Most “AI investing” is just rules or hype. This article explains how Bubble uses a transparent, 11‑step quantitative engine and a chat‑first AI copilot to manage your portfolio like a pro—without taking control away from you.
AI Investing: How Does It Really Work?
“What if you could mirror what a professional portfolio manager does every single day – analysing markets, scoring assets, managing risk, rebalancing continuously – but with full control and at a fraction of the cost?”
That is the real promise people project onto “AI investing”. But behind the slogan, platforms actually do very different things:
- some are just rules‑based automation dressed up as AI;
- some try (and usually fail) to predict markets directly;
- a smaller group uses AI and quantitative methods to optimise portfolios and help humans decide.
Bubble sits in this third category.
The goal is simple:
- give everyday investors access to institutional‑grade portfolio management,
- with full transparency and user control,
- at a fixed cost (around €0–10/month) instead of percentage‑based fees on assets.
This article walks through:
- the three main types of AI used in investing;
- Bubble’s 11‑step system (what a portfolio manager actually does (or should do) every day);
- the multi‑factor scoring engine at the heart of the system;
- the 2005–2025 backtests that validate the approach;
- why Bubble’s AI is designed to educate, not control;
- where we are transparent and where we admit the limits of AI.
1. Three Types of AI in Investing
When you see “AI powered” in finance, it can mean very different things. Roughly speaking, there are three families.
1.1. Rules‑based automation (most “robo‑advisors”)
This is the most common – and the least “intelligent”. Under the hood you often find:
- questionnaires translated into IF/THEN rules;
- a handful of model portfolios (e.g. 20/80, 40/60, 60/40);
- simple rebalancing rules (“if weight deviates by more than x%, trade”).
It is useful automation, but it’s not really AI. Once your profile is set, the system:
- rarely re‑learns from new data;
- rarely adapts to regimes beyond simple volatility bands;
- almost never exposes the logic in a way you can inspect.
Verdict: helpful for basic delegation, but closer to pre‑programmed logic than to true AI.
1.2. AI forecasting / prediction (over‑sold)
The second family is what most people imagine:
- neural nets or other machine‑learning models trying to predict next week’s price moves;
- marketing that suggests the model has “found patterns” others cannot see.
The reality from academic research and practitioner experience:
- markets are noisy;
- most predictive models overfit the past and break in new regimes;
- transaction costs and slippage often wipe out any statistical edge.
There are narrow areas where forecasting helps (e.g. execution algorithms, intraday microstructure), but as a retail investor you mostly see the over‑promised version.
Verdict: interesting research, but in 2025, most “AI that predicts the market for you” is closer to snake oil than to a robust investment process.
1.3. Machine‑learning optimisation (Bubble’s camp)
The third family is more modest and more useful:
- instead of trying to predict exact prices, you use data and ML/quantitative methods to score assets, allocate risk, and adapt portfolios in a disciplined way;
- you focus on structures (diversification, risk parity, factor tilts) instead of magical single trades.
This is Bubble’s approach:
- systematic multi‑factor scoring (momentum, quality, risk‑adjusted metrics);
- risk‑parity style pocket allocation;
- clear rules for position sizing, rebalancing, and risk limits;
- all wrapped in a chat‑first interface so you can question and validate every step.
Verdict: when implemented with discipline and transparency, this kind of AI can genuinely bring institutional methods to individual investors.
2. The Bubble System: 11 Automated Steps (What a Portfolio Manager Does Every Day)
Professional portfolio managers follow a fairly standard workflow. It may be dressed in complex language, but the skeleton is simple. Bubble’s engine mirrors this end‑to‑end workflow in 11 steps.
The key idea:
These 11 steps are essentially what a human portfolio manager does – except Bubble automates them, explains them, and asks for your approval instead of acting behind your back.
Step 1 – Market intelligence (data aggregation)
What a human manager does:
- reads market summaries, watches major indices, sectors, and macro data;
- scans for news and structural shifts.
What Bubble does:
- ingests data from third‑party providers (prices, volumes, fundamentals, factor data);
- organises it by strategy pocket (e.g. global equities, bonds, thematics, etc.).
Steps 2–3 – Asset scoring and filtering (multi‑factor engine)
Human manager:
- screens a universe of assets;
- uses quantitative and qualitative criteria to rank opportunities.
Bubble:
- applies multi‑factor scoring (momentum, quality filters, risk‑adjusted metrics);
- removes assets that fail basic liquidity or quality checks;
- ranks remaining candidates inside each pocket.
Step 4 – Risk structure (risk‑parity allocation)
Human manager:
- decides how much risk to allocate to each pocket (equities, bonds, cash, etc.);
- often uses some form of risk parity or volatility targeting.
Bubble:
- estimates volatility and correlations between pockets;
- applies a risk‑parity style allocation, so that no pocket dominates portfolio risk;
- lets you see and tweak these weights via chat (“less risk in equities”, “more cash buffer”, etc.).
Step 5 – Position sizing
Human manager:
- sets min / max weights per position;
- manages concentration risk.
Bubble:
- constrains portfolios to a reasonable number of lines (e.g. ~30);
- enforces position size ranges (for example 1–10% per line, depending on strategy);
- avoids over‑concentration in a single asset or theme.
Step 6 – Execution planning (multi‑broker routing)
Human manager:
- chooses which broker or venue to use for each order;
- may split orders for better execution.
Bubble:
- prepares orders in the format required by your connected brokers (e.g. IBKR, Alpaca, Saxo);
- structures them so that they can be executed efficiently once you approve.
Steps 7–8 – Order preparation and risk checks
Human manager:
- reviews orders before sending them;
- checks risk limits, exposures, and compliance rules.
Bubble:
- runs pre‑trade checks (exposure by asset class, region, sector, etc.);
- flags potential issues (too much leverage, over‑concentration, liquidity concerns);
- presents a clear summary: “Here is what will change and why.”
Step 9 – Execution (you stay in control)
Human manager:
- sends trades to the market.
Bubble:
- generates broker‑ready orders but does not execute without your say‑so;
- you validate or adjust in the chat interface and/or inside your broker;
- execution happens via API, with you as the legal owner of the accounts.
Step 10 – Post‑trade control (audit trail)
Human manager:
- checks fills and reconciles positions;
- logs trades for compliance and reporting.
Bubble:
- verifies that executions match intended orders;
- updates portfolio state;
- maintains an audit trail of decisions and actions.
Step 11 – Continuous monitoring and adaptation
Human manager:
- monitors positions daily;
- rebalances when drift or regime changes justify it.
Bubble:
- tracks markets and portfolio metrics continuously;
- uses trigger‑based rules (e.g. drift thresholds, regime indicators) to suggest rebalancing;
- notifies you via the chat interface: “Here is why I recommend an adjustment now.”
Result: you get a full portfolio management workflow, similar to what a professional manager runs, but with:
- full visibility,
- explicit reasoning,
- and final control in your hands.
3. The Multi‑Factor Scoring Engine: The System’s Core
At the heart of Bubble lies a multi‑factor scoring engine. Instead of betting on a single signal, the engine combines several dimensions:
Momentum (typically over ~180 days)
- Empirical finance (e.g. Jegadeesh & Titman) documents a momentum effect: assets that have outperformed over the recent past tend, on average, to keep doing so for a while.
- Bubble uses a medium‑term window to avoid short‑term noise while still reacting to regime shifts.
Quality overlay
- Not all momentum is good momentum: highly leveraged or structurally weak companies can also trend up.
- Bubble applies a quality filter (profitability, balance sheet strength, stability) to avoid low‑quality “junk momentum”.
Risk‑adjusted metrics
- Raw return is not enough. Bubble looks at return per unit of risk (Sharpe‑style metrics and volatility measures) to avoid extremely volatile names that look great on charts but behave like lotteries.
Each strategy pocket can tilt differently:
- more pure momentum for aggressive pockets;
- more quality and risk adjustment for balanced / conservative pockets.
The result is a ranking of assets per pocket, from most attractive to least, given the chosen factor mix.
4. Backtesting: 17+ Years of Validation (2005–2025)
Backtesting is about one thing: “If we had applied these rules in the past, what would have happened?” It is not a guarantee of the future, but it is a necessary sanity check.
Bubble’s backtesting framework:
- uses 2005–2025 data, so it includes:
- the 2008 crisis,
- the 2020 COVID shock,
- the 2022 bear market;
- applies realistic assumptions:
- modest transaction costs,
- no look‑ahead bias,
- corrections for survivorship bias (failed companies are included);
- tests several baseline strategies:
- Equal‑weight (naïve baseline),
- Simple risk parity (basic volatility targeting),
- Optimised risk parity with adaptive correlations and factor overlays.
In Bubble’s own experiments, optimised risk‑parity‑style portfolios:
- tend to show better risk‑adjusted returns over 15–20 years;
- often experience shallower drawdowns than simple 60/40 or naïve equal‑weight mixes;
- adjust faster to regime changes (e.g. shifting volatility across asset classes).
Bubble’s philosophy:
- publish these backtests inside the portfolio simulator;
- let you explore different pockets and assumptions yourself;
- avoid cherry‑picking: emphasise drawdowns and bad periods, not just pretty curves.
5. AI That Educates, Not Controls
A critical distinction in Bubble’s design:
Bubble is a decision‑support tool, not an asset manager.
5.1. What Bubble actually is (and is not)
Bubble is not:
- a robo‑advisor in the regulatory sense (Bubble does not have custody or discretionary control over your assets);
- a bot that trades without your knowledge or consent;
- a black box that refuses to explain its own logic.
Bubble is:
- a SaaS platform that runs a quantitative process for you;
- a chat‑first interface where you can ask questions about any recommendation;
- a way to delegate the heavy lifting (data, scoring, risk checks) while staying in command.
5.2. User control: the core difference
Workflow:
- AI analyses markets and scores assets.
- AI proposes an allocation: “Given your strategy and constraints, here is what I suggest.”
- You review: you can ask the agent “why?”, request more details, or try “what if?” scenarios.
- You decide: approve as is, tweak, or reject.
- Only after your approval do orders get prepared and sent to your broker.
You also:
- own your brokerage accounts directly (IBKR, Alpaca, Saxo, others later);
- can stop using Bubble at any time: your assets stay where they are.
You pay Bubble for infrastructure and intelligence, not for giving up ownership or control.
6. Build in Public: Radical Transparency
Most financial products are built and run behind closed doors. Bubble takes a different route: build in public.
What that means in practice:
- sharing the methodology (11‑step process, factor choices, risk framework);
- publishing backtests and assumptions instead of just a single performance number;
- documenting product progress (what shipped, what broke, what changed) in public posts;
- being explicit about limits and trade‑offs (no PEA/AV today, focus on brokerage accounts, etc.).
The goal is simple:
- earn trust by showing the reasoning;
- make it possible for users to challenge the model and ask better questions;
- turn Bubble into a tool you feel comfortable interrogating, not just “consuming”.
7. Honest Limits: What AI Can and Cannot Do
AI is powerful, but not magical. Being honest about limits is part of Bubble’s design.
What AI cannot do:
- predict the future with certainty;
- eliminate risk (equities and other risky assets will still be volatile);
- guarantee outperformance every year (there will be underperforming periods);
- replace personalised human advice for your entire financial life.
What AI can do in Bubble’s context:
- enforce systematic discipline (avoid emotional, ad‑hoc trades);
- run rigorous backtests over long histories;
- monitor markets and portfolios continuously and unemotionally;
- educate you by explaining why a change is recommended;
- help you align your portfolio with a clear, evidence‑based framework.
Bubble’s explicit bias:
- long‑term investing rather than day‑trading;
- diversification across assets and regions;
- evidence‑based methods over stories and hunches;
- low, transparent, fixed pricing over %‑of‑assets fees.
8. Conclusion: AI as a Copilot, Not a Pilot
“AI investing” can mean anything from a glorified rebalancing script to a black‑box prediction engine. Bubble’s stance is narrower and clearer:
- Use AI and quantitative methods to do the hard work (data crunching, scoring, risk checks).
- Keep humans in charge of goals, constraints, and final decisions.
- Make the whole process transparent and inspectable.
- Charge a simple flat subscription instead of skimming a percentage of your wealth.
If you want a tool that:
- explains instead of hiding,
- helps you think instead of thinking for you,
- and lets you delegate the work, not the responsibility,
then a quantitative copilot like Bubble is likely closer to what you need than yet another “AI that trades for you”.
Next steps:
- experiment with Bubble’s portfolio simulator;
- explore the build‑in‑public blog to see how we evolve over time;
- join the waitlist if you want to be among the first to use Bubble with real brokerage accounts.
