Every company you interact with likely uses artificial intelligence (AI) chatbots at some level — usually to handle your customer service issue, answer questions about a product, or help guide you through an FAQ log. Palona AI wants to take that approach further: turning those chatbots into professional-grade AI sales agents who double as charming brand mascots.
Also: AI agents may soon surpass people as primary application users
The startup, officially launched today, aims to give small businesses like pizza shops and wellness centers the power of consumer-facing human sales experts, which they likely can’t afford. Palona “onboards” or trains the agent using your business’ knowledge base, including employee manuals, company policies, and existing FAQs.
The goal is to create what the company calls a highly customized “AI character” that proactively engages with customers in the company’s voice, using its guidelines as personality traits. Available for web, iOS, and Android, Palona agents are built into a business’ backend, and customers can text, call, or chat with them 24/7.
Users can text a pizza chain’s Palona agent with their regular pizza order.
Palona AI
Palona’s agents can reportedly remember customer information after an initial interaction — say, if you asked a camera retailer’s chatbot about pricing on Wednesday, but got distracted and couldn’t return to the conversation until Sunday. Agents develop customer profiles full of conversational tidbits, like the name or gender of the customer’s pet, that they can use in future interactions.
“Building relationships with customers requires a precise mix of human skills that even the best salespeople train for years to achieve,” said Maria Zhang, Palona CEO and co-founder and former VP of engineering at Google, in the press release. “We captured that skillset and turned it into models and actions our AI agents can consistently perform.”
According to Palona, its base model, built using Llama and certain OpenAI large language models (LLMs), is trained on psychology and persuasion literature, similar to human sales experts.
Also: Autonomous businesses will be powered by AI agents
This approach is what the company claims makes its agents adept at upselling and building brand loyalty. After customization, Palona supplements each agent’s knowledge with general industry and competitor information for a “more authentic” customer interaction.
“Using Palona, businesses are able to quickly create a customized ‘Employee of the Century’ that combines soft sales skills with 24/7 availability, unlimited capacity, and infinite patience,” a Palona rep told ZDNET.
Also: Managing AI agents as employees is the challenge of 2025, says Goldman Sachs CIO
If you prefer to immediately escalate to a human representative during a customer service interaction (like I do), the joke’s on you: Palona says its base bots can handle most of what you throw at it.
Using supervisor agents, the autonomous system monitors the customer-facing agent, escalating the conversation to a human when necessary (based on complaints about a pizza shop bathroom or human employees, for example, which the AI cannot address). Palona says its bots have a much higher threshold for escalation than many others on the market.
This multimodal system results in “AI that learns about each individual customer, adapts to their needs, and takes action on their behalf with precision and accuracy,” the release says. Supervisor agents also monitor for hallucinations, which Palona says they reduce by 98%.
EQ claims
The tech industry is rife with AI agents that help employees do everything from write code to prep presentations. Salesforce released its own solution last fall, as did Microsoft, Asana, and countless other IT firms. But Palona claims that what sets its agents apart is their “unrivaled” emotional intelligence (EQ), which includes “humor” and “current-day texting etiquette”, explained Professor Steve Liu, Palona’s chief scientist, in the release.
Also: AI agents might be the new workforce, but they still need a manager
During a demo with ZDNET, CTO and co-founder Tim Howes explained how the company fine-tuned available models using “an eight-point definition of what EQ is”, as well as markers of human likeness and “delightfulness”.
“In order to do what we’re doing — which is apply AI and AI agent technology to the sales part of the customer experience as opposed to customer support — we need a much higher degree of EQ, of human likeness, of persuasion, built into the model,” he added.
In the demo, Palona’s agents congratulated a user on getting a new kitten, pulled up photos of a pet camera it recommended, and likened its subscription cost to that of a cup of coffee. To compare, I told ChatGPT that I had just gotten a kitten (unfortunately not true), and it responded with similar attention and enthusiasm:
When I asked ChatGPT to suggest kitten toys, it provided a list and asked if I wanted recommendations based on my “kitten’s personality? 😊”
To measure proficiency, Palona created “an objective benchmarking system” to compare its agents’ EQ with other commercial or open-source LLMs. The company’s researchers conducted experiments measuring agents’ patience, context awareness, conflict resolution, empathy, humor, problem-solving, and more — finding its own AI to be “the industry leader in empathy and intuition”.
Also: DeepSeek’s new open-source AI model can outperform o1 for a fraction of the cost
Of course, any company’s internal benchmark should be taken with a grain of salt. That said, Palona has submitted its paper, pending review, to the Nations of the Americas Chapter of the Association for Computational Linguistics (NAACL).
A Palona rep told me its agents “consistently outperform both human staff and other AI solutions on industry benchmarks,” but when asked, did not clarify which benchmarks (such as EQ-Bench). “We created the benchmark and run tests against our models and other models in the industry. There is not current industry standard for EQ, that’s why we created one,” Zhang commented via email.
AI agent limitations
Personality training aside, are there simply some items only a human can sell?
Also: The journey to fully autonomous AI agents and the venture capitalists funding them
When I asked Palona’s founders about agents’ ability to sell products that are especially embodied or experiential — like bikes or cars — they didn’t exactly have a read (unlike an aggregating AI that scrapes human product reviews from the internet, Palona agents work off of company data).
One of the company’s early customers is a bike shop; they cited the agent’s ability to direct consumers to its closest location or refer them to an expert in the shop. Ostensibly, it’s been successful.
Also: Why some companies are backing away from the public cloud
Whether a human wants to buy a bike from a bot remains to be seen. However, as AI agents continue to infiltrate all levels of commerce, the question may yield interesting or unexpected answers.
That said, humans tend to enjoy engaging with bots. One of Palona’s customers, a West Coast pizza chain, created an AI brand mascot named “Jimmy the Surfer”, who got as many questions about what the waves were like on a given day as it did pizza orders. Zhang also noted that users appeared to ask Palona agents questions they may be too embarrassed to ask a human employee, like whether they have to wear clothes in a sauna.
“People are not patient innately,” she said. Much like in other human-computer interactions, agents can make social anxieties and norms less present for users.
To learn more about Palona and explore solutions, visit palona.ai.
