How to Build an AI Agent Strategy: Lessons from Colgate and Mondelez

Artificial intelligence has caused a seismic shift in the consumer goods sector, with AI agents promising to further enhance customer experiences and save time by automating processes for greater operational efficiency and strategic decision-making. By 2027, 50% of business decisions will be augmented or automated by AI agents for decision-making, according to Gartner.The AI switch was turned on early at Colgate-Palmolive, when, in 2023, officials decided to build a platform to let employees sink their teeth into creating Gen AI agents, or as they are known internally, assistants.“This was a risk-based judgment,’’ says Kli Pappas, senior director of global predictive analytics and global head of AI, Colgate-Palmolive. “People were going to use the tools anyway.” The platform, known as AI Hub, is a set of instructions and background documents, and it’s open to all employees. “Literally thousands of these assistants have been built’’ around the world by employees who have identified use cases, Pappas says, adding that if people want to share them with colleagues, there are some governance steps they have to follow.What’s in a name?AI copilots and agents are sometimes used interchangeably, but they are not the same thing, says Oliver Wright, senior managing director of global consumer industries at Accenture. All AI copilots are agents that use agentic approaches to do activities on a person’s behalf — but not all agents are copilots, Wright says. “The copilot is the interface that a human is going to use to be able to access AI … to execute tasks.’’ AI copilots are very different from any other technology in the sense that they will change how every employee works, including up to the C-suite, Wright stresses. Currently, about one-third of employees are gravitating toward the technology for “experimentation, adoption and advocacy for it,’’ he says.Learn the steps for implementing AI agents and copilots to enhance internal processes across the enterprise.1. Understand what is possibleCompanies need to figure out how agents can improve work and make a decision based on that; a concept Wright calls “right-to-left thinking.’’ “We need people to understand the switch that’s being made … in the way in which they currently do work and then say, ‘How could we use agents to improve different parts of the existing process,’” he explains. That is left-to-right thinking — the idea that you will get some incremental value from deploying agents but you are “fundamentally anchoring in the way in which you operate today.”It’s also important to test out the technology and show how early adopters are using it day to day, Wright says. “The idea of leading from the top is very different and very important.’’The organization should be very focused on prioritizing use cases, agrees Forrester vice president and principal analyst Craig Le Clair. “There’s never been a more important focus on use cases than in the agent world.” He suggests forming an AI working group to define use cases and focus on data. Le Clair also recommends starting with internal processes rather than customer-facing agents, saying “there is less risk and, often, you’re erasing inefficiencies and getting things better.”Once an organization is comfortable using them, he suggests looking at deploying them in more strategic areas. “Doing demand forecasting is going to get harder and harder given the unstable environment we’re in,’’ Le Clair notes. AI agents can make significant gains for companies in demand forecasting, optimizing inventory management and supply chain.Since April, Mondelez International has been using AI platform Arthur Vibe to introduce AI-powered assistants, which have unlocked entirely new ways of working, says Matt Tomlinson, head of extended reality and metaverse. In workshops, the agents act as intelligent collaborators to boost productivity by organizing ideas in real time, surfacing insights instantly and helping to align perspectives, Tomlinson says. Additionally, they can act as “customized trainers” by significantly reducing the learning curve for advanced tools.2. Get your data ready and select a platformNo matter what you create, AI tools are only as good as the information you feed them. This means organizing and ensuring you have the right data that AI can learn from. CGs must look at data complexity and risk, as well as any regulatory and privacy concerns, Le Clair says.“This is about exposing data in the right, safe way to models, and that requires governance and data engineering and things we haven’t thought about in the same way,’’ he notes. Some CPGs are setting up an architectural engineering function “because they are worried about so many agents sneaking in that it will be chaos.”This team is typically charged with selecting the development environment — whether to use application-embedded copilots or custom agentic platforms, says Le Clair. This can be a dicey proposition.“There’s a lot of confusion around platforms,’’ he notes, adding that there are about 300 custom agentic agent platform startups, “and 30 will survive, like every explosion in tech markets, so there are risks there.”Considerations include whether the app-embedded copilots will get companies trapped in ecosystems, Le Clair observes. Other concerns are whether there will be “endless costs without any material value, and which [platform] is right for a particular use case. So it’s the normal kind of confusion but accelerated by the speed of this market and sustaining excellence.”3. Create the new ‘performance frontier’Once use cases and platforms are determined, don’t jump in and push the adoption of agents to the nth degree, Wright advises. Instead, create “the company’s new performance frontier,’’ he says. “That gives us, in effect, the ability to decide how good you should be at each individual process.” There should be a discussion around the sequence and priority for the agents you’re going to apply across the company.“Everybody needs to understand what you can do in every area,’’ he says. Then, based on your strategy and prioritization, “you describe how much you’re going to aggressively move into being agentically oriented or agentically powered in different parts of the value chain.” It’s also critical to get employees on board. Word of mouth tends to be a powerful asset for getting people to adapt to the AI era. Pappas calls the process of creating agents at Colgate-Palmolive “extremely easy.” Anyone interested simply goes to the AI Hub, which shows all the assistants that have been built around the company. They click a button that says, “Create your own” and there is a chatbot to help write the instructions for your system. Employees can also click a deploy button to upload any documents they want to include in their own assistant. If someone wants to share the assistant they have created, they need to write up an evaluation, answer 25 questions and share the assistant with at least 20 people who confirm it works. Once they feel good about deploying it, they can pick the groups they want to share it with, Pappas says. For example, if someone wants the marketing group in Germany to be able to use their assistant, the appropriate person will get a notification email and it gets deployed. “It’s all done, automated on the platform, end to end,’’ he says. “We spent a lot of time to make it as easy as possible.” 4. Start your pilotsBut if you’re not creating an agent development hub like Colgate-Palmolive did, once there is consensus on what people want from their agents, it’s time to deploy pilots. “You want people to use [the agent] in the way that you intend,’’ Wright says. For example, if you want the agent to provide recommendations, IT needs to tell employees, “I want you to use the recommendations that come out of the tool, but the first cut of those recommendations are obviously not going to be nearly as sophisticated as they could be,” he says. Snow Cosmetics started its use of AI agents to boost sales while increasing customer and employee satisfaction. Too much employee time was being spent on low-level tasks while trying to keep up with customer support.Eventually, officials realized they could automate simple inquiries like “where’s my order,” and “what’s your return policy,” and deployed plug-and-play system Rep AI, which can also be customized. About 85% of Snow’s site visitors want to learn about teeth whitening solutions, so the company developed a tooth color chart. When a shopper indicates they’re looking for information on whitening, the chart pops up and the shopper picks the color closest to their teeth. This enabled Snow to guide shoppers to the most relevant products, which has resulted in an up to 15% conversion rate, according to Thomas McCray, director of customer success at Snow Cosmetics.The company also used Rep AI to solve the problem of abandoned checkout. Typically, there is no way to reach the customer once they’ve left the site, but Rep AI can determine whether the visitor comes back from the IP address and present them with a special offer, according to McCray. That has contributed to a 30% conversion rate, he says.5. Learn and finesseThen it’s time to further enhance the quality of those recommendations. This is the move to the new way of working. Once one department sees how well an AI agent handles routine tasks, other departments will follow. This creates a ripple effect of adoption and innovation throughout the company.Colgate-Palmolive looks at how the AI assistants are improving time savings, which Pappas says is really “reallocation of time because I can produce something in one hour that would have taken me four hours to do.” Mainly, he says, officials are tracking the quality of output creativity and the offsetting of external costs. About 70% of employees have reported the assistants have improved the creativity of their work, and on average, a savings of about four hours per week from all the assistants they use, Pappas says. What the future holdsAgents will continue to evolve because of their interactions with humans, and their sophistication will grow, according to Wright, including their level of autonomy. Right now, agents are more passive and reactive, but when they start to understand and gain more context about the person using them, they will become “more emotionally intelligent” and more proactive.“Already, if you and I prompt ChatGPT with the same question, it will come back with a different answer,” he notes. Pappas confirms the next step at Colgate-Palmolive is for the agents to become more autonomous.They need to determine how much they will allow agents to do as employees become more comfortable using them, he says. Wright has no doubt that people will gain a high level of comfort and trust. An AI agent will become “almost like a best friend” and more empathetic, he says. “That’s why we’re finding more of a level of trust from the consumer perspective. And frankly, expect to see more adoption in the workplace because of its ability to be more relatable.”

How to Build an AI Agent Strategy: Lessons from Colgate and Mondelez

Related Topics

Business Topic