Agentic AI has shown potential to detangle fields of thorny, unstructured data brambles. Marketers for large enterprises sometimes face this issue as they sift through vast, unsorted datasets to inform localized campaigns, websites and sharply cut market segments.
Lenovo's sales span 170-plus countries; its dense technology catalog can be challenging for even the most astute buyer. Its marketers need to sort through data to find insights for digital campaigns. UX designers on the e-commerce side need to understand how to narrow catalog searches so results aren't overwhelming -- and are well-targeted -- for consumers and small businesses.
That all keeps Lenovo's data analysts busy, as they aggregate data and tease out the insights marketing and UX teams need for their various regions.
Derek Gominger, Lenovo's global e-commerce chief operating officer, discussed how his company deploys AI -- zooming in on how it has deployed Adobe Data Insights Agent -- to drive more precise customer experiences, and to put more data in the hands of marketers, so they can explore new and varied campaign ideas without analyst assistance.
Editor's note: This Q&A was edited for clarity and brevity.
What problem was this agent meant to solve?
Derek Gominger
Derek Gominger: Our data had been really disaggregated. There's lots of different data. [Our operations are] in 35 countries and China -- which is a separate operation -- trying to sew together information from marketing campaigns across all those different regions. Plus, all the data that we have internally. Then we track on the website.
Adobe is core to that, but it's a lot of data -- and it's a lot of different campaigns trying to understand at a macro level, "How do we simplify the journey for the customers?" to make it easier for them to transact and find Lenovo products.
So, you had this problem of fragmented data, spread across many systems that you wanted to run analytics on and make it closer to the front-line, so people can ask questions about the data to design experiences.
There are probably 10 ways to attack this problem -- such as a data lake. What made you decide that an agent would work?
Gominger: We tried some of those 10 other ways to answer those questions. You take the elements of structured data, unstructured data, trying to understand what content is actually working in the media and what's driving the right traffic -- and then marry that data with the actual traffic data of the website, the conversion rate, the click-through rate, the attach rates and all of the other data that comes with running a website.
That's a ton of data. It takes a long time to analyze all that data.
Even in a data lake, you're not going to get the unstructured data. So, agentic AI came in able to add the unstructured data to the analysis and expedited it.
Is this agent for data analysts, for marketers at the regional and local level, or both?
Gominger: Both parties are really using it -- and then I'll throw a third party in there: Our journey UX, customer-experience focused team is also using this data to improve our customer experience.
The data varies, but around 70% to 80% of all SMBs start online regardless of where they buy, right? So, having a strong digital presence where they can find information about our products is essential. We want to make sure that when customers come to Lenovo.com, they can find the right product. I can't stress it enough -- little insights scale tremendously. A couple of basis points of efficiency in our media spend scales tremendously to the business and the P&L.
Moving on to agentic AI in general: Which came first for Lenovo, using agentic AI for marketing or e-commerce?
Gominger: I was organizing the e-commerce site around making a frictionless path after identifying who the user is. With such a comprehensive portfolio, it can get confusing. How do we help you find what you want fast and easy?
Agentic adoption is growing fast. There will be an inflection point where it's going to become the norm, where traditional Google search goes away, and it's going to be ChatGPT, Claude, Gemini search. A conversation with an agent is going to become much more normal. At Lenovo.com, consumers and SMB buyers will want an agentic experience. They're going to want a concierge.
People don't really want to talk to someone unless they have to. In an agentic world, all of that can happen online now. All the technology is out there to pull out an experience that is concierge-selling in a few questions.
It's a lot like a wine sommelier. You have a really big wine list. You don't have to know all the bottles on the list. A sommelier is going to come over and ask four questions: "Red or white?" "What are you eating?" "What have you liked before?" "How much are you looking to spend?" and then recommend a bottle. That's it; they can give a good-better-best scenario, and you choose. You don't have to know the entire list. Agentic AI enables this experience to assist customers online.
How do you strike the balance between thinking of agentic AI as a powerful general technology that can do everything -- versus tactical applications in niche workflows?
Gominger: Honestly, that has been one of the hardest things we've been dealing with. There's a lot of excitement around AI, and so there are a lot of teams that are going out and trying to solve, you know, one use case, and then there's another team trying to solve another use case, and there's another team trying to solve another use case. It creates many boats on the water. I don't know if they're all going in the right direction -- or the same direction.
How do you get the boats or marketers or UX designers to move in the same direction -- and get consistent results -- when generative AI generates a different answer every time?
Gominger: I'm a fan of guardrails and guidelines to make sure that the probabilistic world gets developed within the right framework. The data has to be ready. We all need to be working off the same data. Then there's a governance element to the AI development. There has to be some governance that says, "This is the way we're going to go do knowledge management," and "These are the use cases we're prioritizing."
That's less of governance and more like orchestration of activities within an organization, the business management of the agentic solutions.
So, after all that thought and investment, how is an enterprise supposed to measure the real-world benefit of an AI agent, using what KPIs?
Gominger: KPIs drive the outcomes. They're certainly an element of the decision matrix. I think there's also an important strategic component. Do you really want to build and own [this on-device personal assistant Lenovo recently released named] Qira? We want to build and own an on-device agent, because we own the device. We want to own the agent. We want to ensure the data is secure and develop that core competency.
From a KPI perspective, that's more of a strategic direction. There will be KPIs to follow as we maintain and develop Qira. It was a big investment.
Don Fluckinger is a senior news writer for Informa TechTarget. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.
