UPS is among the largest courier and logistics companies in the world, delivering more than 22 million packages on an average day. The Atlanta company traces its roots back to a delivery business founded more than a century ago. But amid a competitive landscape, UPS has come to experiment with new business models like dynamic pricing and technologies like cloud computing and, more recently, generative AI. UPS’ 2023 revenue was $91 billion.
We spoke to Sunzay Passari, UPS’s Senior Director for Digital Innovation, about how his team carefully studies, tests, and deploys innovative technologies without slowing down the deliveries its customers — and their customers — depend upon. This interview was part our latest research initiative, The Future of the Innovation Team.
• • •
Can you tell us a bit about what you’re working on at the moment?
My current year’s focus is around GenAI capabilities, big data, and advanced analytics.
The second biggest piece is also about a whole bunch of digital transformations, and these two are not independent of each other. I’m doing digital transformations of a lot of business support systems — rating, charging, billing capabilities, customer insights, customer data, and security.
We are…116 years old. You can imagine the kind of legacy we have in terms of technological systems: applications as old as 40 years old still running. We are moving things to the cloud, but still a few things are on the mainframe as well. We’re working on a transformation where we get rid of our old technologies and old technical debts and get into the latest and greatest technology.
What do those transformation processes typically look like at a high level?
Let me start with where my role sits within our operations. Different organizations have different ways to look at this, but my organization is basically under our chief commercial and strategy officer, which is a pretty big role.
And in our day-to-day interaction, it is really important to understand the pain points we can address. There are those growth-focused areas where we do things proactively, and there are certain pain points where we are doing things in reaction to certain situations. Both ways we can drive innovation, and we very closely work with stakeholders to understand requirements.
My guidance to my team is the deeper you discover—the granularity point—the better and faster you design.
What’s a good example of that?
This morning I was talking with my team about the capability for doing a performance incentive for our customers. Think of a performance incentive at a very simple level as, “You achieve X volume, you get Y discount.”
That’s a simple way of defining a performance incentive, but in the logistics industry, as in other industries, the scenario could become infinitely complex. I might try to incentivize a route which I want to promote, or a lane where my perishable capacities are higher than average. If I start adding layers of complexities, it could get infinitely complex.
So we have to draw a line — this is a good starting point, an MVP — and then start getting into level 1, level 2, etc. And beyond some point, the law of diminishing returns starts to apply, because it could become too complex for the customer to understand the value proposition, or it could lead to a lot of accounting or reconciliation issues. So we start drawing a line and say, “OK, that’s the best-case scenario.”
And then as far as generative AI, where are you seeing that potential and how have you been testing it out?
We are being a little conservative.
One reason is that most of our business is basically B2B — our B2C interaction is limited. People like you and me are not often on UPS buying a label — it’s more that Amazon or Shopify are buying a label to deliver to your home.
One of the use cases which I would potentially like to try building, since we are a B2B company, is some kind of generative AI capability for my leadership to gather customer insights. Where did we grow and lose business and why? Getting from information to an insight: The information is that we lost 10 percent of the business from Customer A; the insight is what led to that loss. So I’m trying to build some use cases and capabilities around that as we speak today.
But we are moving conservatively, because we want to observe how things are emerging from a technology perspective, and being a very large company and holding a humongous amount of personally identifiable information, we want to be very respectful and compliant with how we use our data.
…It’s not about going into ChatGPT and asking a generic question — it’s about trying to get a very specific insight, which means I have to train the model and infuse it with my proprietary information.
And are you testing and evaluating different tech vendors, or what does that process look like?
It starts with, do we build or do we leverage an existing LLM? Versus are we going to use a RAG architecture, or build our own internal language model?
And then everything boils down to data strategy. Everything will run off the strength of the underlying data, because it’s not about going into ChatGPT and asking a generic question — it’s about trying to get a very specific insight, which means I have to train the model and infuse it with my proprietary information.
And on top of that, thinking about the use cases and interfaces.
Q: You’ve written and spoken a bit about implementing dynamic pricing. What does that look like — was that a big shift in the organization?
It’s a pretty important initiative for us. Dynamic pricing as a concept is not new. Hotels and airlines, that’s exactly what they’ve been doing for decades. But in an industry like logistics, it takes a different approach.
In the first place, the pricing has to be a little more surgical: For every transaction, I’d like pricing where my customer wins, and none of us lose. That’s easier said than done.
And for a larger customer shipping millions of packages, they will say, while I appreciate the intent of the surgical pricing, I want predictability into my operational expenses. So we have to walk a very tight rope between predictability and dynamic price, with some kind of fixed pricing and incentives built around that. And then for a smaller customer, if I’m able to save him $1 a shipment and save him $10 a month, he’ll be happy.
So we are trying to address all these complexities in the modern world of artificial intelligence and machine learning. Rather than a fixed rate chart with an offer and discount, make it a little more dynamic, but not as dynamic as an airline.
How have you rolled it out so far?
There are certain segments we have done, and we have been very successful in them. Initially we started calling it an experiment, but they are more than experiments today. Our customers are happy, and we are happy because we see a lot of growth. In certain segments, we were able to win business.
How do you typically run these types of experiments?
A lot of these are done on a short-term basis to test out a hypothesis. As a very high level example, say, whether 10 percent off makes any sense or not. So we create a hypothesis, we run a program around that for the short-term, and we study the results of that. If it works, great. If it doesn’t, think of something else.
The first [metric] is volume: is it helping us win more packages? The second thing immediately becomes, is all that volume profitable?
What kinds of results are you looking at?
Mostly quantitative. Because most of our customers are B2B customers, that customer feedback is extremely important to us, but it’s not as relevant as it would be in a B2C scenario. So it boils down to tracking quantitative metrics rather than qualitative metrics.
The first one is volume: is it helping us win more packages? The second thing immediately becomes, is all that volume profitable? And then some of these hypotheses are based on certain objectives. So volume is really important, because that tests whether something is swinging the needle for us, but once we that needle is swinging, we have to figure out the right swing.
What’s the process like when you’re working on deploying new systems?
We have a complete infrastructure team, so I play more of the product and strategy role, but the actual heavy lifting is done by them. But we work together on the plan, strategy, and all of that.
Things can take [anywhere] from a few months to a few years. We are moving millions of packages a day worldwide, so there’s a lot of moving parts and a lot of compliance. We have to be extremely conservative. You cannot afford for a single package to get lost.
When we move to production systems, we do it in bits and pieces. In the simplest example, say I had a new pricing mechanism. I would not move everyone to that new system — I would move some smaller customers, where I can manually fall back or if something goes wrong, I can rectify something really quick.
And how is your team organized?
There are really three large blocks to it. One is your product management practice. I have a whole bunch of innovation managers. And they can have multiple sets of product managers reporting into them.
The second is product operations. We can build a certain technical capability, but the stakeholders don’t know how to use that capability to run an experiment. So my product operations teams help operationalize the capabilities in the backend.
And then I have a smaller team working more on go-to-market, and user interfaces — physical touch points for the users and trainings and go-to-market and all of that. Some of that work creates content for the salesforce to help them launch in the market.
Are there particular skillsets you find particularly valuable?
People with great common sense are the ones who can lead.
And then I would say, one pillar is technology, one pillar is understanding the business, and the third pillar is creativity. I typically look for someone with a good amount of skill in each of these pillars, and then of course management comes into it, because you’ve got to deal with stakeholders.
[We have] internal stakeholders and external stakeholders, because we’re not going to build everything ourselves. We evaluate a lot of vendors. We do a lot of proofs-of-concept. So we have to be really adept at managing people and processes in organizations. Half of my calendar is meeting stakeholders, whether internal or external.
