Sara Is All You Need: How Slow Shopping Shapes AI-Powered Decision-Making

We’ve observed that most conversational assistants are optimized for speed.
They answer questions, complete actions, and close transactions.
They are mostly efficient, predictable, and often forget previous interactions.

In many cases, that works. But not always.

Many decisions are not purely functional. Choosing a pair of glasses, a piece of clothing, or a personal care product isn’t just completing a task; it’s an act of self-expression. Yet, many digital experiences treat these choices as micro-operations to be optimized: fewer clicks, less time, and more conversions.
We’ve made interactions frictionless, but often superficial as well.

The result is paradoxical. Decisions happen faster, but not necessarily better.
Instead of helping us understand what we want, systems help us decide sooner.
Which raises a simple question: are we optimizing the right variable?

The Philosophy of Slow Shopping

Within the team, we started calling this idea Slow Shopping: designing systems that slow down just enough to encourage reflection, context, and intention. Not less efficient, but more aware.

The idea has roots in the Slow Food movement, which emerged in 1980s Italy as a cultural response to industrial fast food and the acceleration of daily life. Slow Food reframed eating as an experience to be savored, one intertwined with community, origin, and meaning. Slow Shopping does something similar for purchasing. It treats a buying decision not as a task to complete, but as a process that deserves emotional and contextual depth. A good shopping experience shouldn’t just be fast. It should be intentional.

There’s a useful lens for this in behavioral economics. Daniel Kahneman distinguishes between two modes of thinking: System 1, which is fast, intuitive, and automatic, and System 2, which is slow, deliberative, and effortful.

Most digital commerce is designed to exploit System 1, triggering impulse through scarcity cues, countdown timers, and one-click checkouts. Slow Shopping asks a different question: what if we designed for System 2 instead? What if we gave people the space to think before they buy?

This isn’t about adding friction for the sake of it. It’s about creating the conditions for better decisions, the kind that people feel good about not just in the moment, but afterward too.

Over the past year, we at WordLift have begun to explore a different hypothesis.
What if an agent could think differently, slow down, and not just respond?
What if it could engage in a multi-turn conversation, understand intent over time, and help people make informed decisions rather than rush them into action?

This is what sparked an internal R&D project we named SARA (Shopping AI Research Assistant).

Starting From A Question

Like many of our projects, we started with research rather than a product. There was no commercial roadmap or anything like that. It started as a space to explore and test ideas, guided by four main concepts.

1. Voice interfaces: we wanted to understand how spoken interaction changes the dynamics of shopping. Voice makes exchanges feel more natural and lowers the cognitive load of navigating product features compared to clicking through menus.
2. Multi-turn conversations. Instead of single question-and-answer exchanges, we were interested in sustained dialogue where the agent remembers context and builds on previous turns over time.
3. Reasoning chains. We wanted the agent to think step by step, processing what the user needs, their constraints, and their emotional signals before jumping to a recommendation.
4. Human-centered interaction design, meaning we prioritized user well-being, reflection, and satisfaction over conversion metrics.

But we didn’t want to build another chatbot; we wanted to go deeper.
The goal was clear: to keep the agent engaged in the conversation long enough to be genuinely helpful.
The initial interactions were intentionally minimal, with nothing complex behind them, so we used speech recognition, basic dialogue management, and lightweight reasoning steps. Nothing was polished yet.
We weren’t trying to impress anyone. We wanted to see what would break first.
From there, we doubled down on context, intent tracking, and conversational continuity. We didn’t just want to automate; we wanted more reasoning and thought.

Principles That Guide SARA

The architecture, hypotheses, and evaluation methods were then formalized, albeit through early, small-scale testing, in my bachelor’s thesis, “Sara Is All You Need,” which positioned SARA not only as a conversational agent but also as a reasoning system. It drew on interdisciplinary theory, pulling together ideas from affective computing, behavioral economics, ethical design, and consumer psychology.

Since then, we’ve been guided by four principles.

Engagement over conversion, which means measuring success by the quality of the dialogue, not the speed of the checkout. Multi-turn over single response, sustaining conversations across multiple exchanges so the agent can understand evolving needs rather than react to isolated queries. Support over persuasion, helping users clarify what they actually want rather than nudging them toward what’s most profitable. And Slow Shopping as a core philosophy, designing every interaction to encourage reflection, emotional awareness, and intentional decision-making.

What makes SARA fundamentally different from traditional conversational AI is this last point. Most assistants are built to minimize time-to-action. SARA is built to maximize the quality of the decision. While many assistants encourage quick, impulsive decisions, SARA is designed to support a more thoughtful process, a conversation that helps clarify intent rather than anticipate it.

This framing turned SARA into something we could measure and improve, not just demonstrate. Early pilot testing across 45 sessions showed that when the system worked well, conversations reached a median of 11 exchanges and lasted over a minute. That’s a strong signal that users were willing to engage in the kind of reflective dialogue the Slow Shopping framework calls for. When sessions failed, the cause was overwhelmingly technical, not motivational. People wanted to talk. The technology just needed to keep up.

From Lab to Life: Real-World Experiments

Shortly thereafter, the idea moved beyond our internal lab to be tested in real-world contexts and with real-world limitations.

SARA started to become a technology, something we could label “powered by SARA.” The first experiment involved developing an early prototype for a major eyewear retail partner.

The challenge was concrete: how does an agent integrate with an existing product catalog and in-store experience? How much does intentional friction, the kind that encourages reflection, actually aid the decision-making process? And where does guidance cross the line into annoyance?

What we built was a conversational assistant that could support the entire retail journey, from initial online discovery to in-store engagement. Think of a customer walking into a store looking for stylish, high-performance polarized sunglasses for driving. Instead of browsing dozens of options alone, they could talk to SARA, get personalized style advice, and then filter by brand, price, and key features like polarization and lens color. The agent bridged the gap between the digital and physical experience, what the industry calls “phygital” (physical + digital). offering both customers and staff instant, accurate product guidance powered by structured data.

For the brand, this meant something practical: the ability to offer a richer, more guided shopping experience without replacing any of their existing systems. The agent layered conversational intelligence on top of what was already there, improving customer experience and giving staff a tool they could rely on for real-time, accurate product information across all touchpoints.

The Result

The conversational model could hold up outside the lab: users engaged with the agent, explored options, and stayed in the dialogue. At the same time, integration with real inventory, real constraints, and real customer expectations exposed limitations much faster than internal testing ever could. And perhaps most importantly, the underlying principles of multi-turn reasoning, intent tracking, and reflective pacing proved to be genuinely transferable to a live retail environment.

SARA was no longer just an experiment.

Beyond an Assistant: SARA as a Platform

At some point we noticed a shift in how we were using SARA internally. We had gone from “shipping SARA” as a standalone product to reusing its components in almost everything: dialogue logic, reasoning modules, orchestration layers. Different projects required different agents, but all of them were built on the same foundation. The focus had shifted from a single assistant to an infrastructure, a set of systems powered by SARA.

SARA evolved from a product concept into an internal infrastructure, a reusable layer for creating agentic experiences. This transition from assistant to platform likely marked a turning point for the WordLift Lab.

Agentic AI Beyond Retail

We then began exploring how the same principles could apply outside of ecommerce. Through collaborations in cultural and non-profit settings, we prototyped agentic experiences focused on guidance, education, and interaction rather than transactions.

In the cultural sector, we explored how a reasoning agent could help visitors navigate complex exhibitions. The idea was to provide context-aware explanations that adapted to a visitor’s curiosity and pace, much like a knowledgeable guide who listens before speaking.

In the non-profit space, we worked on agentic approaches for social impact initiatives. Here the challenge was very different from retail. It wasn’t about selling a product. It was about helping people access information, understand their options, and feel supported in making important decisions. What did we learn? That the same core capabilities, multi-turn dialogue, contextual reasoning, and a patient, non-pushy interaction style, proved just as valuable when the goal was social impact rather than a sale.

These projects reinforced a key insight: the value of an agent that reasons and maintains context is not limited to retail. Whether someone is choosing a pair of glasses or navigating a social service, the need for thoughtful, multi-turn interaction is the same.

The pattern is consistent: when an agent can reason and maintain context, it becomes an interface, not just a tool.

SARA Today: A Foundation for Human-Centered AI

Looking back, SARA was never intended to be the final product. It was a starting point.

Today, the WordLift Lab brings together what we learned from those experiments: building agents that reason, designing conversations that span multiple turns, testing prototypes with real partners, and focusing on human-centered outcomes rather than pure automation.

The broader landscape is moving fast. Voice commerce is projected to grow significantly in the coming years, and agentic commerce, where AI agents participate in the full shopping cycle from discovery to payment, is becoming a reality. Major players are already enabling purchases directly within conversational interfaces, with secure tokenized payment systems that let agents transact on behalf of users.

But speed alone isn’t the answer. We think SARA points to a third path: commerce that is fast enough but remains ethically anchored, connecting commercial activity with user well-being and sustainability goals. 

Imagine an agent that, before completing a purchase, summarizes the product’s provenance, price, and environmental impact, and simply asks: “Would you like to continue, or do you need more time?” That single question transforms checkout from a reflex into a conscious affirmation. That’s the frontier of agentic commerce we’re most interested in.

SARA remains the foundation, not because it’s perfect, but because it gave us a place to explore. And that exploration is still ongoing.

The future of commerce isn’t just faster—it’s more intentional. At the WordLift Innovation Lab, we are turning the Slow Shopping philosophy into a reality. Explore our latest experiments in agentic AI and see how SARA is redefining the relationship between humans and machines.

Discover the Innovation Lab →