An example of that that we've been working on is telling different stories of attribution in the marketing space, where there are a lot of different sets of data counting revenue for an affiliate program, for example. We have data coming out of Impact or Rakuten or whatever affiliate platform a brand is in, and that's one set of revenue. It's going to be generally over-attributed, because these walled-garden marketing platforms only know about the traffic they generated. They're going to count every last click toward a certain publisher or a certain campaign. That's one set of data. There's Google Analytics data, where there's data-driven attribution and last click, and that's another set where credit is a little more shared across channels.
But we've been doing some cool analysis with the click path data coming out of GA, where we export the click path report, which is kind of a nasty set of data. It says there were 100 conversions where the click path was Google organic, email, and 98 conversions where the click path was affiliate, email, organic, paid search, and so on. It's just this mass of how conversions are generated. So we're plugging that into Claude. In our case, we've gone with Claude as our AI platform, and we have Claude extract influenced revenue for various channels. So we're looking at affiliate and specific channels of publishers and saying, okay, here's one set of data where the affiliate platform is claiming $100,000 in revenue for content publishers. GA is attributing $65,000 in revenue. If we look at everything that content touched with this click path, we have $120,000 of influenced revenue. So it's about creating more lenses with which we can answer questions about performance.
I was on a webinar with Triple Whale last week where their main point was, yes, have a source of truth with your data, but also have other lenses ready with which you can view your data. You may be asking one question where one lens is appropriate, and you may be asking another question, maybe incrementality, where you look through a lens of new customer acquisition, not total revenue generated. So there are a lot of different ways we've been using those large data sets to have an impact.
Using AI to find where email engagement levels off (9:49)
Kevin Dugan
Another good one is an email client, looking for some stats on when engagement really falls off in a user journey. A user signs up for an email list, we get an email every day, and a couple of journeys are happening behind the scenes that are triggered based on our use. At what point does engagement plateau? We had a really big set of data, all sends, all opens, all clicks by user, with timestamps and when they subscribed as well. Again, we put that into Claude and explained where we wanted to get. We got this very clean curve of email opens and another curve for email clicks, and the data told us that at email five, you're basically getting all the clicks you're going to get and all the opens you're going to get. So we're able to really prioritize those first five emails, which need to be dialed in and personalized, doing all we can. In the past that may have been a week-long project that gets dumped into an analytics queue, and eventually we can turn that around in Excel. But the acceleration and speed with which we can deliver these insights is really cool.
Turning call recordings into client health insights (11:02)
Kevin Dugan
The last one I'll give is related to call recordings. Everybody has their note-taking app and their call recording app that joins calls, and that's great for pulling out notes and to-dos and insights. But what we've been doing is taking our call recordings and using them to generate a client health report. We had been doing that with more subjective measures, where account managers would say, "Okay, this client is a four on general client health and a five in terms of likelihood to renew, and here are some notes related to our last conversation, and here's the return on ad spend for this client. It's great," or "it's trending down," or whatever. That was one way we were doing client health. Another way that seems to be almost better is putting an AI model on those call recordings and pulling out sentiment analysis, and our conversations coming up about contract renewals, and having that turnaround. Client health has been really powerful and almost better than the manual work the account managers were doing.
So the takeaway I would give here is to think about places where you have large sets of data and an old process, or no process, to get at them. Then think about what questions you could be asking of that data, and have that light bulb moment that everything is possible now thanks to AI. In the past you may have had a cool idea and thought, okay, that'll be impossible, to listen to the 40 hours of client calls generated every month. The tables have turned in terms of how much we can do. Hopefully those are good examples.
Claude, Orion, and why AI adoption gets harder before it gets easier (13:07)
Lucas Thelosen
Yeah, no, this is great. I couldn't help but notice you mentioned Claude a couple of times, on how you use Claude in your analysis, and I assume analytics-minded people are using Claude with the data. I know, of course, you're using Orion as well. Where does that come in? Where is that important, and why that instead of Claude maybe, or where do you see the advantages?
Kevin Dugan
Yeah, I think where we are right now is that there's a very good Harvard Business Review article that came out last week, and I'm going to try to pull it up so I can get the title. My boss, who's also a Harvard guy, shared it. The title is "Your AI Marketing Transformation Will Get Harder Before It Gets Easier." I would recommend everyone check that out. It's really on the nose in terms of where a lot of organizations are on AI adoption. The thesis is that there's a J curve of productivity when brands are adopting AI and investing in IT. It's the paradox of productivity, I think it references, where productivity goes down when incredible new technologies are introduced. The light bulb was one example. Productivity actually goes down because we're still holding on to old processes and trying to implement these new technologies on top of them, and we're maybe overspending a little bit. Then we hit our stride a little bit, everyone's sharing skills, efficiencies are being gained, and we're really replicating work. Then we start to really improve when we get into agentic system development, where we're removing the human touch from a lot of the processes that don't need someone clicking "execute this skill" in Claude.
So I think where we are is at that tipping point of getting our legs under us in terms of Claude. We've got a lot of users in there, we've got a lot of good skills going, and people are finding efficiencies. But I think the next phase, where we really find a lot, is the agentic path of having agents do a lot of this work unsupervised, still with reviews of course, and anything client-facing is getting eyes on it. The manual work that was taking up a lot of time, I think the next phase is removing the human clicking from the elements that get it done. So that's where we are.
The Orion piece has been awesome, where, again, it gets back to what questions we should be asking. Orion has been super valuable in surfacing those questions proactively. Our team can log in and Orion will say, "We noticed a downturn in AOV over the last 14 days compared to the prior period. Dig into this?" It's a quick yes, and you can deliver some insights on that very quickly. It's a shortcut to getting the right questions to ask. That's the internal side.
On the external side, the Orion piece has been great because I think there's a gap in the market for AI-based automated strategy for brands that may not have a place for a full-service agency to run their marketing. In this case, we've built it for affiliate marketing so far, but it's something where we can leverage all of the learning we've done over the years. We have hundreds of pages of training documents in the Orion instance for brands to ask questions about their strategy overall as well as their performance data, because we're hooking up to their data through our Looker environment. So it's been a fun journey of getting embedded with Claude while also building out this Orion instance.
How data teams and the analyst role are changing (17:24)
Lucas Thelosen
Okay, so then you use Claude for maybe some ad hoc analysis by the data team, and you use Orion where it's more curated and guided and controlled, especially when you do external work for your customers.
Kevin Dugan
Yeah.
Lucas Thelosen
Okay, that makes a lot of sense. In that case, how do you see teams changing, and the job of the data person changing, in this new paradigm?
Kevin Dugan
Yeah, I think the data team is way more spread across the organization than it's ever been. We're a small team at DMi, so I'm sure there are still traditional data teams with data engineers and data scientists and data analysts, with everybody fitting their role. But what's maybe more relatable for smaller organizations is that the data analyst skill set is now extending to folks where that's not really their job description. Whether or not someone has a strong analytical background, they can still contribute and find insights that previously would have just gone to IT or an analyst on a data team to put into a queue and turn around.
Now the role that I have, and that a data analyst would have, is figuring out how you can enable everyone at the organization to find answers on their own, in a quicker and better way than ever before, while also making sure those insights are accurate and grounded in real data. There's still that power behind the semantic layer, of identifying what metrics mean and how data is joined in an accurate way. But the data analyst can do so much more now that the ceiling has been raised so much, and the concept of a data team is much more flexible than it used to be, I think.
Spreadsheets, semantic layers, and analytics power users (19:32)
Lucas Thelosen
Early on you said you were a spreadsheet guy. My happy place is still the spreadsheet, many times. Things are black and white, and you get to play around and find things.
Kevin Dugan
Yeah. Sometimes I'll have something where I think Excel is the right thing for it, and it's a fun time being back in Excel. I'm a little rusty with my latest and greatest formulas. Lookups, I feel like, aren't even the right path now, but it's ingrained in me to do that. So it is nice.
Lucas Thelosen
Well, yeah, we had this whole thing where you're not allowed to touch the mouse, you can only use the keyboard, all the shortcuts.
I think from what you're saying, and from what I'm hearing from other people too, the analyst now can do so much more, as you said. The way I translate that is that you're highly leveraged. But you can be highly leveraged because you set up the foundations. You mentioned the semantic layer. I also think a lot about context, business context, which you have in your case in Orion, but you also use Claude internally and use it to actually service your customers. So you have to have the right context in there, who this customer is, what they care about, what template they should be using, all this information that needs to be set up, and the data foundation. Those two pieces come together, the context about the person being served and the data that is actually accessible to this person, and you bring that together. So you really become a context architect, or data architect, or whatever the right phrase would be for that. Do you have a title that you use internally already?
Kevin Dugan
No, not really. I have my analytics power users, or my analytics Avengers. You have these people within the company who kind of rise to the top. In my email team, I have someone who's generating great skills and running scheduled tasks, and when we look at the analytics of the team, people are using it more and better than others. You kind of pick these people out from the organization and have an AI team that I meet with every couple of weeks to brainstorm and do show-and-tell. We don't love titles for these things, but you get your power users and assemble.
Lucas Thelosen
Avengers.
Kevin Dugan
Your Avengers, who you can really lean on in each service. So that's been our model.
Who owns context, and leadership versus bottom-up adoption (22:09)
Lucas Thelosen
In some companies, maybe the engineering team now owns context engineering. In a lot of the organizations I talk with, the data team was pushing the frontier, the envelope, on context and how context would be structured. I don't know if that's been the case in most organizations. It's not settled yet who owns what and who controls all the agents running around. That's another question. How is that in your organization, with the analytics team pushing the thinking on how context can be managed, versus maybe engineering?
Kevin Dugan
Yeah, within our instance, it's been coming from leadership, in terms of recommending the tools we use. There are kind of competing mindsets of bottom-up growth versus top-down instruction, and I think we've been a little bit of both. Our leadership team decided on Claude, and our CEO Patrick and I are probably the most hands-on in Claude Code, working on agents and trying to take the next step. But there's still the bottom-up approach, where really cool time-saving things are going to come from the people who are actually using and building them, and that's a great way to get adoption across the organization. It's like a peer telling you, "Hey, I just saved myself two hours a week on building agendas because I built this skill, and it uses Looker, and it uses our brand skill, and it uses PowerPoint. It's a great agenda that I can get generated and review and put in some insights from Orion." I'm generating better decks than I ever have, and they take me half an hour or 20 minutes, pulling out cool insights, instead of an hour where 50 minutes is dragging elements around in PowerPoint. So I wouldn't say we're necessarily typical, and I don't know exactly how most organizations are adopting this. But as a smaller data team, we're guiding the team from the leadership level while also relying on really passionate users to find their way.
Predictions, from the analyst role to marketing to AI agents (24:36)
Lucas Thelosen
Yeah. As you can tell, I'm doing this whole thing in essence like an analysis. Looking at the past, looking at trends, looking at the present, and now, what is the future going to look like? What are your predictions? What is your directionally correct forecast of where things are going to go in terms of people and technology? Feel free to swing wide here. It's a low-risk environment, but I think it's always interesting to at least have an educated guess.
Kevin Dugan
Yeah, I think in terms of the role and the job of the data analyst, that's kind of going away already. It's almost like prompt engineer and some of those things you were throwing out, in terms of what the role actually is. People sometimes joke, "My job is Claude now." It's an exaggeration, but the idea is that everybody can get to the answers super quick now, and that's no longer the challenge. It's knowing what data you have and how you can pull out interesting insights that will help with whatever you're doing, in our case digital marketing for brands, and how to turn those into accurate and actionable insights. That's now the job, I think.
In a more general prediction, I think the future is agents and building agents, but also marketing to agents and assuming that agents are going to be browsing the internet, and they already are. It's kind of like when we were all building websites for mobile, and everything had to be responsive because people were using their phones. Now the internet is going to become primarily agentic rather than real people. How do we respond to that, and have those agents see what we want them to see? That means GEO, and coming up with how our brands can be recognized in AI searches and how their products can be recommended. When someone tells an agent to order a certain amount of toilet paper, how does the agent pick what to buy and what to recommend? I think the next phase of marketing is marketing to agents, almost. That's where we're headed, and it's a little daunting, but there have been many daunting moments in marketing history where we had to figure things out.
Starting out in 2006 versus now, and advice for newcomers (27:15)
Lucas Thelosen
Well, you started out in an incredibly interesting time, in 2006. I remember, as a first job, you get the first job and you're so happy you got that job. I was an analyst, but I was also a young guy, so I was like, why don't we do some social media stuff? Everything was a cent, super cheap, and nobody had figured anything out yet. It was an interesting time to be starting out. In some ways, here we are 20 years later, and once again it's an interesting opportunity to figure things out. This is uncharted territory. Nobody has written the rulebook on how things are, because it is evolving and changing. Anyone who would write a book right now about it has to update something in three months. So it's a good time to be in the wild west and just embrace it and learn as much as you can. On that note, imagine you were starting out right now again, after an incredible 20-year journey at DMi up to VP. What would you tell yourself if this was your starting year?
Kevin Dugan
Yeah, it's a good question. Professionally, I'm glad I've stayed where I am. I wouldn't want to do anything differently in terms of the journey I've taken at DMi. But in terms of personal growth and trends, what I could have done better is probably to just be on the front lines of AI. I think that's the obvious answer. Being ahead of the curve is the skill now, where you can be a step ahead in being able to develop agents and think about what's next. I think being on the front lines of that would be the greatest skill I could recommend. And thinking about software from the command line. That's a lot of the skill set that I would have had as a developer, and as an account manager I don't have as much, but I think that's a super valuable skill to have. So in general, staying on top of AI as early as possible would have been the best advice.
Staying engaged with AI, and why prototypes are easier than production (29:43)
Lucas Thelosen
Well, sometimes it seems daunting, because every week something is announced. Not just every week, every day something is being announced. But at the same time, I was just looking back at the last three years. We're coming up on three years of ChatGPT being announced, and I remember all my friends were doomsday, like, "Jobs are going to be gone by January. What's going to happen to the world?" And here we are three years later, and the world is of course changing tremendously. I do think it's good to be engaged and actually not hold on to the past too much, because the train has left the station. It certainly has. You're much better off getting experience with how to manage context and how to think like an architect for this, versus being the passive recipient and the guy who's pushing back the whole time.
It's very similar to cloud adoption maybe in 2013 or so. I remember giving a talk in Berlin in 2015, and these people were so hesitant. They were saying, "No, no, no, I think it's going to be on premises. It's going to be Hadoop." And I thought, guys, this is 2015. Maybe we could have had this conversation in 2012, but this is a long time over, and this is what's happening right now. Again, you don't need to be reading every single tweet out there, but get your hands on the technology and get your experience up. It's funny, one of my engineers was telling me I should build my own little product on the side, and I had a lot of fun with that. It's so fun to just learn the challenges. It's so easy to prototype, and so hard to actually fine-tune and get it just right. I think the main reason he told me that is so I would really appreciate all my engineers too. But it was good.
Kevin Dugan
See how hard this is?
Lucas Thelosen
Yeah.
Kevin Dugan
I was talking to somebody last week, and they said getting 90 percent of the way there is really only getting 10 percent of the way there. You think you have a working prototype and it's a really cool thing to show off, but how are you maintaining the code? And if you need to market it, getting users is obviously the hard part. But even with an internal tool, we've noticed on our side that we can get something really cool to show off to people, and then somebody asks for a little tweak, and that's really hard. How do we make that tweak? We need to change the whole database. So prototyping is now easy. It's a good exercise.
Closing thoughts (32:21)
Lucas Thelosen
Okay, well, thank you so much for jumping on with me. This was really fun. It was really good to chat with you again. And I think we'll record this and make this available.
Kevin Dugan
Nice. All right, thanks for having me. I had a great time.
Lucas Thelosen
Thank you.

