Stop Analysis Paralysis: Getting ROI from Generative AI

Shaown Nandi:
Welcome, everyone. Thank you for joining the Executive Insights Podcast, brought to you by AWS.

I'm Shaown Nandi, head of technology for AWS Global Sales. And I have with me here today Jeff Traylor, who leads the field engineering teams in the Americas for Databricks, a pretty incredible AI and data company.

Jeff, welcome.

Jeff Traylor:
Hi, Shaown. Thanks for having me.

Shaown Nandi:
Databricks is one of our biggest and most important AI and data partners, as well as one of our customers. Jeff, tell us a little bit about your story and what you do.

Jeff Traylor:
Yeah. As you said, I lead field engineering at Databricks. I've been at Databricks for almost two years. Chicago's home. Don't sound like it. Grew up in Georgia, but been in Chicago for 22 years.

And prior to Databricks, I was at AWS. So, I'm an AWS alum. Almost seven years there in a similar role. Prior to that, I was at a leading infrastructure and a SaaS company, ADP, Automatic Data Processing, and a division of theirs, leading global infrastructure, and basically, bringing, transforming what were, otherwise, on-premise systems and taking them to the cloud. And did that for actually a decade and a half before going on to AWS.

Shaown Nandi:
So, Jeff, you and I have been in the same business for a while. You've done it at a couple firms now. Tell me what field engineering is about, how do you help customers, what's sort of the purpose of the role, and how do you see yourselves?

Jeff Traylor:
All right. I get a lot of energy from this question. I'm impassioned. This is the world according to Jeff Traylor. I'll give you that disclaimer.

I see the fundamental role of a field engineer, and this can be a bit controversial. It's primarily three pillars or three dimensions for me anyway. And I would argue that they're perfectly a third, a third, a third in importance and weight. One of the thirds is technical depth and breath.

And this is, oftentimes, when field engineers want to throw things at me because they think, you're asking me to learn all this stuff, it feels like my whole job is technology, and I see myself as a technologist, that gives me energy. But the job of a field engineer, for me or an SA, is one third technology depth, so you have to have the chops.

You have to have the skills. But the other third is you have to have that deep customer-centricity that, really, unpacking the customer, understanding the customer's customer, really having the vision and really, deeply understanding what your customer's trying to accomplish, not necessarily what they're asking for, but like five-wise, unpacking what they're trying to accomplish, why it's important to them, how they measure success, who their competitors are, who their customers are, what are the conditions by which they're currently operating, what is their ability to execute?

So, once you've deeply understood the customer and what they're trying to do, what their ability to execute is, then, you can, then, pair that with a technical solve, and you can have the technical imagination that they may not, otherwise, have, because they're doing their job working in their industry, they don't get the benefit that you or I to get to see across many different customers.

So, you can bring that technical imagination, understand what happens, pair it with the technology that you have. Okay. So, that's pretty powerful. But still, you're not done. The last third is, now, you've got to go communicate it and influence it, and give the customer the confidence and the conviction that, yes, we're going to do it. We're going to do it with you. We're going to partner with you. We're going to go faster than we're comfortable going.

We're going to think bigger than we were otherwise thinking, and we're going to get started. And we're not going to do the analysis paralysis. We're not going to wring our hands perpetually. But we're going to get going, and this is how we're going to go about it.

Shaown Nandi:
I love this breakdown. So, having the technical chops, having the business acumen and empathy to really understand where the customer needs and where they're coming from, and then, having that voice to be helping them on that journey and to be convincing them that they just need to move it forward in a way that helps them.

Well, excellent. I mean, you've seen it all. You've been a customer, you've helped customers from AWS, now, you're helping them from one of our partners. What are you seeing happening right now in the industry trend-wise? What is Databricks thinking and doing?

Jeff Traylor:
Yeah. Databricks is really... Our foundation is we're a data and AI company, that's kind how we were founded, or we're a founder-led organization. All the founders were working at Berkeley Labs, PhD students or faculty, launched Databricks. They were data scientists. And so, they basically... We’re trying to come up with ways to make data science easier. So, they come up with Spark.

So, one of the founders is the original committer of Spark.

Shaown Nandi:
That's awesome.

Jeff Traylor:
MLflow as well. Delta format. So, it's a very open-source lineage, originally, built on AWS from the foundation. But it's really, fundamentally, at our core, is about simplifying the data experience, taking what is otherwise a very disaggregated estate, oftentimes, in large enterprises, difficult to manage, difficult to secure, and basically, trying to harmonize it as much as possible, simplify it, keep it secure, and then, help customers ring value from it. That's fundamentally what we're about.

I would say, most recently, probably, the explosion that's happened in the last 18 to 24 months is generative AI has kind of hit the world by storm. And so, companies obviously have huge ambitions around AI, and generative AI in particular. As we know, there is no real generative AI outcomes without data. I mean, data, it is the classic garbage in, garbage out. So, it's absolutely foundational and central to customers realizing those capabilities. And so, that plays right into kind of what we're about and how we go to market.

Shaown Nandi:
Data is absolutely king. You know, I've heard from a lot of customers, they are anxious to have that simplified journey where they can. How are you helping customers get to sort of outcomes from their data? What sort of things is Databricks unlocking and how are you doing it?

Jeff Traylor:
Yeah. To me, it starts... It's not dissimilar to when I was at AWS. I think there's a lot of common themes, right? And one of the things we talked about, that Andy used to talk about at AWS was really just get started. Just get going. Get started, don't try to boil the ocean at first.

It's great to have a huge ambition, it's great to want to innovate with the greatest innovators on the planet, but first of all, just get started. Stop digging. Get started. Work with an AWS, work with a Databricks, work with a partner that will help you that's done this thousands of times for customers, help you avoid the easy traps and avoid the pain of change that happens.

But the first thing I would say is to like land your vision, like, what do you want to accomplish? Anchor it to a business outcome. It's not just pursuing the latest Vandy project or the latest technology trend, because that really, ultimately, doesn't really serve you. If it's not anchored to a business outcome, you're not going to succeed and you're not going to be doing it for very long.

So, number one is, what do you want to accomplish from a business standpoint? So, cast the vision, and then, communicate it, like, what are you trying to do, why it's important. Bring your organization along, because so many, I've seen so many technology transformations and initiatives never get off the launch pad, because either they started too big in their vision, and it was analysis paralysis-kind of concept, so they really just can never get off the dime, like where to start. It was a perpetual debating society, and/or they didn't bring people along with them, they didn't bring the organization along.

But really what it's about is trying to... Most transformations are around disrupting status quo, and trying to leap ahead to what's possible. And most of the organization understands the pain of status quo, because they live it. So, they're ready to do it, but it's just human nature, it's like, what does it mean for me?

So, there's going to be some reluctance, and I've seen, too often with companies, big initiatives and a lot of fanfare, and they kind of burn out.

Shaown Nandi:
Yeah.

Jeff Traylor:
So, there's going to be some skepticism a little bit, and there's going to be, what is my part in it? But if you land the vision, here's what we're doing, here's why it's important, and here's your part in it, that's huge. That's super important. And then, get going, and then, kind of figure out how are you going to measure, like, what does success look like? How are you going to hold yourself accountable to achieve those outcomes?

And then, as I said, start small, get some early wins, get some momentum, because that will help you overcome the inertia that's inevitable. But once you have that vision, you also, it's super important to have, what is your technology estate? Like what is the strategy? What is the architecture? Because it's super important to get that land. And that's again, where you come into work with AWS, work with Databricks, work with partners that have done this, and it helps you build that foundation that's super important.

And once you get that foundation in place, it's like the magic of compounding. You get the foundation in place, you get some early wins, you get some organizational momentum, you get some learnings, you realize the things that you were fearful of aren't nearly as nasty and gnarly as you thought, and the things that you weren't even thinking about can bite you.

But you get some learnings, and then, it starts just really accelerating from there.

Shaown Nandi:
And those learnings are, probably, really, levers that let you accelerate. There's a lot you said there that we can start to unpack. I think the first that I heard that's super exciting is something we love to do here as well, which is identify the problem, and maybe, work backwards from it.

Jeff Traylor:
That's it.

Shaown Nandi:
And get to the right opportunity for the customer, rather than just do tech for tech's sake.

Jeff Traylor:
That's right.

Shaown Nandi:
On the second, you mentioned the change in how you have to grapple with organizational stasis. With generative AI sort of being everywhere, and it feels to some customers like overnight, how are your customers sort of dealing with that change in both opportunity and from a fear factor perspective, how are you helping them with that journey?

Jeff Traylor:
Yeah. Again, we start off with, again, what are they trying to accomplish, right? And then, ensuring that they have that foundational, like do they have their data in a manner that's useful? Has it been deduped to the extent possible? Has it been cleansed? Has it been secured? Do you have readiness to do generative AI? And then, what are you trying to accomplish with it?

We, typically, tell people because the first instinct is, oh my gosh, I'm going to be an industry disruptor, and I want to go to market, and I want to change the way we do our product delivery or customer experience. And that could be awesome. But oftentimes, the best way to use it is to start inside, like help your company run more efficiently. How do you get more effective to execute faster? And then... It's indirectly helping the customer experience, but get good at that. Help your organization be more efficient, more quicker time to market.

Then, once you build that confidence, then, you can start dabbling or being more ambitious in changing how you actually interface with customers. So, that is a couple of things. Number one, it allows you to kind of learn, get some experience, avoid some reputational...

Shaown Nandi:
Risk.

Jeff Traylor:
... damage and risk, and mistakes, maybe, some legal risk, et cetera. So, it kind of helps de-risk it initially.

Because there's a tremendous amount of value.

Shaown Nandi:
Yeah.

Jeff Traylor:
I mean, there's a tremendous amount of organizational value of real dollar costs, opportunity costs, just driving efficiency. You think about so many processes inside of companies are spent doing low or non-value added activity, but there's just the repeated drudgery and just looking for needles in haystacks type of activity. It's not great work for the people, but it's also slow and creates a lot of cost and inefficiencies.

So, if you can take some of that out with generative AI, that's a huge lift for the company.

Shaown Nandi:
So let's talk tech for a minute.

Jeff Traylor:
Yeah.

Shaown Nandi:
I'm super curious, like how is Databricks sort of engaging, moving from being data and classic AI-centric data, generative AI, what sort of products are you bringing to market? How are we helping you? Are we helping you with that?

Jeff Traylor:
Yeah, no, absolutely, you're helping us. One of the things that when... As I said, go back to 18, 24 months ago when LLMs and generative AI kind of entered the world from... It became ubiquitous... Our CEO, Ali Ghodsi, basically, challenged... at the time, we had about a thousand developers, so about a hundred sprint teams.

And he basically just challenged them to reimagine, how we build our platform today, how do we deliver products for customers, what's all the pain, undifferentiated heavy lifting, as you like to say at AWS, that customers have to endure, the expertise that they have to have to run the product, the skills and the difficulty to stand it up.

And then, just managing the efficiency of like... They challenged every team to reimagine what you do, but using generative AI. So, basically, we've infused the entire platform with generative AI from the inside out. And then, now, we're also making those experiences and capabilities available for customers. So examples of that would be a product called Genie that we have.

So, imagine, so just basically allowing anybody in the company to access the data using natural language. So, as an example.

Shaown Nandi:
A game changer for them?

Jeff Traylor:
Yeah, totally. And so, historically, you would have to go through your MIS department or your BI teams, and give me a report, and then, hopefully, within a day or weeks, or whatever the case may be, you'd get your report. But if you could just access it immediately, self-service.

So, an example of this is, at Databricks, we call our employees “bricksters”.

Shaown Nandi:
Love it.

Jeff Traylor:
That's just like a unique language to Databricks, right? And so, let's imagine somebody in HR, our chief human resource officer says, “How many bricksters did we hire in Latam last quarter?” Just a simple example. Latam was a geographical region that we organized, by bricksters and employees. So, that semantics is built in to the platform. We basically use generative AI to build those semantics where they don't exist, and it'll basically yield a report, or the feedback to her that says, here's how many employees were hired in Latam last quarter.

That's a very simple example, but it's that ability to access it, engage with the data using natural language.

Shaown Nandi:
I love it-

Jeff Traylor:
But done in a secure way, giving you the access to the data that you, as a user, as Shaown, has access to... You do that through Unity Catalog.

Shaown Nandi:
So, you have fine-grained access controls effectively?

Jeff Traylor:
Yeah, fine-grained access control through Unity Catalog. So, Unity Catalog, basically, understands and manages the permissions, the governance, the security, and also, as I said, helps build that semantics. Basically, creating almost like a neural network for everything that's in the platform.

Shaown Nandi:
So, the outcome for customers and sort of the outcome you described internally, Databricks, is you've sort of democratized access to the data. Now, more and more users can get to it and work with it.

Jeff Traylor:
That's right.

Shaown Nandi:
Are you seeing customers start to change their mental models around what they can make happen with this data because of generative AI?

Jeff Traylor:
Yeah, absolutely. It really is about exposing... Just simplifying the experience and, truly, bringing more value from the data, getting access to the data much quicker in a more simplified way.

As I said, making use of natural language. The days of needing to know SQL or Python or whatever the case may be to get at the data understanding, looking at the data library and understanding what to access, that's no longer required with the generative AI tools like Genie, as an example. We have an AI BI utility as well.

It's a BI capability that allows you to use natural language, and drag and drop and create dashboards, like big business users are able to do this, where, historically, you had to have departments, skilled people that would do that on your behalf.

Shaown Nandi:
It sounds like a really big change in how companies have the potential to operate, and really, get to their data and have actions come from that data.

What about the people, and specifically, like... I'll just turn to your team for example. How are they keeping up with this rate of change? How are they learning about this? How are you attracting and retaining talent in this sort of amazing time where there's, I think, a new model being released every week, it feels like, and...

Jeff Traylor:
It does seem that way. I think there's a couple of things. Number one, it's... I'm going to tell you anything given your role at AWS, it's very difficult to attract, there's a hunt for talent right now, as you might know. How they do it, number one, they're super bright people, they're super energetic. Obviously, we look for that. But I think the pace of change gives them energy, actually.

I think what seems like an endless conveyor, being on a treadmill of learning what seems to be... Just as soon as you think you've got it, something new drops. But that also kind of creates a bit of energy, and it acts as a catalyst for them to do it.

Obviously, just using generative AI and large language models helps you to get access and get smarter faster as well. So, a lot of our teams use it. A lot of our developers use it for code correction.

Shaown Nandi:
They're ramping quicker-

Jeff Traylor:
Yeah, yeah. They're constantly... It's dramatically changing the developer experience as well.

Shaown Nandi:
That's incredible. We've seen that, too. I'm curious, any words of advice for your customers, our customers, all those enterprises and ISVs and mid-market firms out there, how should they be bringing talent in? What kind of talent should they be looking for to take advantage of generative AI and all the changes that are happening?

Jeff Traylor:
Yeah. I think, it's obviously, to the extent you have classic... IT training and skills and experience is always desired, but I think the days of everyone needing to know how to code and being that as a requirement, I think with prompt engineering and just being able to access them through conversational chat experiences, I think will help simplify that.

But it's the same old, same old, like looking for people that want to learn, that are ambitious, that are comfortable being uncomfortable, or leaning into the technology, because it's just super disruptive, but there's constantly capabilities that are being brought to market that help simplify it.

Certainly, if you can code and all those things, it's wonderful, but I think it's going to become less and less of a barrier to entry for a lot of people, doing what would otherwise be technology jobs, but are made much easier.

Shaown Nandi:
Technology is slowly going everywhere. I've been seeing that for a while.

Shaown Nandi:
So, Jeff, tell me, one thing I've heard from customers is, this jump in fast and start experimenting is great, but they're trying to figure out how to like have production value out of this, how they want to take a use case, and actually, put it out there at scale. How are you seeing customers think about things like return on investment, and what are some of the metrics and measures and how are they doing that?

Jeff Traylor:
Yeah. One of the things that we see... One of the big emerging opportunities that we see in the data and AI space and generative AI as well is personalization. That's kind of somewhat the holy grail for all user experiences and just making that a better user experience, be it internally or externally.

If you think a customer like Skechers, to manufacture... They use Databricks to do personalization. And so, personalization, if you think about it, personalization is really hard, because it requires a ton of data wrangling and accessing data, doing it in a way that is secure, that respects privacy, but, basically, teasing many different data signals from different areas of your technology stack, and accessing, maybe, third party data as an example.

It could be weather, or any other kind of influences that might change the personalization experience, bringing all that together, and that can have huge ROI, huge impact, because, as a buyer or a consumer, it feels very different. Right?

Jeff Traylor:
It just feels magical almost.

And that's huge ROI if you think about that. The customer lifetime retention, customer lifetime value, if you can retain that, and personalization can help you do that, that's huge.

Shaown Nandi:
I am curious, with your customers as they go on this journey, what are things that they're nervous about or worried about, and how are you alleviating those fears?

Jeff Traylor:
Yeah. I think using the generative AI space, or just data broadly, security's always, hundred percent, always going to be a huge thing, just the reputational damage and liability that can come from that, which makes it all the more important that you have a complete understanding of your data and where it resides, deduping it, making sure it's secure, fine-grained access control lineage.

So, if you think about an example... And again, using generative AI as an example, so you can imagine... So, on Databricks, let's assume you're using your data that are pre-trained models, and you've got data and you've got tables that are not meant to have PII data in them.

So, if you talk from one end of the stack all the way from your ETL processes and landing the data, and you're not meant to have... This is data that's not meant to, or it should mask PII data, as an example, but a mistake happens. Somebody makes a mistake. It starts injecting PII data. It's, now, going downstream, it's being used to train models, and now you've trained a model using PII data. That could be huge reputational impact and liability impact.

Well, with things like Databricks and using Unity Catalog, we have that visibility across the entire stack, across all what would, otherwise, historically, be disaggregated data in different silos. It's all unified. And so, basically, we have technologies within the application that will identify that quickly, expose it to you, and say, “Shaown, you've got PII data in this table that you weren't meant to have, and here's when it started.”

Basically, you can create a lineage map and it will bring you up, share the map when it got introduced, where it got introduced, and then, upstream, since that happened, where it's been used. Oh, no, we started training this model last night using this data. Stop. And you can roll back the change, and quickly-

Shaown Nandi:
I love it.

Jeff Traylor:
Quickly avert what otherwise would be a huge, potentially, a huge impact.

Shaown Nandi:
So, the tooling and the mental model all together...

Jeff Traylor:
That's right.

Shaown Nandi:
Jeff, thank you so much for being with us today, and giving us some vision and inspiration from how you operate at Databricks, and how Databricks helps customers with AWS. It's really great to hear.

Jeff Traylor:
Well, thank you for having me.  

Listen now

Listen now

Listen now