Interview with a former Pepsi brand VP: How CPG marketers can build an AI advantage competitors can’t buy

Angelique Bellmer Krembs spent 23 years in marketing at PepsiCo, including as a VP on the Pepsi brand during the shift to digital. Today she advises CPG leaders on AI through A.Team's CxO network, and when one of them tells her their company is "doing AI," her first question is how they're defining it. Often the answer is that they've brought in some AI tools and everyone has a ChatGPT account.

We sat down with her to talk about solving the gap between adopting AI and changing how the work gets done. In this interview, Angelique Bellmer Krembs shares where marketing leaders should start with AI, why using it only for efficiency is settling, and how to build an advantage competitors can’t buy.

Key takeaways

• Leader churn and fear keep many CPG companies stuck in discussions about AI. A written blueprint for the intelligence layer keeps progress from resetting when a leader leaves.
• Start with a vertical pilot on one small use case, redesign it top to bottom, and prove it in 90 days. Uses where AI pays for itself, such as media planning, make an easier first step.
• Using AI to do the same tasks faster with fewer people is settling. The value comes from rethinking why the work is done and redesigning the workflow.
• Counting how many employees use a licensed AI tool is a vanity measure. Competitive advantage comes from redesigned workflows and proprietary data.
• On build versus buy: buy the commodity, build the advantage, and work with a partner who can do both.
• If AI takes over entry-level judgment work, companies need to build feedback loops so junior talent still develops judgment.

Why companies get stuck

What is the biggest challenge you see when enterprises get started with AI?

I see a lot of companies through my PepsiCo alumni network, and I have a fascinating seat at the table through my advisory work with A.Team. I'm definitely seeing a lot of companies stuck in paralysis instead of actually building with AI.

Churn is one of the things happening. You'll see progress being made, and then the leader leaves and they have to start over. You can make three to six months of progress with discussions, and then they're gone and you're starting over again. I've seen it a dozen times. It's a real factor to think about: how do you keep the momentum going? How do you hand over the blueprint so they're not starting from zero?

There's also a ton of fear in big CPG companies: not wanting to put their data into an AI, not trusting their employees with using AI, to different degrees. I've talked to some CPG companies that are like, "Oh yeah, we're totally into it," and others that are like, "Oh no, we're not allowed to do anything." There's fear of investing with a partner: have we chosen the right ones?

And there's leaning toward legacy partners. It feels comfortable to go with the big players, the Microsoft, the Accenture. No one gets fired for hiring McKinsey, I guess, is the phrase. They're already embedded in the company. They have huge MSAs. But what they're missing is the agility and the innovation that comes from a startup, and I've heard this from the enterprises themselves: if only they could have the agility and the innovation of a startup. A partner like A.Team, which is a startup AI solutions company with the backing of Microsoft through the Pegasus Program, is the best of both worlds.

What are the biggest fears you hear from marketing leaders, and how do you get past them?

Some of the biggest fears I've heard from other marketing leaders about AI are, honestly, about getting started: getting the alignment to get a pilot going. There are so many stakeholders who need to get on board. There's so much fear of stepping in, of maybe going too far too fast. That fear is probably the biggest barrier to getting going. There's nothing more frustrating to a marketing leader than seeing the potential of this technology and not being able to do much with it, when you feel like your hands are tied because the vetting process is so long or because there's such a risk intolerance.

Some of the successes I've seen through A.Team come from charting out a pilot that is vertical, not horizontal, because trying to build something that goes across is going to take way too long to prove out. Start with something vertical, where you could say, "Hey, just for this tiny little use case, let's go top to bottom, redesign it, and prove it in 90 days." Start a pilot that doesn't try to boil the ocean. I've heard A.Team's CEO, Raphael Ouzan, say that more than once, because he's seen it play out. You bite off something where you can see the impact, and then it's so much easier to scale from there. Get a pilot going fast, prove to the stakeholders that it will pay out and that it will be safe, and then you can get on to bigger things.

If you were back in a brand lead role at PepsiCo, where would you start?

I would be looking at ways for the AI to pay for itself quickly. One of the things I've seen be most successful is using AI in media planning and media marketing. AI can identify unused dollars and wasted dollars, and be more efficient with the dollars, so you immediately free up money that pays for the AI. Then you can do other great things, like connecting to social and automating the creative. Looking for ways for the AI to pay for itself quickly feels like an easier place to start. It'll make the CFO happy.

Efficiency vs. transformation

What changed in your thinking as AI came onto the scene?

The first epiphany I had was, "Oh my gosh, you can talk to your data like a human." I was running around saying it. But then I learned that it's actually your data talks to you. So the potential is just tremendous.

The trap, though, is to think of AI as a tool, because then you think of it as a product, as a technology. The technology is super important, and the dev teams are super important, but it can't be technology-led. If you're thinking of AI as a tool and not a workflow reinvention, you're missing out on most of the potential. You have to have the subject matter experts at the core. The business partnering with technology is really the thing that's going to create true transformation. It's not just AI adoption. It's AI transformation that you're going for.

What is the most expensive wrong assumption marketing leaders are making about AI right now?

It's got to be around tools versus workflow: thinking they would just replace current processes with AI versions of the same thing and string them together in the same way. You will get some efficiency out of doing that, but you're really missing the point on the transformation that's possible: rethinking the workflows and redesigning them. They didn't even realize that's not how AI works. This is not a product you're going to be getting. This is a layer of intelligence, and there's iteration you have to do to get the AI to create the workflow that replaces what you're doing.

What does it look like when a company settles for "good enough" AI?

When a company settles for good-enough AI, it's all about the efficiency. It's like, "Oh, I was doing all these things, and now I can do them faster with fewer people." Efficiency, efficiency, which sounds good for business, but it's really missing out. It really is settling, because if you're not transforming the workflows, if you're not rethinking what we're doing differently now that we have this AI capability, then you're just scratching the surface.

It's not about efficiency. It's about transformation. If you just automated everything you're doing now, you're scratching the surface. But if you actually rethink why we're even doing this, or, now that we can do it this way, what changes in the workflow, that's where you get real transformation.

When a marketing leader says they're "doing AI," what do you think?

The first thing I'm thinking is, how are you defining that? "We've automated, we've brought in some AI tools, we're using ChatGPT, everyone has an account." That's often the case, and that's barely doing AI. That's training wheels, I guess. You have to start somewhere.

Adoption alone is not really a great measure. What they're trying to get at is how many people are using AI in the company. It feels like a vanity measure, because it's really about: is AI transforming anything you're doing? Are you really rethinking the workflows now that you're using AI? Checking the box on using AI feels like a waste of money.

We've seen enterprise companies just getting an enterprise license for an off-the-shelf tool and calling it a day, thinking they're doing AI: "Look at the adoption of our employees using our GPT." Gosh, that's not the point of AI at all. I guess it'll make a tiny dent in efficiency. People will write shorter emails, maybe. But the power of AI is in compounding intelligence. The only way you get any competitive advantage out of AI is if you actually redesign your workflows and use your proprietary data to build something unique. Otherwise everybody is going to have that advantage, so there's no advantage. How do you build a moat out of AI? There's got to be something proprietary about it.

I've also seen companies go to the other extreme, where they've allowed all their employees to use AI, and then they've got a handful of power users who are just blowing up the tokens because they're vibe coding on things they really shouldn't be doing. They're not the most efficient person to do it, but it's possible to do it.

Building an advantage

What does a good AI strategy sound like?

There are a lot of surface-level answers you can get to "doing AI." The more transformative things I've heard are from people who are blueprinting: "Here's how we're going to build our intelligence layer, and here's how it's going to compound over time." It's not the usual answer I get, but I've actually seen it in a couple of cases.

That has several benefits. Sketching out the blueprint makes it easy to work with a partner like A.Team, who knows how to make that happen and can make it better. It also survives any churn. If that leader walks out the door, the blueprint is still there. It's really great evidence of doing AI in a way that will actually bear fruit.

How should enterprises think about the question of build vs. buy?

It's easier to buy, but everyone wants to own, especially when you're talking about proprietary data, and people get worried about that. Somehow the image that comes to my mind is a toddler: "I got it. I got it." You want to be able to do it on your own.

I'm seeing companies working with AI startups who initially do the pilot, see the potential, and then bring it in-house, thinking, "Okay, we got it from here. We're going to build it." And then they come back, because it's harder than you think to keep the most cutting-edge AI builder talent in an enterprise company. Plus, they've got all these other technology priorities, so it's really hard to maintain that level of innovation and forward thinking in-house.

In the build-versus-buy debate, it's really: buy the commodity, like any off-the-shelf tool that can help you get started, whatever is commoditized, and be as efficient as possible. Then build the advantage part: what is unique, what is ownable to you. And ideally, work with a partner that allows you to do both.

What are the top marketing AI use cases you've seen?

There's the obvious stuff, which is removing the tedious tasks. I mean, the deck writing. Sadly, there are enterprise companies out there that aren't even using AI for help with that because of the fear, but it feels like a no-brainer. It's best used when it's embedded in the tools you're using, like PowerPoint. It makes the process more focused on the higher-level thinking (what am I saying, what are we doing) as opposed to the tedious task of creating decks.

Then there's using AI for all the update scorecards and reports that are constantly being asked of marketers, using AI to serve up insights, to serve up what happened. First of all, it's way more timely, and it saves all that scrambling for information. Everyone then has a single source of truth, which is transformational at a company. You also can't hide, but it's probably best in the long run if everyone is looking at the same source of truth.

Across the creative process, it's transformational, because it can hold you accountable to being true to the brief and true to the brand guardrails better than ever before. It's reading in social signals faster than ever before, absorbing insights and translating them, creating more options to work with, narrowing down options faster. In that whole creative process, AI can be so powerful.

It's also really great to be able to look back over past work and see what led to growth, or to look across a company: across brands, across markets. And it's not just marketing, because marketing can't solve anything by itself. It's always connected to sales. How do you cross functions with that compounding intelligence? There's so much potential to bring people together with better information than they've ever had before and faster, and to elevate the humans to really be thinking about what we do with this.

Failure, knowledge and judgment

Can AI improve your team’s judgment?

Whenever I think about marketing failures, I feel like it's always a lack of intelligence or a lack of insights. I don't mean lack of intelligence like people are stupid. I mean they didn't have information they could have had. And not just the information I might have as a brand manager, but from the other functions that might feed into a project I'm working on.

I don't think failure is a bad thing. We always say failure is what makes you realize you're taking risks, and some risk is necessary to try new things. It's maybe a question of managing the failures. How are you processing them?

One of the things I remember from the New Ways of Working we did at PepsiCo, learning from more agile startups, was the debrief process: looking back over what just happened. It was so hard to make time for that. Now, with AI, it should be so much easier, not only because a good AI intelligence layer should be able to look back over all the work and serve up cautionary moments to consider, but also because it can do that quickly. The excuse of not having time to do the debrief should be way less.

What happens to an organization's institutional knowledge with AI?

It's really fascinating to see how institutional knowledge is going to be nurtured over time with AI. There's a compounding of intelligence you get with the humans who are there, but it's all very distributed, and when the human walks out the door, so does their knowledge. One of the potentials of AI is to be that intelligence glue that keeps the institutional knowledge of the company and never walks out the door.

The potential of compounding intelligence is one of the most exciting parts of AI, because it's not just taking one project and making it faster. It can do that at the scratch-the-surface level, but the idea that it can look over multiple projects, over everything you've ever done, looking back over history for patterns you couldn't recognize yourself, and start to predict the future — this compounding intelligence is so powerful. We couldn't do that easily back in the day. You had to know who to talk to, or you had to be lucky enough to have stumbled on the thing that was tried before. Now AI can know it all.

What does AI taking on more tasks mean for junior talent?

The flip side is, how do you get young talent to gain the experience they need to have good judgment, especially if you're outsourcing a lot of the judgment work that is usually entry-level work to AI? That's a real risk. One of the things I've been looking for is how people are using feedback loops. How are they creating processes where the AI is working, but there's still the opportunity for junior talent to be checking for judgment, or validating their own judgment? It has to be built in now. That's a very different way of thinking. Now the trick is, how do we get the next generation to learn alongside that powerful compounding intelligence?

How do we equally invest in the human skills as we're investing in AI? I want those conversations to come up in parallel. If there's a huge push to invest in an AI pilot, where's the parallel investment in how we're creating learning loops for our younger employees, or in the human skills that AI can't teach or do? How do you get a young leader to grow in their confidence and get coaching to be the leader of tomorrow? How do we make sure their judgment is being supported? Our good friend Eric Solomon teaches a class at Fordham on judgment and really investing in those human skills. My Band of Sisters is creating field guides for building the Human Advantage we need in the age of AI. This topic is something I would like to see talked about more in parallel with AI as a technology.

Leading the change

You were a VP on the Pepsi brand during the digital transformation. What parallels do you see with AI?

Everything exploded, everything was different, and there are a couple of things I remember very clearly.

There were people quickly on the bus, and then people really dragging their feet. In an organization, that's very disruptive, because not everyone is equally eager to do it all differently, take on new workflows, trust the AI. It's very hard culturally to get everyone in that same mindset.

I remember there being dictates like, "You will spend 20% of your marketing budget on digital." I'm sure the parallels to AI are there today. But then you don't really know what you're doing, so you're now using 20% of your money in a way you're not sure is going to work, and you're feeling your way around.

There was a handful of people who were definitely at the forefront of this change, and they would look for conspirators. We called them change conspirators: people who were courageous enough to try some percentage of their budget on new things.

One of the things Pepsi did at the time, which we called New Ways of Working, was to actually talk to startup companies: "What are you doing that we could learn from? What would make us more agile?" One thing that really stuck with me was risk thresholds: having levels of risk that every employee was enabled to take, up to a certain point they were comfortable with. It just allowed people to be more free to try things. Then at some point, obviously, you have to get the whole enterprise on board.

I feel like that could apply today. There should be some level of risk threshold where AI experimentation should happen, and then maybe a very regular process of elevating the things that are working so they can get further funded.

What have you changed your mind about in the age of AI?

First of all, I should say I do not have the answers. Honestly, anyone who says they do right now is kidding themselves, because so much is changing. I definitely remember, back in the digital transformation when I was leading Pepsi, that there was just so much we didn't know, and we made tons of mistakes. Making mistakes is part of it. Obviously, you want to make it safe to try. That was a big phrase from our New Ways of Working: how do you make something safe to try? You don't have to get consensus. You have to get consent to try. In the age of AI, how do you make it safe to try? How do you get consent, not consensus, to take a step?

Something I try to remind myself I just don't know is the role of AI in creativity. A couple of months ago, there were a lot of things where you'd say, "Well, AI will never be able to do X well. The humans will always be able to do Y." I just don't think you should have that definitive a mindset. I've seen examples of creative teams using AI to astounding effect. Not always, not consistently, but the potential is there.

We've got to keep an open mind about what's possible and try to create that world, really thinking about what's the best, highest use of the humans. If we have this powerful technology, is there a best, highest use of human potential, so that the humans feel really fulfilled? They get to work on the stuff they always wanted to focus on, and they get to do truly net-new thinking. AI is a technology that allows them to do more work, better work, broader thinking work. It's not taking their jobs. A good leader today, someone who's managing a team and thinking about the impact of AI, should be thinking about what best, highest use their humans are going to be doing. Then hopefully it's a win-win.

What changes in the CMO's role with AI?

CMOs have always been in the role of innovation and future-proofing. Obviously, they're stewards of the brand, so they have to be thinking about how to protect that equity and leverage it for growth for the business. With AI, that role includes how AI is going to help us grow and help us protect our equity. It's a continuation of that innovation mantle. Not just a chief marketing officer, but a chief magic officer. I like to call myself a chief magic officer.

The magic is in that spark of inspiration, that culture of trying things, trying hard things, doing hard things. It's the magic that comes from finding the thing that's never been done before, the hard thing, which is harnessing AI to really make a new path with competitive advantage, with efficiency. I really think every CMO out there is tasked with leading that charge.

I wouldn't be surprised if, at a lot of companies, other functions are looking to the marketing leader for how to approach AI. The marketer as a business lead can really be the SME who is thinking about the use cases, thinking about the business case that makes the technology make sense. If you can do that as the marketing lead, in partnership with technology and in partnership with the CFO, I think that creates a lot of job security.

Who should marketing leaders be listening to?

You should be talking to startups. You should be going to conferences. You should be reading what people are writing about what's going on, and have your advisory board, your personal set of who you're listening to. For me, I learn a ton from having a seat at the table with A.Team and seeing what they're learning and what challenges they're trying to solve.

See how A.Team AI Solutions works →

A.Team is building a new model for how the world’s largest consumer brands staff and scale AI. Your teams get agentic teammates that live inside the tools they already use and take on the work that eats their time. Behind every teammate is your agentic system, built by A.Team's forward-deployed engineers inside your own environment. This intelligence layer connects your company's knowledge, records every decision, and learns from every outcome – getting sharper each cycle, for a compounding advantage you own outright. Trusted by Fortune 500 and high-growth companies, A.Team is backed by $60 million from Insight Partners and Tiger Global, recognized by the OpenAI and Claude Partner Network, and a member of Microsoft's invite-only Pegasus Program.