Summary of Key Points
Recently, there has been a growing demand from companies for knowledge libraries. Two technical terms that have frequently emerged are "knowledge graphs" and "ontology," with ontology gaining particular attention due to the promotion by Palantir. However, implementing ontology is challenging and costly, making it suitable only for businesses with stable operations that require advanced reasoning capabilities. Companies that blindly follow this trend may encounter pitfalls; successful implementation requires a thorough understanding of their own business processes and ongoing maintenance.
1. Why Has Ontology Suddenly Become Popular? The Industry Trend Driven by Palantir
Ontology is not a new concept, but its recent popularity is mainly due to Palantir, the so-called "miracle company":
- Despite being dubbed an outsourcing firm, Palantir was once valued at $370 billion, with a Price-to-Sales (PS) ratio ten times that of typical SaaS companies (60-70 vs 4-7 times).
- The CEO has repeatedly emphasized that ontology is the key to success, leading the industry to eagerly adopt its technical approaches, business models (platforms with high per-user prices), and organizational structures (FDE teams working directly with frontline staff).
- Essentially, the shift in companies' needs from simply searching for documents to gaining a deeper understanding of their businesses has driven this trend. What used to be achievable with AI document searches now requires AI that understands business entities, relationships, and rules, which ontology can effectively address.
2. What Exactly Is Ontology? Explained in Simple Terms
Ontology is essentially a "list of everything" along with an "instruction manual for how things are related" designed for machines to understand:
- For example, if you were in a fantasy world and asked about the available resources, someone would provide a list (monks, cultivation methods, spiritual energy) and explain the relationships (monks use spiritual energy to practice cultivation, which consumes that energy). This "list + relationship" structure is what ontology represents.
- Technically, ontology uses concepts like "classes," "properties," and "relations" to describe entities. For instance, on a shopping platform, "product" would be a class, "price" a property, and "buyer-purchase-product" a relation. Machines can use these rules to infer that people who buy electronics are likely to prefer similar products.
3. The Three Major Challenges in Implementing Ontology
Despite its potential, many companies hesitate due to significant barriers:
1. Business abstraction: Transforming complex processes (such as ordering and logistics) into clear classes and relations is difficult, especially when different departments use different terms for the same concept (e.g., "lead" could mean different things in sales, operations, etc.), leading to prolonged debates.
2. Data cleansing: Corporate data is often messy, with multiple definitions for the same item across departments. Mapping this data to new ontology rules can be costly and may require significant rework.
3. Business complexity: Changes in business processes have a cascading effect on ontology. Modifying even one relation (e.g., changing from "one customer per order" to "one order per multiple customers") can invalidate all related AI systems, data flows, and pages, requiring the recalculation of historical data.
4. High maintenance costs: Ontology requires skilled professionals who understand both business and technical aspects. Once established, changes are limited, and the knowledge base can become static over time.
4. Ontology Is Not a Panacea: It's Only Suitable for Certain Scenarios
Ontology is worthwhile only in situations where:
- The business logic is stable: Medical diagnoses, where incorrect interpretations can be fatal.
- Advanced reasoning is needed: For example, in supply chain management, understanding the impact of supplier disruptions on production and sales.
- A small, specific problem needs to be solved: In e-commerce, determining whether a filter fits a particular device model.
5. The Right Approach for Companies Using Ontology
- Don't rush into it: Use ontology only when document searches or SQL queries are insufficient.
- Start with small, well-defined projects: Verify its value by tackling a specific, manageable issue before scaling up.
- Maintain it regularly: Ontology needs constant updates to reflect changes in business processes and technology.
- Understand your business thoroughly: Ontology ultimately challenges you to truly understand your own business logic; if you can't articulate it clearly, neither can AI.
Conclusion
The next two years will see a surge in the development of corporate knowledge libraries, but ontology is not a silver bullet. To enable AI to understand business processes effectively, companies must first clarify their own logical frameworks and choose the right technology based on their needs. Blindly following Palantir's approach can lead to costly mistakes. The real key lies in continuously maintaining a deep understanding of your business operations.