Ontology Velocity: The New AI Investing Moat Hiding in Plain Sight
The next generation of AI winners may not be the companies building the biggest models. They may be the businesses learning the fastest from real customers—and turning every interaction into a better product.
For years, investors were told that data was the new oil. The companies with the most data, the argument went, would possess the strongest artificial-intelligence advantage.
That idea was directionally right, but incomplete.
Data sitting in a warehouse does not automatically create intelligence. A company can collect billions of clicks, searches, purchases or health records and still fail to turn them into a durable advantage. The real value appears when a business can repeatedly observe behavior, make an intervention, measure the outcome and use that result to improve what happens next.
That is a closed learning loop.
And the speed at which that loop improves may become one of the most important—and least understood—metrics in AI-era investing.
Call it ontology velocity: how quickly a company converts real-world activity into structured knowledge, better decisions, stronger customer outcomes and another round of higher-quality data.
“The moat is not simply the data a company owns. It is how quickly the company can turn experience into intelligence.”
After Palantir and AMD, These Three Companies Could Lead the Next AI Wave
AI Models Are Becoming Inputs. Learning Loops Are Becoming Assets.
The cost of accessing capable AI is falling. Companies can choose among proprietary models, open-weight systems and specialized models, then route different tasks to whichever option provides the best combination of performance, speed and cost.
That weakens the idea that merely having access to a powerful model is a defensible advantage. If competitors can buy access to the same intelligence, the model itself starts to resemble a commodity input.
What cannot be purchased so easily is a living system built around millions of customer relationships.
Imagine two companies using the same underlying AI model. The first feeds it generic internet data. The second connects it to years of proprietary interactions showing what users chose, what the company recommended, whether the recommendation worked and how the user responded afterward.
The second company is not merely generating an answer. It is generating an answer informed by a continuously updated map of its market.
That map is the ontology: a working representation of customers, products, behaviors, relationships and outcomes. Its velocity is the rate at which the map becomes more useful.
The Compounding Loop
A strong AI-native platform increasingly follows the same pattern:
- More users create more interactions.
- The platform observes what users want and how they behave.
- It makes recommendations, interventions or product changes.
- It measures the outcome.
- The result becomes labeled feedback.
- The system improves.
- Better outcomes increase engagement, retention and referrals.
- Those gains attract more users—and the loop begins again at a larger scale.
This is more valuable than raw data because the information contains context and attribution. The system does not just know that something happened. It begins to understand what action preceded the result and whether changing that action improves the outcome.
If the loop works, learning compounds. If one platform learns slightly faster than its competitors, its product becomes slightly better. That advantage attracts more activity, which creates more feedback, which increases the learning advantage.
Over time, a small lead can become extremely difficult to close.
Spotify: Building a Map of Taste
Spotify is often valued as a music-subscription business. But that description may miss the deeper asset: a continually evolving model of human taste.
Every skipped song, replayed track, saved album, abandoned podcast, completed audiobook and shared playlist offers a signal. Individually, these actions are minor. At enormous scale, they help Spotify understand relationships among listeners, creators, moods, formats and moments.
Its central problem is not simply delivering audio. It is solving curation: selecting the right piece of content for a particular person at a particular time.
As recommendations improve, users engage more. Greater engagement generates more feedback. That feedback improves the recommendation system, making Spotify harder to replace—not because competitors cannot license music, but because they cannot instantly reproduce the same history of observed preferences and measured reactions.
AI can also change Spotify internally. Agents that help developers test features, identify bugs or route workloads among lower-cost models can increase the rate at which the company experiments. The result is a double flywheel: the consumer product learns from listeners while the organization learns how to build and operate faster.
The investment question is therefore not only, “How many subscribers did Spotify add?” It is also, “Is Spotify increasing the speed and economic efficiency with which it converts listening behavior into better experiences?”
Duolingo: Hundreds of Small Experiments Become a Machine
Duolingo provides a different example. Its visible product is a colorful language-learning app, but underneath it is an experimentation engine.
The company can test lesson structures, notifications, streak mechanics, difficulty levels, subscription prompts and AI conversations across a large user base. Most individual improvements may appear insignificant. A fractionally better lesson sequence or slightly more effective reminder will not transform the business in a single quarter.
But retention compounds.
If many small experiments help more learners return tomorrow, those users generate additional sessions. More sessions give Duolingo more opportunities to observe where learners struggle, test interventions and personalize future lessons. Better learning and engagement can then support subscriptions, advertising and new subjects.
This is why a quarterly user-growth number can sometimes be a lagging indicator. By the time an improvement appears clearly in reported financials, the product system may have been quietly strengthening for months.
The deeper question is whether the experimentation machine is intact:
- Is the company running more high-quality tests?
- Is it learning which changes produce durable retention rather than temporary engagement?
- Is AI lowering the cost of delivering personalized instruction?
- Are product improvements expanding lifetime value without damaging the learning experience?
If the answers are yes, a temporary slowdown in a surface metric may not mean the underlying engine has stopped.
Hims & Hers: From Transactions to a Healthcare Learning Loop
Hims & Hers is the most ambitious—and the most sensitive—example of the three.
It can be viewed narrowly as an online seller of treatments. A broader interpretation sees a developing health platform connecting customer acquisition, intake, clinician access, prescriptions, fulfillment, follow-up care and ongoing engagement.
Each additional specialty can deepen the relationship. A patient who initially arrives for one need may later use the platform for weight management, dermatology, mental health, sexual health or another service. With appropriate consent, governance and clinical oversight, the platform can learn from longitudinal interactions rather than a single transaction.
AI assistance could help answer routine questions, support adherence, reduce administrative work and determine when a patient needs human clinical attention. If implemented responsibly, the same deployment can improve engagement while lowering the cost to serve.
That possibility makes healthcare a powerful closed-loop opportunity—but it also makes restraint essential. Music recommendations can be imperfect with limited consequences. Health recommendations cannot. Privacy, clinical validation, regulatory compliance, model bias, escalation protocols and patient safety are not secondary concerns. They are part of the product.
The bullish thesis is not that an online platform will suddenly become a flawless “AI doctor.” It is that a well-run healthcare system could gradually become better at matching the right patient with the right intervention while operating more efficiently.
That is a much more credible—and still potentially enormous—opportunity.
What These Three Companies Have in Common
Spotify, Duolingo and Hims operate in unrelated industries, yet their potential AI advantages share a structure:
| Company | Domain ontology | Core feedback signal | Potential economic result |
|---|---|---|---|
| Spotify | Taste and media preferences | Plays, skips, saves, completion and discovery | More engagement, retention and monetization |
| Duolingo | Learning behavior and proficiency | Answers, errors, lesson completion and return frequency | Better retention, learning outcomes and lifetime value |
| Hims & Hers | Patient needs and care journeys | Intake, engagement, treatment adherence and outcomes | Broader care, lower service costs and greater retention |
All three are attempting to move from distributing a product to optimizing an outcome.
Spotify is not only delivering songs; it is trying to predict what a listener wants next. Duolingo is not only serving lessons; it is trying to determine what keeps a learner progressing. Hims is not only facilitating a prescription; it is trying to support an ongoing care relationship.
The better each platform becomes at its respective prediction problem, the more value it can potentially deliver per customer—and the more difficult the experience becomes to replicate.
The Metric Investors Will Not Find on the Income Statement
Ontology velocity is not a reported accounting metric. That makes it useful, but also dangerous. Investors can easily turn an appealing concept into a justification for any valuation.
A disciplined analysis needs evidence. Look for signs such as:
- rising engagement or retention after product improvements;
- a growing volume of measurable customer interactions;
- faster experimentation and product-release cycles;
- declining AI inference or service costs per useful outcome;
- expanding revenue per customer without deteriorating satisfaction;
- successful entry into adjacent products using existing infrastructure;
- management discussion focused on measured outcomes, not vague AI branding;
- improving free cash flow per share over a reasonable time horizon.
Equally important are the failure signals. A supposed flywheel may be weak if the data is noisy, outcomes cannot be attributed, users can easily switch, regulations prevent the data from being used, management cannot translate insights into product improvements, or competitors have access to comparable feedback.
More activity does not always mean more intelligence. More engagement does not always mean more customer value. And faster iteration can accelerate bad decisions as easily as good ones.
The Real Question: What Is the Company Learning?
Traditional financial analysis remains essential. Revenue growth, gross margin, dilution, cash generation and valuation do not stop mattering because a company has an AI flywheel.
But financial statements mostly describe what has already happened. In periods of rapid technological change, the more revealing question may be what the business is learning now—and how quickly that learning can affect future economics.
The strongest AI companies may not be those that mention artificial intelligence most often. They may be the businesses with a valuable customer problem, enormous interaction volume, a closed feedback loop and an organization capable of acting on new information faster than anyone else.
That is the essence of ontology velocity.
The model creates an output. The company observes the result. The system learns. The customer receives more value. The business earns another opportunity to learn.
And when that loop compounds, what looks like an ordinary app today can begin to resemble the intelligent operating system for an entire category tomorrow.
“In the AI economy, the winner may not be the company with the smartest model. It may be the company that learns the fastest.”
Investor note: This article presents a conceptual framework, not a recommendation to buy or sell Spotify, Duolingo, Hims & Hers or any other security. Competitive advantages, reported metrics, regulation, execution and valuations can change. Conduct independent research and consider your time horizon and risk tolerance.
