Neural network
The AI advisor
Learns the pattern in the cases and explains each recommendation afterwards by working out how much every answer counted.
What counted for one example investor
Investment horizon 18 years
Age 38 years old
A research simulation. Not a financial service
Know when to trust AI investment advice
Meet an AI investment advisor that explains itself. Describe an investor and get a recommended mix with the reasoning behind it.
- Runs entirely in your browser
- Nothing about you is stored
- No real money involved
- About two minutes to try
Researcher or reviewer? Start here instead
Age 38 · 18-year horizon · moderate risk tolerance
Recommended outcome
Growth
confidence94%
- Global equities70%
- Bonds20%
- Cash and money market0%
- Real assets10%
What drove this
- Investment horizon: 18 yearssupports it
- Age: 38 years oldweighs against it
Example investor 1 of 4
What the advisor does
Three steps from a person to a recommendation
- 1
Describe an investor
Age, investment horizon, risk tolerance and a few facts about their finances. Made up is fine.
- 2
Get a recommendation
One of five investment mixes ranging from capital preservation to aggressive growth. When the case calls for it the advisor refers you to a human adviser instead.
- 3
See why in the style you choose
Pick one of four explanation styles. Then decide whether you would trust what you were told.
Two advisors
Two ways of thinking about the same data
Both learned from the same 400 cases reviewed by a panel of financial experts. One reasons like a network and the other like a checklist.
Try the AI advisor
Scorecard
The interpretable advisor
A points table you can read in full. Every answer adds or removes points for each outcome and the highest total wins. No step is hidden.
Three rows of the real table: risk capacity
| Conservative | Balanced | Growth | |
|---|---|---|---|
| Low | -1.1 | -0.7 | -0.3 |
| Moderate | +0.6 | +0.9 | -0.5 |
| High | +0.4 | -0.2 | +0.8 |
Explanations
Four ways to ask why
Pick any of them on the advisor page. Each is computed from the advisor's own numbers.
Why
Which of your answers pushed the advice and by how much.
What would change it
The smallest change to the situation that flips the advice.
How sure
The advisor's confidence with every other outcome shown beside it.
Ask it
A conversation about the recommendation grounded in the same numbers.
Outcomes
Five mixes and the option to say no
Each bar shows how a mix splits the money across four kinds of asset.
Mix 1 of 5
Capital preservation
Capital preservation first: mostly bonds and cash, a small equity sleeve.
Most of the money sits in bonds and cash, so the value moves little. Over long periods it may barely stay ahead of inflation, which is the price of that steadiness.
15%
50%
30%
- Global equities15%
- Bonds50%
- Cash and money market30%
- Real assets5%
Why this matters
An explanation can help you judge advice or just make it sound convincing
Nobody yet knows which kinds do which. That is the question this advisor was built to answer.
Investment apps increasingly recommend what to do with your money and the recommendation comes from a model. Following it helps when the model is right. Following it costs real money when the model is wrong. Most people cannot tell which is which from the recommendation alone.
Explanations are meant to close that gap. Yet an explanation can also make wrong advice sound convincing. Which kinds of explanation help people judge rather than simply persuade is still an open question.
The study
Judge the advisor on six short cases
Read a case. See the recommendation with one kind of explanation. Say what you would do. Some recommendations are deliberately wrong and the debrief tells you which. The study is anonymous and takes about fifteen minutes with no account needed.