Why Data Work Compounds, with Frank Tingley

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.

In this episode, Lucas talks with Frank Tingley, Founder and Principal Advisor at Tome Advisory, about the analyst's path from answering requests to owning data products, why small gains compound over time, and why executives should spend three to five percent of their time engaging with the data. Frank has spent more than a decade in data analytics, across Fortune 500 companies and high-growth direct-to-consumer brands.

What you will learn

• How segmenting products across sales channels lifted dot-com revenue by about five percent
• Why analysts are becoming full-stack
• Why context management matters more as AI automates hard skills like SQL and dashboard building
• What to look for when hiring analysts in the age of AI
• Why the 95th percentile of data work is not good enough

Resource mentioned

95%-ile isn't that good, by Dan Luu
https://danluu.com/p95-skill/

Chapters

0:00 Welcome and guest intro
0:31 Career path from finance to analytics and advisory
4:01 Where analytics has driven impact
4:44 A channel segmentation example and how AI shifts the work
8:49 Team structure and the full-stack analyst
13:15 Context management and why hard skills matter less
15:42 How to hire and interview analysts in the age of AI
17:44 Advice for a 25-year-old analyst
18:25 Compounding returns in analytics
23:07 Data-first culture and the urgency of every week
25:40 Why the 95th percentile is not good enough
29:54 What executives should do to engage with data
34:25 Closing thoughts

Transcript

Welcome and guest intro (0:00)

Lucas Thelosen

Hello, and welcome to another episode of Directionally Correct. I'm Lucas Thelosen, your host. Today I'm being joined by Frank, who got to work with me a little over 10 years ago, at the heyday of the modern data stack. Many of my friends in the Looker world will look back to that moment in time as when things really went through the roof. Frank and I got to work together back here in Colorado. I'll leave it to him to introduce himself and give us a bit of his background.

Career path from finance to analytics and advisory (0:31)

Frank Tingley

Hey Lucas, thanks for having me today. When you said 10 years, I was thinking that can't be right, but yes, it was 10, if not more. I appreciate you having me. Like you said, we worked together more than a decade ago. I've spent the past dozen or so years in the world of data analytics and now AI. I talk about my career as a bit of a sandwich. The bread has been two Fortune 500 companies, and all the meat and fillings in the middle have been VC-backed, high-growth, DTC companies. It's been an interesting journey, and I'm excited to talk about all those experiences with you today.

Lucas Thelosen

Okay, well, thank you so much. Where were we? We were the meat in the middle, probably one of those VC-backed companies.

Frank Tingley

We were the meat in the middle, yeah.

Lucas Thelosen

Okay, so what came before, and what came after?

Frank Tingley

Sure. I actually started my career, and I don't like to talk about this much, as an investment banker. I had a passion for it, I think, and I started in the early 2000s because I like turning decisions into business problems and quantitative outputs. That was the way you did it back then. And I knew I didn't want my life to be as an investment banker, so I went back to grad school. In the early 2010s there was this new, weird discipline coming online called analytics, and I thought, let me give that a try. I thought I would go into grad school to become a corporate FP&A person, like a lot of ex-investment bankers do. But this new analytics came along and I fell in love with it, because it got to the heart of what I wanted to do, which is take problems, quantify them or put them in a framework or a model, and make decisions based on that. It seemed really cool.

I got my first job out of grad school working at Dish Network, a Fortune 500 company in Colorado as well, doing marketing analytics. This was at a time when a lot of marketing measurement was just coming online. Think about Google Analytics and attribution, MTA. MMM was just coming online. So I cut my teeth in the marketing world and grew into more of the database world, the traditional SQL-based world we know today. After that I went and worked at Craftsy, where we were together for a few years. I started off as an analyst there but grew to run the analytics team. We ended up exiting that company, like a lot of VC-backed companies do, to a Fortune 500, and I worked there for a little bit. I went to a few other similar companies, VC-backed, PE-backed, high-growth DTC and CPG companies, before ending up at The North Face for the past, I don't know, five, six, seven years. Time flies.

When you look back on the COVID time, you don't know how many years it was. You think, yeah, it was a few years. How long was COVID? I don't know. Same kind of thing. I ran an analytics team, making data-driven decisions, at what I think of as the bottom part of that bread, the last Fortune 500 I was in. Today, I'm taking everything I've learned and helping other CPG and DTC companies navigate this wild world of data analytics and AI.

Where analytics has driven impact (4:01)

Lucas Thelosen

Perfect. There are two things I think everybody in the audience always appreciates. One is what team composition looks like, and what it used to look like, maybe pre-AI. How were you thinking about that, and what is needed to staff a team correctly? The other is examples of where analytics really drove impact in an organization. Maybe we touch first on the impact of analytics in an organization, and then we get to team, because then we can also talk about the transition we're seeing right now. So where have you seen analytics drive impact in an organization? You don't have to be specific, and we don't need to disclose any trade secrets here, but what have you seen be really helpful?

A channel segmentation example and how AI shifts the work (4:44)

Frank Tingley

I'll look back on an example from a recent company I was a part of, and I'll try to keep this as generic as possible to keep trade secrets out. Like most companies that might be listening to this today, we were multi-channel, omnichannel. We were selling on our dot-com site and on Amazon, and we were selling a lot of the same merchandise, a lot of the same products. When I joined the company, the first thought was that we were probably cannibalizing ourselves a little bit, but we also wanted to meet consumers where they are. So there were two conflicting issues at play. We had the biggest e-commerce platform in the country in Amazon, but our margins were a little lower, because Amazon always takes a cut of everything. We wanted to meet consumers where they are, and they're on Amazon. But we also had our own dot-com site, with better margins, where we could tailor the experience a lot better, although obviously far fewer people were visiting our own site. How do we think about bifurcating these two channels and segmenting them in a data-driven way?

This was a classic data analytics exercise. Go in, organize the data, query the data, dashboard the data, compare the two side by side, and, most importantly, work with your business partners on what we can act on. "Hey, I'm seeing this in the data. Is this real, or is this just an anomaly? Can you tell me what happens here? What are the minimum order quantities, what are the colors," and so on. How does the business flow? You can't just look at a dashboard, wave a magic wand, and make things happen. Otherwise analysts would just be running every business, instead of having a marketing team and a product team and a merchandising team and whatever other team.

So we went through, and based on a lot of different findings, we started bifurcating our product segmentation between our own dot-com site and the Amazon site. That came about through a lot of dashboarding, a lot of working cross-functionally with partners, and a lot of looking at numbers, making judgment calls, and being business leaders, not just doing data querying. The result at the end of the day was about a five percent lift on top-line revenue in our dot-com business. At a company that scale, you're talking about tens of millions of dollars of top line. I think there was also a couple million dollars of cost savings from not having to ship to different channels.

This was in a pre-AI time, so a lot of the work was on the data side. I'd say roughly half, maybe 60 percent, was organizing our data, querying our data, getting it in the right places, and dashboarding it correctly. The other 40 to 50 percent was working cross-functionally with our business partners and setting context, getting the business rules we worked out with our merchandisers, wholesale partners, and sales partners back into our data layer. If I think about today, that 50 to 60 percent we spent on the data work, a lot of that has been automated away. If I did this today, that would probably be 10 to 20 percent of the work. The context management, the business decisions, working with stakeholders, and feeding that back into the system would be a much, much larger part of it.

Team structure and the full-stack analyst (8:49)

Lucas Thelosen

What kind of team composition does an organization need? What did they need pre-AI, so to speak, and what might they need post-AI? What is the ratio of data engineers to data scientists, business analysts, and data product managers? How do you see an organization structuring itself? Maybe we can talk first about what a small team, a small organization, would look like, then go to a growing one and where you would actually put the staffing. We don't have to spend too much time on that, because we can quickly shift to what it should be today.

Frank Tingley

Yes. The way I think about it, in the past, and I think a lot of teams are currently set up like this, is that teams are set up left to right in the architecture stack. You start with your data engineers, who extract the data, load the data, and handle the source. Then you do your data modeling, and then you move right, and you move right. I think that's still an important part of what data does. But for analyst teams especially, who used to sit on the right side of the architecture, what's more important today is becoming what I call a full-stack analyst. I think you need to know how to do all the things to the left of the architecture. As an analyst, it's so much less important that you know how to query SQL well or build a pretty dashboard than that you know how data modeling actually works, how data engineering actually works, and how to add context back to all those things. So step one, when I think about how analysts' skills evolve and how that relates to a data team today, is analysts becoming more full-stack analysts. They need to be able to move all the way left and all the way right across the data lifecycle. That's one vector, the left to right.

Then you brought up product management. This is where I'm really bullish, and where I really think the future of the data profession is going. Previously, being a data person was a career with a low ceiling. Very rarely will you find an end-to-end data career trajectory. There's very rarely a chief data officer at every single company, the way every single company has a CFO or a CHRO, whatever it is. I think they say only about 10 to 20 percent have a chief data officer, and those are usually tech-heavy companies or information-based companies. So being a pure data person had a low ceiling. Analysts or data people, whatever they are, used to stay on the edge of the business. They were fact checkers. They weren't necessarily owners of the business. Requests would come in, like, "Can you look up sales for last week and exclude all these things?" You'd write your SQL, export the results, and put them in a dashboard. You'd probably grumble about it. You'd say, "My God, this request."

But I think today, and the future of the role, is that data product owner you talk about. Owning a data product means having real users, having a roadmap, and having an outcome you're actually accountable for. That's how an analyst, and I say analyst meaning any sort of data practitioner, goes from checking other people's work to deciding what gets built, to having an ROI put on your own work, to having actual ownership over the output, which you can take both in your career at your current company and leverage in future opportunities.

Lucas Thelosen

So we have less one-off analysis and more structured thinking, architecture thinking. We have the data product manager on one side and maybe the context architect on the other side. I think that's completely aligned with how I think about the world and where I see things going, so I really appreciate hearing that from your end as well.

Context management and why hard skills matter less (13:15)

Frank Tingley

Yes. Let me double-click and draw a line between two things I just said. I talked for a moment about the full-stack analyst and the data product owner, and I think the two go hand in hand. The reason I started off talking about the skills an analyst needs is that you asked what a team should look like. I think the line and the description between traditional data professions is blurring a lot. Between data engineer, data modeler, data analyst, analytics engineer, and so on, the lines are blurring because of what you said, because of that context engineer, that context management focus. I started this podcast talking about how I got into investment banking, then analytics, and now AI, because I wanted to take business problems, quantify them, and solve them in a more systematic way. That is where the role is going, and where the role is today. I think that syncs up with the context management you brought up, which means writing down what a smart hire would or wouldn't know on day one, keeping it current, and handing it to both people and AI so no one has to relearn it.

In other words, you have to know the business a lot more than you did before. You have to be able to take that business context and manage it, context management, all the way to the left of the data chain, a lot more than you did before. That's what I brought up about being more full-stack, taking the context and moving it all the way to the left. Hard skills aren't nearly as valuable as they used to be, and that's a hard pill to swallow, for a lot of us, myself included. Lucas, based on our time together, you too. A lot of the time we were rated on our ability to execute on our hard skills, SQL, whatever BI tool you were using, whether it's DAX or LookML, whatever it might be, and our ability to build dashboards. That has been automated away for the most part. I'll pause.

How to hire and interview analysts in the age of AI (15:42)

Lucas Thelosen

That brings us to an interesting question. How should anyone think about who to hire? And on the flip side, what can someone prepare to really shine in an interview or in the hiring process? If hard skills are no longer as important, how do you evaluate what would be a great hire in this current time?

Frank Tingley

The actual hard skills are less important than how to apply them. A question I always used to ask during an interview, pre-AI, was, "What's your favorite SQL query, and why?" A little of that was asking whether they knew SQL, but a lot of it was asking whether they could take what they know from a hard skill perspective, speak to it critically, and actually apply it. It was less important how technical the answer was. Maybe the answer was even a WHERE statement. Great. Why? Who cares? WHERE is one of the most basic SQL statements out there. But if you had a good answer, I thought that was golden. The same thing applies today to who you go out and hire. Can you take a hard skill and apply it to context? Or the reverse, can you take context, such as whether sales are gross or net of returns, or whether they include Amazon, and apply that to the data stack, to the left of the architecture, all the way back down to the data engineering, so you can write rules that will ultimately be fed back into the context of a dashboard, AI, whatever your end-user delivery vehicle might be.

Advice for a 25-year-old analyst (17:44)

Lucas Thelosen

If you were to give your own 25-year-old self some advice right now, what would that be?

Frank Tingley

I would say that compound interest is the eighth wonder of the world. Someone much smarter than me said that, so I don't take credit for it. I would apply that both to your savings and your investments. Savings obviously compound over time, but the same is true of what you learn. Tools depreciate, but judgment and the things you learn compound over time. So remember that, and just keep learning.

Compounding returns in analytics (18:25)

Lucas Thelosen

There's actually a related topic I just talked to a really interesting company about. In my tenure at Looker and then at Google, I got to work with many different interesting companies, and there's a really clear distinction between the ones that are successful and the ones that are not. There's a correlation between how seriously they take analytics and how they do, and you can tell quickly who takes it seriously and who does not by, for example, who pushes out meetings. We were there with Walmart during the transition when they wanted to become an e-commerce company, very successfully so. They were far behind Amazon and really did a fantastic job. And then you have some examples, maybe ones I shouldn't mention, of companies that didn't take it that seriously. They kept pushing out the meeting, and everybody on the team was totally fine waiting another month or another couple of weeks.

And that compounds. Two percent every single week adds up so quickly. Before you know it, you have a company that is able to add 20 percent top-line revenue at massive scale, and another that starts falling behind and slipping. You had this fantastic example of an analysis that was able to produce five percent additional revenue. Findings like that are rare. They happen every once in a while, but they're rare, and we have to acknowledge that. But there are hundreds of tiny things you do with analytics, that you improve with analytics, that compound. I got to work with Uber and Lyft at the same time, and it was really interesting to see those two companies side by side. It's fascinating. You could write a whole book on that. I have these stories over and over where the companies that took it seriously, that didn't push out the meeting, that said every week counts, really compounded.

I don't know if you had similar experiences there. One of the interesting things is that probably at VF you got to see multiple brands operate next to each other, and of course you worked at multiple different companies. Have you seen this kind of compounding effect?

Frank Tingley

Yes. And I think the compounding doesn't show up immediately. This investment in data should be treated as an investment. What do I mean by that? I think a lot of organizations treat it as a cost. I put money in and it's gone. It's a million-dollar line item, whatever it is. But it's really an investment. You put in money, and you might not see a return for one quarter, two quarters, three quarters, but eventually your investment will have an ROI. It doesn't happen overnight. It's a bit of a J curve, and there's no real way around that. You do that by building data products that people can self-serve from.

You protect the time. You don't push out the meetings, like you said, because there's always going to be a backlog of those small one-off questions people want answered. There's always going to be a reason not to invest in data-driven decisions and the tools and skills people need. But it becomes an investment over time. You put in a dollar a day, you put in a dollar a day, and eventually you're going to have millions of dollars. I think it's the same thing with your data teams. You invest slowly. I think it was Churchill or somebody who said, what was it, gradually, then suddenly. It's this thing that happens gradually and then suddenly. You have this insight, or all of a sudden you turn around and all these data-driven decisions that you use in your weekly stand-ups or your monthly business reviews or whatever end up changing the culture of your business and how you approach your business and the decisions you make on a day-to-day, week-to-week, month-to-month basis. All of a sudden it's worth 5, 10, 15 percent annually, which compounds over time and is worth 2x in, I don't know, three or four years or something along those lines.

Lucas Thelosen

Fascinating. Absolutely. It's so hard to create the urgency around the compounding, but it is so important. You think about people's personal finances, but it's very translatable to corporations and...

Data-first culture and the urgency of every week (23:07)

Frank Tingley

Well, you said something, Lucas, if you can talk about it. You mentioned Uber versus Lyft, and I think people would want to hear a lot about that. My ear has perked up. I don't know if you can talk about it, but what were a few differences you saw in how they approached data, if at all?

Lucas Thelosen

Well, there are many companies that take data to reinforce existing decisions and the things they already want to do. And then there are companies who say they're totally fine with data showing them something they haven't taken into account before, but those are actually more rare. Uber was very much a company that was totally fine with data showing something they hadn't anticipated yet, something that changes the direction of the company, even in small ways. That was great. There was a very data-first culture at Uber, where it's like, okay, this is what we're seeing right now, so now we're going to do X, Y, and Z. Other companies would say, this is what we want to do, so let's find the data to support this argument. That is one thing.

The second one is urgency, where every day counts, every week counts, where another company might say, we got a lot of funding, we don't need to rush, so let's come back in March and check again. That is a very different approach, and you see that compounding as well. I'm sure I signed a lot of NDAs, so I'm not going to talk about any specific company, but you can draw some conclusions. There is just a cultural difference in the organizations. The data-first company can be tougher to work at, because there are directional changes, and you might get some curveballs. But I think that's okay if you're comfortable being challenged on a weekly basis on your assumptions and understanding and the way you approach things. I think that's a really good thing. It's very similar to being a professional athlete. We all follow some professional sports team. They have a bad game and they reassess. Then there's the bad team that says, let's just forget about the game and focus on the next game, and doesn't take away any learnings. That compounds.

Why the 95th percentile is not good enough (25:40)

Frank Tingley

Yes. Hard agree. I think true progress is incremental gains against harder and harder goals. What you're talking about is incremental gains against harder and harder goals. When you were talking about those really hard decisions that I think you said Uber made, that reminded me of an article, or a Substack, that I go back to every few months. It was written in 2019 or 2020, so pre-modern AI, and it was called "95th Percentile Isn't That Good." I see you laughing, so you might be familiar with it.

Lucas Thelosen

No, no, but it is a classic.

Frank Tingley

Yes. I think that's what you're talking about, and I think it can be applied to data and analytics cultures and organizations. If you have a 95th percentile data organization, that's really not that good. You're one out of 20. That's not that hard. If you stand in line for a concert behind 20 people, behind 40 people, there are going to be one or two who are better than you if you're 95th percentile. Getting from the 95th percentile to the 99th percentile to the 99.9th percentile is the really hard part, and that's where the daily incremental investment and compounding come in.

Another facet of this 95th percentile concept, a little less about org strategy and more about talent and the future of the data profession, is that a good AI dashboard or a good AI program can get your data 95 percent of the way there with minimal work. But 95 percent isn't that good. This is where your talent and your data people come in, getting that last five percent, getting to 99 percent of the way there. You can open a free instance of Claude, feed it a CSV, and say, "Build me a dashboard that looks like this," and it'll get 95 percent of the way there. That's not that good. We've all had that experience with "analyze my data, please." It's not scalable, and it's probably going to make a few critical errors, et cetera. Your data talent needs to take that from the 95th percentile that AI can reach to the 99th percentile, to the 99.9th percentile, whatever it is, so you can make business decisions based on that, so you can find those hidden opportunities.

Lucas Thelosen

Exactly. You've got to go for excellence. It's interesting how we got to work on the consulting side of Google, and at Looker before that, with the companies that were in the most competitive environments, like food delivery for a while, from 2017 through, of course, COVID. Food delivery was ultra competitive, so they cared. An industry that is not as competitive might not care all that much, because it's okay, you're not that competitive. But at some point almost every industry becomes competitive, and then it starts to matter. It's always interesting to see. I got to work with a decent number of streaming companies, and there seems to be a lot of competitive pressure in the streaming market as the market contracts a little bit on streaming subscriptions. You see some competitive pressure in the ad space as well. It's really fascinating to see the heat map of competitive pressure at any given moment.

What executives should do to engage with data (29:54)

Frank Tingley

I agree. We've talked a lot about the analysts on the ground. Can we talk about the executives for a moment?

Lucas Thelosen

Yeah, let's talk about the executives. What does it take for an executive to be successful in this world?

Frank Tingley

Let's talk about someone you and I both worked with. Let's name drop at Craftsy. We'll name drop our COO, Josh Scott. Shout out to Josh Scott. I think he was actually a good prototype for what an executive of a data-driven company in an AI world would look like today. What do I mean by that? If you recall, you worked in the data world, and I did too. When I was there, he didn't run data and analytics, but he was technical enough with SQL. Again, this was the mid-2010s, pre-AI, and before a lot of the BI tools used today. I think you brought Looker online around that time, so we were still heavily relying on SQL querying. But as the COO of a 200-plus person company, he would go in, query his own data, check on things, and ping people. This was pre-Slack, so he'd ping people on whatever the company messenger was, or by email. "Hey, I saw this thing. Can we look into this? What do you think about this? What about this?" For me personally, that made me a much stronger and sharper analyst, knowing that someone was also looking at this stuff. I don't want to say looking over my shoulder, but an executive at the top was looking at the data I was responsible for, the data I was making recommendations and business decisions from. It was probably five percent of his time and job, maybe I'm being generous, but when I saw that email come in, or that Google Hangouts message, or whatever it was, I knew I had to come correct.

So how do I apply this today? I don't expect, and I don't think you expect, all the executives out there to be able to query SQL today. They don't have to. There are BI tools out there today that do it for you. But I think it requires executives in the C-suite or at the VP level to spend three to five percent of their time engaging with the data in a way that goes beyond a dashboard in a PowerPoint. It's actually going and getting deep and sending a little feedback here and there to the data teams.

Lucas Thelosen

Frank, that's a really good point. It has never been easier for an executive to get involved and get their hands on the data, by asking questions of the appropriate AI assistant for a given topic, digging a little deeper, asking a couple more questions, and being more informed as they go in front of their teams. A super common example is that you can tell your favorite AI to review the last couple of sales meetings and see if anything stood out that they should address. On the data side, it's super easy to have a couple of curious questions, have the AI go through your governance layer, ideally Orion, and come back with the numbers you can bring up at the meeting. Not as a way of saying "I'm watching everybody and standing over your shoulders," but as a way of saying "I care about the business. I care about the part of the business you work on." That makes us better at our jobs too, because we know the executives care.

Frank Tingley

Well said. Agreed. It shows that they care. It shows you're not just someone who fulfills requests and pulls numbers, that what you do matters and is being looked at. And everyone has different angles and perspectives on the business. From your most junior new employee to the C level, the more people who can get their hands on the data and look at the results, the more perspectives you can have on the business, and that's a good thing.

Closing thoughts (34:25)

Lucas Thelosen

Well, thank you so much. This was really great. I think we got a lot of really helpful information here for people in the analytics space and in business in general, about how things are changing right now and what kinds of traits are really important to be successful. I want to think a bit more about the compounding component, because that was a new aspect. I hadn't really thought about it, but in this moment it kind of clicked. So thank you very much, Frank.

Frank Tingley

Thank you. Thanks for having me, Lucas. I really appreciate it.