How Hello Uses AI to Find What Humans Miss

On October 7, 2025, three people shared a stage at Variety’s Future of Brand Summit to argue that AI makes marketers better rather than unnecessary: Diana Haussling, CEO of Colgate-Palmolive’s Hello Products, Andrés Ordóñez, then global chief creative officer of FCB, and Jarrod Martin, then CEO of Acxiom and Kinesso. Marketing Dive covered the panel, and the line worth keeping is Haussling’s explanation of why her team needed AI at all: “Because we’re so close to it, sometimes it’s hard to find those things” that consumers actually want.

A year on, the argument has held up better than the org chart. Omnicom closed its acquisition of IPG in late November 2025 and days later announced it would retire FCB, along with DDB and MullenLowe, with FCB absorbed into BBDO. Martin became chief transformation officer of Omnicom Media, and in July 2026 the Kinesso brand was retired as well. Haussling still leads Hello, and Colgate-Palmolive has pushed AI further into the work she described. CEO Noel Wallace put the process plainly: “We start to test these ideas with AI, and we refine them down to the bigger ideas that we can launch around the world,” as Consumer Goods Technology reported in March 2026.

Hello’s AI exists because the team is too close to the brand

The tool is simple to describe. According to Marketing Dive, Hello uses a proprietary AI system that synthesizes data from “hundreds of thousands of different resources” to surface consumer needs the brand is not yet meeting. The team also uses AI to make animatics, rough animated versions of an ad, so concepts can be tested before anyone pays for production.

What the tool does not do is choose. Haussling was direct about that: “You still have to have a gut, you still have to have critical thinking skills.” The AI widens what the team can see. People still decide what gets made.

Put that way, it sounds like every company’s AI statement. The part worth studying is Haussling’s reason for needing the tool: her team knows the brand too well, and the blind spot that creates is about proximity, not skill. That sounds like modesty. It is a precise description of how experienced teams miss things.

A brand team carries a working model of its customer: why people buy, which words they use, which complaints matter. That model is what makes the team fast. It is also a filter. A review that fits the model gets noticed. One that does not gets read as noise, an outlier, a customer who misunderstood the product. After a few years, much of what the team calls research is confirmation of what it already believes.

An AI system reading hundreds of thousands of sources has no model to protect. It counts. If a phrase keeps appearing next to a product category, it surfaces, whether or not anyone at the brand would have thought to look for it.

Haussling named one place this shows up: “The generational divides have really driven different types of language.” Different age groups describe the same need in different words. A team writing in its own vocabulary can miss that younger customers are asking for exactly what it sells, in terms it never uses. Pattern-finding across large volumes of text catches that mismatch, because it is a question of frequency, and frequency is what machines are good at.

How far synthetic consumers can be trusted

Colgate goes further than reading what customers write. It builds what it calls digital consumer twins, AI models that respond to product concepts the way real shoppers would. MIT Sloan Management Review reported that the company compared the twins’ feedback with real consumer responses and found “a high level of concordance between the two groups.”

The most detailed public evidence comes from a study PyMC Labs ran with Colgate-Palmolive, covering 57 personal care product surveys and 9,300 human responses. Using a method in which the model answers in words and those answers are then mapped onto a rating scale, the synthetic respondents reached 90% of human test-retest reliability, Research Live reported. One of the researchers, Nina Rismal, was careful about what that means: language models can mirror human decisions “under specific conditions, and within well-defined contexts.”

That caveat is the useful part. A model trained on what people have already said and rated is strongest where there is plenty of history: established categories, familiar formats, concepts that resemble things shoppers have seen before. By the same logic, it has the least to go on when something is genuinely new. Use synthetic consumers to screen fifty concepts down to five. Do not use them to decide whether a category that does not exist yet will work.

Effectiveness is the harder test

Martin made the point most of the industry skips. He told the audience he would like to hear more stories about AI’s effectiveness and fewer about its efficiency. Faster production is easy to measure and sell. Better decisions are harder to prove, and they are where the money is.

The panel offered examples. FCB built synthetic audiences of 10,000 members to help judge creative work. Acxiom said AI-built audience personas, five of them, produced a 23% uplift in creative performance for a major sports league. Neither came with a published methodology, so treat them as claims, but both point the same way: the value is in choosing better, not in producing more.

The same split runs through consumer goods. Some companies aim their AI money at making more advertising for less, as in Mondelez’s $40 million bet on AI-produced marketing. Hello aims earlier, at finding out what to say before anything is produced. When every competitor can generate content cheaply, the second use is the one that builds an advantage, the argument behind why human copy wins back ground once everyone generates the same AI copy.

You can run the same check without a proprietary tool

You do not need Colgate’s budget to copy the method, because the method is a comparison, not a tool.

1. Write down what your team believes: why customers buy, what stops them, the words they use. Do it before looking at any data.
2. Collect what customers actually wrote: reviews on your site and on marketplaces, support tickets, sales call notes, social comments and the search queries people use to find you.
3. Have AI cluster and count it. Ask for recurring themes, how often each appears and quotes that show it is real.
4. Compare the two lists. The gaps matter more than the overlaps: needs customers mention that your team never listed, and reasons your team is sure of that customers rarely raise.
5. Test the gaps with real money. Write ads around the new language and run them against your current messaging.

The last step is where most teams stop short, and it is the job of performance marketing built around testing one message against another. If you are not sure what customers see and say about you in the first place, an audit of how your brand shows up across search, social and review sites is the input everything else depends on.

Hello never claimed AI is smarter than its team. The claim was that the team is too close to see everything, and a system that counts without caring can show it what it stopped noticing. That is a narrow job, and it is the one AI does best in marketing.

Frequently Asked Questions

Who owns the Hello toothpaste brand?

Colgate-Palmolive owns Hello. It announced an agreement to acquire the oral care brand in January 2020, and Hello has operated as a separate brand inside Colgate’s portfolio since then, selling toothpaste and other oral care products under its own name, led by CEO Diana Haussling. Hello’s AI work sits inside Colgate-Palmolive’s wider push into AI-assisted research and concept testing, while the brand keeps its own voice as a challenger.

What AI tool does Hello use?

Hello uses a proprietary AI tool that synthesizes data from hundreds of thousands of sources to surface needs the brand is not yet meeting, according to Marketing Dive’s coverage of a 2025 panel with CEO Diana Haussling. The team also uses AI to create animatics for testing ad concepts before production. Public coverage does not name the vendor or the underlying model, so the useful lesson is the method rather than the software.

Can AI replace traditional consumer research?

Partly, and within limits. A study by PyMC Labs with Colgate-Palmolive found that language models, prompted in a specific way, reached 90% of human test-retest reliability across 57 product surveys. That makes AI useful for screening many concepts quickly and cheaply. It does not make human research obsolete: the models learn from what people have already said, so they are least reliable for new categories, and their results still need checking against real customer behavior.

What happened to FCB after Omnicom bought IPG?

Omnicom completed its acquisition of IPG in late November 2025. On December 1, 2025, it announced that three IPG creative agency brands, DDB, FCB and MullenLowe, would be retired, with FCB absorbed into BBDO. Omnicom also said it expected to cut about 4,000 more jobs to reach its cost-saving targets. With FCB absorbed into BBDO, the name is one of the merger’s casualties.