Tori

eToro's users come to the platform with very different needs. A new user wants to learn what CFD trading is. A confident trader wants to compare Nvidia and Tesla in depth. Someone halfway through a portfolio review wants to know which sectors they're overexposed to. The product had a separate surface for each — search, the Academy, the Discover page, the help center — but every one of them assumed you knew what you were looking for before you arrived.

Tori is the AI agent built into eToro that meets users where they are. One conversation answers four different kinds of questions: where is this thing in the app, what does this concept mean, what's worth looking at right now, and what's actually happening with this asset. The product principle was strict — Tori doesn't give advice. It helps people think.

Product Designer on Tori. I owned the conversational interface end-to-end: message structure, the icon and touchpoint behavior, the smart follow-up prompt strategy, and the contextual logic that decides which suggested questions appear on which page. Worked closely with product, product marketing, and engineering.

The team started where any consumer-product question should: with the people who'd be using it. Over a few weeks we ran more than 40 user interviews and a competitor scan in parallel — not to validate a solution we'd already drawn, but to find what was actually missing.

Three patterns emerged from the conversations:

Learners. Users who'd never opened the Academy, but were quietly intimidated by the product. They wanted plain-English explanations — what's a CFD, why does leverage matter, what does the risk score actually mean — without leaving the screen they were on. They wanted to invest more thoughtfully; they just needed the language translated.

Power users in a hurry. Active traders who knew exactly what they wanted but couldn't always find it. "Where's the dividend yield filter?" "How do I see the EU40 index?" "Can I compare two assets side by side?" The platform had the answers — buried under three menus or behind a search that didn't understand intent. They wanted speed.

Investors leveling up. Users in the middle: not beginners, not professionals — readers of finance Twitter and watchers of earnings calls who wanted to do deeper research on a stock or their own portfolio without paying for Bloomberg.

The competitor scan filled in the rest. AI chat experiences in finance ranged from chatbot-for-FAQs (useless) to opinionated advisors (regulatory non-starter). Nobody had attempted a context-aware companion that could serve all three of those segments without crossing into advice. That gap is where Tori was designed to live.

Tori is actually four specialized agents under one entrance. The decision to split them — rather than ship one generalist bot — came from research. Users asked questions in clusters, and the cluster told us the answer's shape. "Where can I find earnings dates?" wants a single navigation result, not a paragraph. "Compare Nvidia and Tesla" wants a structured side-by-side, not narrative. The four-agent split lets each conversation route to the right response format.

Navigation answers where is X in the app? — earnings dates, the stock screener, copy trading, the deposit flow.

Academy explains concepts — CFD trading, eToro's risk score, technical analysis, what Polkadot actually does.

Discovery surfaces opportunities — most undervalued stocks, popular investors who mostly trade crypto, top movers in a sector.

Research does the deep work — compare two assets, analyze a portfolio's exposure, walk through everything that's happening with AAPL this quarter.

Three principles shaped every screen:

Hand-holding throughout the conversation. Tori narrows the question instead of guessing wrong. Ask "show me good stocks" and Tori asks back: which sector? what risk level? Smart follow-up suggestions appear after every answer so users don't have to think about what to type next.

Widgets instead of free text whenever possible. When the next input is a category, a date, a risk level, or a comparison target, Tori renders a tile or chip — not "type your answer." Less friction, less ambiguity, fewer dead-ends.

Contextual by where you are. The same Tori behaves differently depending on the page it's opened from. From Home or Discover, it defaults to discovery questions. From a market page, research questions. From Watchlist or Portfolio, the suggested prompts reference the actual assets on the screen. Users get useful prompts the moment they tap the icon — they don't have to think first.

The entry point is a single icon in the top-right of the nav, next to notifications. It glows softly when Tori has contextual prompts ready for the user — for example, after they've spent a long time on a single page without acting. Tapping opens a full-screen experience with rotating contextual questions and an "ask anything" free-text option always within reach.

Tori is in a staged rollout — open to a subset of eToro users while the team tests, observes, and iterates. It hasn't shipped to everyone yet, and that's by design. We measure progress on two parallel tracks:

Functional signals — unique queries per session, follow-up clicks per user, thumbs-up ratio, and "successful sessions" (ones that end on a real action, not a closed tab).

Commercial signals — conversion to first trade, conversion to verification, conversion to first deposit, and the share of new users who interact with Tori during onboarding.

Every couple of weeks the dashboard gets paired with fresh user interviews. The conversation model, the contextual prompts, and even which of the four agents shows up first by default keep changing based on what we hear and what we measure. The screens above are where Tori is today — not where it ends.

The hardest design problem hasn't been conversational UI — it's been restraint. The temptation to make Tori smart and opinionated is enormous, and every advisory-style sentence is one regulatory step closer to a problem. The interesting work is in finding the gap between "useless" and "advice" — a place where Tori can be confident, contextual, personalized, and helpful, without ever telling someone what to do.

The other thing that keeps surprising me: how much of the design work is on the entry, not the conversation itself. A glowing icon in the right corner at the right moment, with the right prompt already loaded, does more for adoption than any improvement to the conversation that comes after. Most of the magic happens before the user types a single word.

The work is ongoing. Every test, every interview, every dashboard cycle moves Tori a step closer to what it should be — and away from what we initially thought it would be.