In the past year, as the Ask-Data project continued to be implemented, everyone's attitude towards Ask-Data became more rational and even began to be suspicious.
I believe many people have encountered situations where the demo was amazing but the implementation was awkward.
It’s okay to ask simple questions, but once you enter a real business scenario, you will encounter problems such as inconsistent calibers, unclear field semantics, complex permissions, and non-traceable answers.
The reason is that, in the final analysis, the most difficult thing about Ask-Data is business semantics. In traditional enterprises, it is difficult for even humans to explain calibers and indicators clearly. AI cannot be expected to naturally understand them.
Therefore, there are also obvious negative voices, saying that Ask-Data cannot be implemented at all.
In fact, Ask-Data can be implemented in another way.
We no longer continue to force Ask-Data into all historical data systems, but look for new scenarios where the historical baggage is lighter, the semantic boundaries are clearer, and the results are easier to verify.
There are roughly two types of new scenarios: one is a new business system built from scratch, especially an AI-native system; the other is an operational assistance scenario centered around external Internet data.
New Scenario 1: Starting with a semantically native new business system
If the biggest difficulty in traditional enterprise data environments is historical baggage, then for new scenarios that are free from historical baggage, the problem can be much smaller, such as various AI-native businesses.
AI-native scenarios refer to new business systems that assume that AI will participate in query, analysis, decision-making, and operation from the beginning of product design.
In AI-native scenarios, data and semantics can be designed from the beginning.
In a new system developed from scratch, we don’t have to wait for the data to get messy before managing it. Instead, we can clearly define it in the business modeling stage:
What are the core events? For example, registration, browsing, clicking, ordering, payment, refund, conversion, retention, and repurchase.
What are the core indicators? For example, GMV, net revenue, conversion rate, retention rate, repurchase rate, customer unit price, ROI, and LTV.
What is the default caliber of these indicators? Who is responsible for interpretation? Which characters can be seen? Does the answer need to return SQL, sources and evidence?
This is a new path more suitable for Ask-Data:Semantic nativeness.
The so-called semantic native is not to first have a bunch of tables and then add a semantic layer; rather, when the business system is born, business objects, events, indicators, permissions and interpretation capabilities are designed together.
The traditional approach is to “treat chaos first and then cure it.” The AI-native path should be "precipitating semantics while building".
The former allows large models to understand historical chaos. The latter makes business systems easier to understand by large models from the beginning.
This will greatly reduce the difficulty of implementing Ask-Data.
New Scenario 2: Starting from Internet Data and Operation Assistance
In addition to new business systems built from scratch, operational auxiliary scenarios centered on external Internet data are also a new type of scenario with relatively light historical baggage.
For example, it is easier to generate value by using Ask-Data in e-commerce operations, competitive product analysis, content operations, advertising, comment analysis, and market trend analysis.
The problem with these scenarios is usually not:
“What was the company’s true profit last month?” "How should the performance bonus be calculated this quarter?"
Instead:
“What changes have occurred in the price of hot items in this category in the last 7 days?” “What are the main areas where negative reviews of competing products are concentrated?” “Which keywords have become increasingly popular recently?” “What do the main images and selling points of similar products have in common?” “Which creatives are being used by more businesses?”
This type of problem is more about operational assistance, trend judgment and opportunity discovery, and does not necessarily require the sole caliber of financial level or audit level.
Of course, external data does not require governance at all, but the focus of governance has changed:
Is the source reliable? Is the collection legal and compliant? How to align products, stores, and brands? Is the data expired? Is there a chain of evidence for the conclusion? Can different sources verify each other?
In other words, Internet data scenarios can avoid the historical quagmire of internal data governance in some enterprises, but they cannot skip data credibility management.
It’s just that this kind of management is lighter, closer to the scene, and easier to productize.
Next generation Ask-Data
From the perspective of implementation, in addition to high-value narrow scenarios in traditional enterprises, future Ask-Data will grow more from new scenarios with lighter historical baggage.
I used to think that as long as a large model was connected to the database, the enterprise would immediately enter an era where everyone could analyze and perform data querying.
The real world is not that simple.
But that doesn't mean Ask-Data doesn't have a future. Instead, its future could be clearer:
Start with a new business, Starting from a small scene, Starting with high-frequency problems, Start with a lightweight semantic layer, Start with a traceable, explainable, and actionable business loop.
Ask-Data has not failed, it just needs to change its implementation method.