Pirouette

Humanistic communication and elegant rebalancing by Native Markets

Product VisionCopy link

Today, our wealth is fragmented across multiple strategies and accounts, each with its own method of access, monitoring, and execution. But each of us has just one set of preferences we seek to effect. Because it’s hard to align our wealth to our preferences — or even understand what we own! — we suffer inefficient allocation and personal anxiety.

What a shame! Our wealth is the sum product of our life’s work. John Locke said property is that which one “hath mixed his Labour with, and joyned it to something that is his own… no man but he can have a right to (it).” When our accumulated property is so directly joined to our very being, we should be able to preserve and grow it in a way that is natural, safe, and comforting.

Perhaps this is now possible. A few things are new today:

1. The study of finance has produced mathematical axioms for how to value and manage assets. These standard procedures are operationalized by quant firms and risk teams everyday.
2. Modern LLMs (and, crucially, even non-frontier models) excel at translating complicated concepts to local contexts, and vice versa. We can use LLMs to translate complex topics to diverse consumers.
3. Finance is increasingly programmable, whether due to the maturation of financial applications on blockchains or the increasing prevalance of APIs in traditional financial institutions. Therefore, once we generate a plan for a consumer, we can execute it.

In 2026, perhaps we can use technology to translate personal preferences into a coherent wealth strategy. And then, automatically maintain that strategy in the background of our entire life. If we could, the end result would be a wealth management experience that naturally extends our beliefs, preferences, and plans.

This product would remove the dissonance between preference and reality, and it would empower us to refocus on the parts of our lives that matter the most: creating new wealth, finding meaningful work, creating shared experiences with friends and family, and pursuing fulfilling hobbies.

Core ThesesCopy link

There are a few embedded theses that drive the vision, scope, and design of this product. These are important to understand, investigate, and debate.

1. Consumers are trying to manage exposures, not assetsCopy link

Most portfolios are presented as sets of assets: $100 of the S&P 500, $20 of NVDA, $20 in a money market fund, $10 in bitcoin. However, consumers may find it more intuitive to think of their portfolio in an exposure (or “factor” framework), i.e., $110 of exposure to US equities, $30 exposure to AI, $10 in cash.

I argue users already do this: when a retail investor buys an S&P 500 ETF, the investor appears to be seeking exposure to the US equity market, not to the 500 specific equities in the ETF.

Modeling portfolios with a factor framework is a canonical tool for finance professionals at trading firms, banks, and hedge funds to make better decisions. We can make this superior framework accessible to consumers with LLMs and a careful product experience.

The factor framework can also be a more rigorous and coherent abstraction around which to situate the user’s financial decisions. Instead of “if you execute this trade thesis, you could make this much money,” the trade off is “if you make this allocation, you’ll increase your exposure to X, Y, and Z themes by amount.” This is a more statistically responsible way to present financial decisions to users (i.e., because that is what is actually happening).

This framework doesn’t preclude a retail investor from adding single asset exposures.

2. Modern LLMs can translate complex decisions to consumers, but a complete product experience is still missingCopy link

As covered in Bloomberg, the initial value prop of wealth advisors (1:1 advice on the trade offs of different financial decisions) is being commoditized by LLMs.

But a chatbot is just a part of a complete wealth management experience:

1. The product should have access to a consumer’s comprehensive portfolio (this is really just table stakes, and a thin moat).
2. The product should help the user maximize her long term wealth, even if the user has a temporary idea for a trade or investment. For example, the product should anchor advice in a statistically robust portfolio model that helps the user understand the true risks and trade offs of their decision.
3. The product might leverage frontier AI capabilities such as fast inference, interaction models, cost-effective models (as frontier models reprice to true cost), or deterministic inference, where necessary, to deliver a better, safer, more reliable money experience.

From a business perspective, the more components that are needed, the harder the product to build and the more valuable the business will be if we succeed.

3. Wealth management can become background software, and the amount of wealth that desires this is largeCopy link

The vision for this product is that wealth management can fade into the background. After the user defines her target exposures, the product keeps the portfolio balanced without her input. Lightweight updates, such as portfolio updates that can be used as conversational threads, relieve worry, but because the user’s preferences are relatively stable, she doesn’t need to manually intervene.

A core belief is that this level of engagement is desired by users who seek more control over their finances without having to constantly think about their wealth. Low manual intervention should also help retail users build more wealth over time because most research shows that overtrading hurts retail users.

A revenue model based on subscription fees or regressive-fee AUM is uniquely aligned to this product goal: our unit economics will improve if users remain happy but don’t need to use the product extensively. Staying away from ads and volume-based transaction fees keep us away from engagement baiting or encouraging the user to over trade.

Product ExperienceCopy link

Pirouette is a wealth management product that helps a user understand, express, and maintain their desired financial posture across all of their accounts.

User/Customer: Who is this for initially?Copy link

Based on these, our ICP can have three specific characteristics:

1. Personal wealth between $1m and $5m
2. Checks market or portfolio 3-7 times per week, trades <1/week
3. >5 financial accounts of some kind (brokerage, bank, crypto wallet, 401k)

What does the product do?Copy link

• Connects all financial accounts and assets.
• Builds a unified view of the user’s wealth and translates the assets into a coherent, unified model of the user’s exposures, liquidity, and risk.
• Explains the user’s current portfolio in those same terms.
• Translates the user’s questions, preferences, and opinions into a new portfolio composition; shows the gap between current exposures and a hypothetical new portfolio.
• Recommends actions to close the gap.
• Executes approved actions where possible.
• Automatically maintains the strategy over time as asset prices change, deposits arrive, withdrawals occur, or the user’s preferences change.
• Selectively updates the user on his or her portfolio to build comfort.
• After any change in portfolio, can explain what changed and why.

Example use casesCopy link

Non-goals/boundariesCopy link

• Not primarily focused on trade execution/showing price charts.
• Not primarily focused on helping the user generate alpha. The user can express a view, and the product can describe the effect of the change, but it isn’t capable of or promising to generate excess returns.
• Not a high-touch wealth management concierge service.
• Not inherently a Hyperliquid product.

WorkstreamsCopy link

This product is not yet concrete, nor is it obvious it’s possible to build. Research is required. Most of the research does not have to do with interface design.

Below are five workstreams (four engineering, one business) and questions to investigate.

Time to preliminary answers: 3-4 weeks.

Time to high conviction: 8 weeks.

InterfaceCopy link

How will users interact with our product? This is likely a combination of charts/GUI, text, voice, and other elements.

• Fundamental UX design: how should the user review their wealth, express their preferences, and make decisions?
• What are the key product requirements that we need to prioritize in product development?
  • Starting assumption: responsiveness is a critical UI requirement that helps users feel close to and in control of their money.
• To the extent we can identify important requirements, like simple UI or responsiveness, what early tricks can we think of to improve on those characteristics
  • For example: LLMs tend to over-output (interesting to wonder why!). Dumping an LLM response into an interface probably feels overwhelming. But we could take a full response, lazily display subsets of the response, and have additional information quickly available upon request, which makes responses feel instantaneous while also making exploration feel “just-in-time” to the user.
• What’s the MECE/full lifecycle of user engagement, and what’s the best interface for the user to use at each step in the process?
  • For example: Once the user has established her target portfolio, do we want her to engage daily? Why? What’s the best way to continue providing value without annoying the user?
• While multi-factor portfolio modeling is standard for large financial institutions and quant firms (see, e.g., BlackRock Aladdin), it’s a new concept to most retail investors. How can we intuitively and immediately communicate this concept to the user?
• Data model: The underlying representation of how we store and deliver data in our backend will trickle out to our users. How should the Modeling team represent user data and the underlying portfolio; that is, what is the appropriate interface between the backend and the frontend?

ModelingCopy link

Given an arbitrary portfolio, we should be able to characterize its comprehensive exposures. This requires rigorous — but doable — math and engineering.

• Is modern multi-factor portfolio modeling sufficiently well-understood that it can be relied upon for a retail investing product with minimal discretionary risk management (is this business software-scalable)? Is the math straightforward enough that we can learn and implement the quantitative frameworks without a background in quant finance?
  • Starting point: Giuseppe Paleologo, Advanced Portfolio Management: A Quant’s Guide for Fundamental Investors
  • Importantly, we’re using these quantitative methods only to describe portfolios, not to generate alpha on which to trade. This should significantly reduce the difficulty of operationalizing these concepts.
• Is historical asset data available and economical for commercial use?
  • Starting point: Databento
• How hard is it to implement a prototype modeling engine that enforces correctness and strict risk management?
• Can we incorporate diverse investor portfolios into a multi-asset framework including savings accounts, bonds, cryptoassets, and derivatives?
• Given the introduction of a new asset, can we efficiently add its history to our multiasset covariance matrix in a robust way?
• Tax rules are asset-based, so a factor exposure model should allow us to optimize tax to generate excess returns for users. Is this true in practice? What are the limits?
  • Starting points: Nathan Sosner, Michael Gromis, and Stanley Krasner, “The Tax Benefits of Direct Indexing: Not a One-Size-Fits-All Formula” or Shomesh Chaudhuri, Terence Burnham, and Andrew W. Lo, “An Empirical Evaluation of Tax-Loss-Harvesting Alpha”
• Is there anything off-the-shelf that fits our needs and is cost-efficient?

InferenceCopy link

LLMs excel at translating complex concepts to local context, but there may be special requirements for LLMs operating in investment finance. Developing custom AI tooling could be a technical moat for Native Markets.

As a practical matter, raising capital and attracting attention as an AI company might be easier than doing the same as a fintech company, at least for now. But it’d be a waste if the AI investment doesn’t improve the product!

This workstream needs careful attention. Custom AI work could be a black hole for company resources.

• What’s the state of robo-advising? This has been a thing for almost a decade. What has emerged that works well and is showing clear PMF?
• What about modern LLMs doesn’t work for a consumer financial application? For example, how can LLMs be adapted for a more natural user experience?
  • Starting point: Interaction Models: A Scalable Approach to Human-AI Collaboration
• Might we need deterministic inference, e.g., for explainability and auditability?
  • Starting point: Defeating Nondeterminism in LLM Inference, Thinking Machines Lab.
• If we can use off-the-shelf inference, which inference products should we use? Do we need SOTA models? Why? How do wek now? What are the economics of these models, and how are the economics changing?
  • For example: Seems like latency of response is incredibly important to the UX. So if we want to use fast inference as much as possible, e.g., Cerebras-accelerated compute, that is more expensive. Research cost diff, power diff of smaller, faster models, which providers exist that can offer lower latency, etc.
• If any of the above is important to the product, what resources (talent, compute) are required to achieve our goals or serve our customers?
• Where should we be using traditional ML instead of LLMs? What data or resources do we need to build up to good ML?

ExecutionCopy link

Once a user defines their preferences, the product should constantly rebalance their portfolio to match those preferences. For example, at payday, the user’s direct deposit should be instantly split into their existing factor exposures.

• What TradFi systems and financial institutions can be accommodated, and which cannot?
• To what extent, if any, do blockchain systems (e.g., Hyperliquid, Ondo stocks) provide operational leverage for this product? Can the seamless integration of blockchain and TradFi systems be an operational advantage for us?
• What does academic literature say about optimal execution for retail traders? What is the annual cost of poor or better execution?
• Mock a prototype execution engine that enforces correctness

BusinessCopy link

Good product… good business? Not always.

• Even if everything is possible and the product is good, how do we make money? Will we make money? How much?
• What are plausible exit paths for our business, e.g., acquisition by an incumbent financial institution?
• What are the terminal applications of our product and its component? Is it just retail wealth management, or can one or many of the components grow into other business lines at some point (when?)?
• How will we acquire customers, and how much will they be worth?
• A team strength is strategic partnerships. What strategic partnerships can accelerate our business, and what moats can they provide?
• How much capital do we need to grow this business?
• What team do we need to grow this business?
• What outside expertise, i.e., consultants and advisors, would be helpful or necessary for scaling this product?
• To the extent we’re doing or using AI, what are the economics of that consumption given rapid changes in pricing and use (e.g., Cerberus inference)?
• Do we need any licenses? How would we need to design or guardrail our product to avoid the [time, money] cost of licensing?

RisksCopy link

Most startups fail. What are the risks that we need to eliminate or mitigate to win?

• We aren’t able to effectively communicate new models for wealth management, or new interfaces for managing wealth.
• There’s some incontrovertible statistical impediment to providing the product in a safe and automated way.
• The tech isn’t ready for our product vision.
• We can create a good product, but we can’t convert it to a good business.
• We’re blocked by some kind of regulation, and we don’t think we can overcome or change it.
• Good financial decisions take years to pay off, and luck may obscure an objectively poor financial decision. Even if we’re helping consumers make better financial decisions, they may not care or realize.
• Our product does well, but no one wants to purchase our equity.