How Thry Works

Thry takes a plain-language investment hypothesis and pressure-tests it against real market data, financial news, and the opinions of 93 tracked analysts and commentators —then delivers a fully weighted, backtested portfolio in under 60 seconds.

This page explains every step of the process: what's automated, what's deterministic, and where the limitations are. No black boxes.

Step 1: Hypothesis Parsing

Every hypothesis passes through a safety gate before anything else happens. The system scans for personal information —card numbers, emails, phone numbers, identity strings —and rejects the request instantly if any are detected. Nothing leaves the system.

Once cleared, an AI agent parses your thesis into a structured investment framework:

• Stance detection. Bullish, bearish, or neutral —including the core directional claim and the reasoning behind it
• Bull and bear cases. The strongest arguments for and against, with near-term catalysts and warning signs
• Themes and drivers. Macro themes (AI infrastructure, deglobalisation, rate cycles), sector exposure, and conviction drivers
• Trade implications. Who benefits if you're right? Who gets hurt if you're wrong?
• Evidence framework. What data would validate this thesis? What would disprove it?

This structured output drives every subsequent step. Whether you write "European defence will outperform as NATO procurement accelerates" or a single sentence with less detail, the parser extracts the same analytical signal. Everything downstream works from structure, not raw text.

Step 2: Narrative Search

Thry searches for what the market is actually saying about your thesis across two parallel tracks —covering financial news and current web sources alongside the published opinions of independent analysts and market commentators.

Headlines (Financial News & Web)

Financial news headlines are pulled from market-data news feeds and supplemented with a real-time web search to catch breaking coverage. Each article is scored for relevance against your parsed thesis using deterministic term matching, then classified as bullish, bearish, or neutral.

Alignment classification is polarity-aware: a bearish article supports a bearish thesis. This sounds obvious, but most sentiment tools get it wrong.

Voices (Analysts & Commentators)

This is where Thry goes deeper than any headline scanner. We maintain a continuously growing database of 1,600+ pre-extracted investment opinions from 93 tracked channels —financial YouTube creators, independent analysts, and market commentators with a combined audience of over 94 million subscribers.

When your hypothesis arrives, the system performs a two-stage search:

• Keyword matching with intelligent synonym expansion (e.g., "defence" also searches "defense," "military," "arms") —casting a wide net across the database
• Recency sweep to catch recent opinions that may use entirely different terminology but address the same investment theme

An AI model then scores each matched opinion on two dimensions: relevance (is this actually about the same topic?) and alignment (does this voice agree or disagree with your directional view?).

Each source carries its own individual credibility weight —a respected macro analyst and a hype-driven personality are not treated equally. The aggregate voice score reflects the quality-weighted consensus across all matched opinions.

On the report, channel influence is displayed on a logarithmic scale. A channel with 1 million subscribers gets roughly 2x the visual weight of one with 1,000 —not 1,000x. Subscriber count informs context, but it doesn't overpower the analysis.

Step 3: Equity Discovery

This is where Thry finds the right stocks to express your thesis —and it works very differently from a simple stock screener.

An AI research agent receives your full parsed hypothesis and identifies 10-20 publicly traded companies that express your investment view. The agent doesn't just match keywords. It reasons about your thesis to find:

• Direct beneficiaries. Companies that win most directly if your thesis plays out. For "European defence stocks will outperform," that means Rheinmetall, BAE Systems, Leonardo, Thales.
• Indirect beneficiaries. Second-order exposure —component manufacturers, logistics firms with military contracts, infrastructure enablers that benefit from the same tailwind.

Each equity receives an alignment score (0.0-1.0) measuring how directly it expresses your thesis, along with a written rationale explaining why it was selected. The agent includes a mix of large-cap and mid-cap names from any relevant market —it's not limited to US stocks.

Step 4: Technical Analysis

Once equities are identified and validated, the system fetches real-time market data for each one:

• RSI (Relative Strength Index). Momentum indicator that flags overbought or oversold conditions
• MACD (Moving Average Convergence Divergence). Identifies trend direction and momentum shifts
• Bollinger Bands. Measures price volatility and identifies potential reversal zones
• ADX (Average Directional Index). Gauges trend strength regardless of direction
• SMA-50 and SMA-200. Short-term and long-term trend context
• Annualised volatility. Risk measurement used in alignment weighting

These indicators are combined into a composite technical score using a deterministic formula. Each indicator contributes independently across three dimensions (momentum, trend, and volatility positioning), and the result is clamped to a standardised range from strongly bearish to strongly bullish.

This score is fully reproducible. Same market data in, same score out, every time.

Each holding also receives a technical narrative: a plain-language explanation of what the technicals are saying, key support and resistance levels, what catalysts to watch, and where the risks are. This makes institutional-grade technical data accessible without requiring you to read charts.

Step 5: Alignment Analysis

The equities identified in Step 3 are ranked and weighted based on how strongly they align with your thesis.

The primary signal is thesis alignment: how directly each company expresses your hypothesis. This is supplemented by technical context from Step 4 —a stock with strong alignment and supportive technical signals ranks higher than one where the technicals diverge. A volatility adjustment ensures that no single high-risk name disproportionately dominates the analysis.

The output is a ranked, weighted view of the equities most relevant to your thesis —capped at 10 holdings, with minimum alignment thresholds to exclude weak matches. Think of it as a structured lens on your hypothesis, not a portfolio recommendation.

Step 6: Backtesting

The constructed portfolio is backtested against SPY using up to a year of daily historical prices. This tells you how the portfolio would have performed if you'd held it —with all the caveats that implies.

The report includes:

• Total return. Raw percentage gain or loss over the period
• CAGR. Annualised compound growth rate
• Alpha. Excess return above the benchmark
• Sharpe ratio. Return per unit of risk
• Maximum drawdown. Worst peak-to-trough decline —the most painful moment
• Annualised volatility. How bumpy the ride would have been

Beyond headline metrics, the report includes contribution analysis at seven time horizons —7 days, 30 days, 60 days, 90 days, 6 months, year-to-date, and 1 year —showing which stocks drove performance and when. This tells you not just how the portfolio performed, but why.

A daily time series is generated for charting, showing the portfolio's equity curve against the benchmark.

Step 7: Report Generation

An AI agent synthesises everything —your thesis, the sentiment landscape, the portfolio, the technicals, the backtest —into a concise executive report:

• Key findings and overall confidence assessment
• Thesis validation with supporting and contradicting evidence
• Portfolio rationale with sector and allocation logic
• Performance outlook grounded in backtest results
• Risks, caveats, and data limitations
• What to monitor going forward

This is the "so what" that ties the entire analysis together —written to be useful whether you agree with the conclusion or not.

What's AI and What's Not

We think this distinction matters. Most AI tools don't tell you where the model ends and the maths begins. We do.

If you ran the same hypothesis twice with the same market data, every deterministic stage would produce identical results. The AI stages may vary -- and we think being upfront about that is more honest than pretending everything is a formula.

Limitations

We're transparent about what Thry can and can't do:

• AI can be wrong. Language models produce probabilistic outputs. The equity research may miss relevant companies or overweight irrelevant ones. Sentiment classifications are imperfect. Always verify independently.
• Sentiment isn't prediction. The fact that most voices agree with your thesis does not mean your thesis is correct. Markets frequently and enthusiastically reward contrarian positions.
• Backtests look backwards. Every performance figure uses historical data with perfect hindsight. They don't account for transaction costs, spreads, slippage, market impact, taxes, or liquidity constraints. Past performance tells you what would have happened, not what will.
• Coverage has edges. News coverage is deeper for large US equities. The voice database is English-language and weighted toward YouTube and independent analysts. Niche or emerging-market theses will have thinner data.
• This is not financial advice. Thry is an educational research platform. Nothing on this site constitutes a recommendation to buy, hold, or sell any security, and our analysis does not take into account your individual objectives, financial situation, or needs. Thry does not hold an Australian Financial Services Licence (AFSL) and is not registered as an investment adviser in any jurisdiction. You should independently verify any information before acting on it and consult a qualified financial adviser where appropriate.

For full details, see our Disclosures page.