#pymntsAI

With the rise of generative artificial intelligence (AI), will true personalization finally be within reach?

“Personalization has long been a digital promise that’s failed to fully deliver,” Ja-Naé Duane, lecturer at Brown University and research fellow at MIT CISR, told PYMNTS. “We’ve been told for decades that our clicks would unlock tailored experiences, but if you’ve ever scratched your head at a bizarre Netflix recommendation, you know we’re not quite there yet.”

That may be changing. According to Duane, the latest AI systems are moving past rule-based segmentation toward real-time learning and contextual awareness.

“Today’s systems don’t just track what we do; they infer how we feel,” Duane said. “They shift tone mid-conversation, rewrite content on the fly, and evolve as our needs change.” But she also acknowledged AI systems’ limitations. “True personalization is a moving target because human desire is fluid, contextual and often contradictory.”

Past efforts also have come up short because they rely more on segmentation instead of truly understanding the target audience.

“Most so-called ‘personalization’ over the past two decades has been little more than categorization,” Puneet Mehta, CEO of Netomi, told PYMNTS. “Show sci-fi to the sci-fi crowd. Offer 10% off to frequent buyers. That is not personalization. That is pattern matching.” He said what’s changing now is the ability to add a “true human element” to digital interactions, including emotional intelligence and memory of prior exchanges.

According to a PYMNTS Intelligence report, while 83% of consumers are receptive to personalized offers, only 44% find them “very relevant” to their needs and 17% said these were “completely irrelevant.” Moreover, personalization of offers can be a more persuasive selling tool than the discount amount. “The true value of personalizing an offer to a consumer lies in how tightly the merchant tailors it to that consumer’s preferences rather than how much the discount is,” according to the report.

Meanwhile, an SAS survey showed that while 75% of marketers are using generative AI in their daily tasks, only 19% are using it to target audiences. This is “not optimal” because when marketers use gen AI for personalization, 92% said they are seeing higher ROI.

The culprit? Lack of understanding about what gen AI can do at the CMO and senior management level, according to SAS. A separate report by Data Axle cited additional barriers such as organizational silos, legacy systems and regulatory complexity as blocking true personalization. That’s even though 91% of marketers believe that blending a consumer’s personal and professional data will lead to better audience targeting.

Read more: Personalized Offers Are Powerful – But Too Often Off-Base

Moving From Data to Action Is Complex

Jackie Walker, retail experience strategy lead, North America, at Publicis Sapient, told PYMNTS that personalization faltered for several reasons.

“Part of that is about inputs, like the quality of the data that a business has about its customers and how usable it is, and the other part is about outputs, such as what meaningful changes can be made to the interactions that a business has with its customers across the wide range of customer touchpoints,” Walker said. “The complexity inherent in both of these areas has resulted in redefining personalization to just mean better segmentation. … Customers, however, are smarter than that.”

For example, if a diner always orders a burger with no pickles and a restaurant recommends a new Southern-style sandwich with extra pickles, this isn’t meeting the consumer’s needs. For a business, understanding this data for thousands of customers is complex: They have to stitch together the data and perhaps infer some attributes of the customer, then figure out how to act on the data to make an offer.

“AI truly has the power to tackle both sides (consumer and business) of the problem,” Walker said. “With AI, data aggregation and interpretation become far more accessible. You can consider more scenarios than were previously possible. There’s the ability to tie together data assets you couldn’t before.”

But Lei Gao, CTO of SleekFlow, told PYMNTS that while AI has made strides in delivering true personalization, “there is still a disconnect between promise and delivery.”

That’s because artificial intelligence is only as good as the data it ingests. The problem is most businesses still have their data in silos, or lack the infrastructure to make their customer data available for use in real time across systems.

“Without combined, high-quality streams of data, AI recommendations will always fall short,” Gao said.

Second, Gao said true personalization is not just about what a customer likes or dislikes. The context is important as well. “In conversational commerce, for example, a product recommendation will be a function not just of a customer’s purchase history but also of their location in the purchasing process at the time,” Gao said.

Data also should be as fresh as possible to be most effective.

“Personalization is only as effective as the data behind it,” Dean de la Peña, vice president at Resonate, told PYMNTS. “If it’s working with outdated or incomplete information, the insights you receive can be off base.” For personalization to work, he argued, brands need access to real-time signals that reflect not just past behavior but current context — particularly in volatile times when consumer sentiment shifts rapidly.

For example, if a retailer based this year’s marketing campaigns on last year’s back-to-school spending trends, it risks missing the market, de la Peña said. Consumer priorities can shift quickly in response to factors like inflation, fashion trends, shifting cultural attitudes and more.

But not all personalization efforts have failed. Some have succeeded too well.

“This is a two-sided story,” Siamak Freydoonnejad, co-founder at Sprites-AI, told PYMNTS. “On one side, AI has already made significant advancements in personalizing content for audiences. One of the most compelling examples would be TikTok or Instagram. The recommendation systems of these platforms curate visual content that feels like it gets you. That’s precisely why it’s so addictive.”

However, “Spotify’s Discover Weekly AI algorithm results are either extremely relevant or horrifically out of touch with reality,” Freydoonnejad said. “These algorithms are highly adaptive and precise and will get it right in most cases. You don’t have to search for what you like; it’s already there in your feed. And that’s the beauty and the curse.”

So will generative AI fix personalization? “While we’re definitely closer today, ‘true personalization’ is still a moving target,” SleekFlow’s Gao said. “It’s less about a magic algorithm and more about building robust, thoughtful systems that combine technology with human-centered design.”

Read more:

Stitch Fix: AI-Powered Personalization Will Overcome Any Macro Challenges

Retailers Embrace the Power of Hyper-Personalization

Retail Personalization Is About More Than Just Sending Offers

According to a report Wednesday (July 16) by Bloomberg News, the artificial intelligence (AI) startup is planning a new investment round that could bring it to that valuation.

The report, citing sources familiar with the matter, said that Anthropic is not formally fundraising, though pre-emptive funding offers from venture capitalists to high profile AI startups have become standard in Silicon Valley. Investors have approached Anthropic signaling they would invest at a valuation north of $100 billion, the sources said.

Anthropic was valued at $61.5 billion earlier this year following a $3.5 billion funding round. PYMNTS has contacted the company for comment but has not yet gotten a reply. 

The report notes that the investment discussions are taking place amid a surge in revenue from Claude, Anthropic’s chatbot, with annualized revenue jumping from $3 billion to $4 billion in the past month.

Founded in 2021, Anthropic’s investors include both Google and Amazon, the latter of which is said to be weighing a new multibillion dollar injection into the company on top of the $8 billion it has already invested.

Meanwhile, PYMNTS spoke earlier this week with Jonathan “JP” Pelosi, head of FSI at Anthropic, about the company’s new Claude for Financial Services offering, an AI solution designed for analysts, portfolio managers and underwriters at large financial institutions.

It’s the first industry-specific service that Anthropic has formally introduced, Pelosi said.

“Where we saw a lot of traction early on was with these high-trust industries,” he added. “Our models, our solutions, are just very well positioned to help these firms.”

Still, the AI industry has a lot to prove before chief financial officers grow comfortable with the technology, according to the PYMNTS Intelligence report “The Agentic Trust Gap: Enterprise CFOs Push Pause on Agentic AI.” 

That report found concerns about the technology that include hallucinations, where an AI agent can go off script and expose companies to cascading payment errors and other inaccuracies.

For highly regulated industries such as financial services, the accurate responses of generative AI models are critical, and Pelosi said the new tool comes through on that front.

The worry over hallucinations has “significantly stymied meaningful adoption in the financial industry,” Pelosi said. “If you and I are in the business of making very large investments or analysis on very high-stakes transactions, we don’t have the luxury of saying, ‘Hopefully that [calculation is] right.’”

Still, Pelosi stopped short of saying Anthropic has completely solved hallucinations.

Anthropic isn’t saying that “Claude would never hallucinate again,” he said, though “it’s making it easier and easier to validate the numbers that you’re making very big decisions on.”

That’s according to a report Wednesday (July 16) by the Financial Times (FT), which says this move is part of a plan by the artificial intelligence (AI) startup to derive new revenues from online shopping features.

The company already displays products on its platform with the option of clicking through to online retailers, and inked a partnership with eCommerce platform Shopify in April.

Now, sources told the FT, OpenAI hopes to integrate a checkout system into ChatGPT, to make sure users complete transactions within the platform. Sellers who make sales this way would pay a commission to OpenAI.

PYMNTS has contacted OpenAI for comment but has not yet gotten a reply.

The FT notes that this marks a shift in strategy for OpenAI, whose revenue has chiefly come from subscriptions to its premium services. Getting a slice of eCommerce sales would let OpenAI earn money from people using its free service.

The report adds this move is another threat to Google’s business model, with consumers increasingly turning to AI chatbots for online search and shopping. OpenAI is already building its own browser in a separate challenge to Google, per a report last week by Reuters.

While the feature is still in development, the sources told the FT that OpenAI and partners like Shopify have been presenting early versions to brands and discussing financial terms.

Writing about the OpenAI/Shopify partnership in April, PYMNTS noted that rival Perplexity already offers its own AI shopping assistant, as well as a free merchant program to allow retailers to share their product specifications so shoppers can find their products. 

Meanwhile, PYMNTS CEO Karen Webster spoke recently with Rezolve Ai CEO Daniel Wagner about the potential for AI to transform online shopping.

“Taking a customer from a query to a purchase is an art form,” Wagner said. “We’ve trained our model on psychographics and closing techniques. We’re giving retailers the best salesman on the planet… It understands not just product specifications but the utility of features like why an OLED screen matters, why a fast shutter speed benefits sports photography.”

Modern AI retail systems, that report added, are anticipating needs as well as responding to queries. While at one time chatbots offered limited, scripted responses, AI-powered sales assistants are now virtually indistinguishable from their human counterparts, argued Wagner.

“This is the evolution of digital engagement,” he added. “Today’s online shopping is old-fashioned and ineffective. Consumers will start to demand the new way.”

The system, known as Athena, is the U.K. bank’s first major deployment of generative artificial intelligence (AI), according to Finextra.

Designed to serve as a centralized knowledge base, Athena can sift through 13,000 internal articles to provide quick answers to frontline employees in branches, call centers and other customer-facing roles.

The bank said this has cut the average time to find information from 59 seconds to 20 seconds, a 66% reduction.

For telephone banking teams, the bank estimates that Athena will save 4,000 hours annually that would otherwise be spent searching for information or keeping customers waiting.

“Athena is a monumental leap in our digital and strategic evolution, as we harness the power of generative AI to supercharge efficiency and elevate the customer experience,” according to Ranil Boteju, group chief data and analytics officer at Lloyds. “This technology isn’t just an upgrade – it’s a revolution.”

The system has seen wide adoption across the organization since its rollout earlier this year. According to the bank, 21,000 employees have used Athena to conduct more than 2.1 million searches so far in 2025.

Lloyds says it plans to expand the tool’s use to more employees in customer support roles, targeting as many as 40 million searches by the end of the year.

“We are freeing up thousands of hours as Athena puts critical information at our colleagues’ fingertips, leaving them free to help our customers with more complex, bespoke needs,” Boteju said.

Athena builds on Lloyds’ broader strategy to embed AI across the organization. Earlier this year, the bank announced it was developing a new machine learning and generative AI platform on Google Cloud’s Vertex AI. The platform is expected to drive both revenue and productivity gains, with a projected value of at least 50 million pounds ($70 million) in 2025.

Lloyds is developing an underlying architecture for AI agents that can be deployed for different use cases, including financial advice, software development or underwriting.

Working with Google engineers, the bank was able to develop a working prototype of an AI agent after a 12-week sprint. The agent would interact directly with consumers to give financial guidance. The bank is planning to launch a consumer-facing AI agent as early as August.

Read more:

Lloyds Shows Interest in Buying FinTech Curve

Lloyds Launches Virtual Card Partnership With Taulia

Lloyds Shuttering 136 Branches Amid Digital Banking Shift