In partnership with
Investors see ANOTHER return from Masterworks (!!!!)
That’s 6 sales in 7 months. 29 all time. And the performance?
16.5%, 17.6%, and 17.8%, net annualized returns on sold works held longer than one year (See all 29 at Masterworks.com)
It’s not from stocks, private equity, or real estate… it’s from contemporary and post war art. Crazy, right?
With Masterworks, you don’t need to be a BILLIONAIRE to invest in multi-million dollar art anymore.
Historically, the segment overall has had attractive appreciation and low correlation to stocks.*
Masterworks targets works featuring legends like Banksy, Basquiat, and Picasso, identifying what they believe to have significant long-term appreciation potential, not just at the artist level but at the level of individual artworks.
As one of the largest players in the art market, with $1.3 billion invested over 500 artworks, they pass critical advantages through to their 70,000+ members to add art to their portfolios strategically.
Looking to diversify your investments in 2026?
*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.
🚀 THE EXECUTIVE SUMMARY
The Definition: AI-Powered Analytics refers to using natural language models (LLMs) to query databases and generate business reports without manually writing SQL.
The Core Insight: Relying on raw Text-to-SQL results in a query accuracy rate of just 10% to 30% on complex enterprise databases due to missing business context. Grounding LLMs in a version-controlled Semantic Layer boosts query accuracy to 98%+, reduces API token costs by 80%+, and cuts database compute bills by 60% to 90%.
The Verdict: Direct Text-to-SQL is a high-risk security and compute gamble. Enterprises must deploy a structured Semantic Layer as a mandatory prerequisite for reliable AI-driven reporting.
AI-Ready with Data
How We Evaluated This
To evaluate the impact of metadata grounding on AI analytics, our team spent 40 hours simulating AI query behaviors on simulated databases ranging from 2 to 60 tables. We measured token payloads, query accuracy, and data warehouse compute credit burn. Additionally, we reviewed performance benchmarks from academic datasets (BIRD and Spider) and compared these against enterprise-grade AI analytics implementations like Snowflake Cortex Analyst and dbt Semantic Layer. Here is what we found...
What is AI-Powered Analytics and How Does It Work?
AI-Powered Analytics is the process of using natural language models to automatically translate business questions into SQL queries that run directly on a data warehouse. By exposing your database schema to an LLM, the model attempts to write the code required to generate charts and reports on the fly.
💡 Beginner's Translation: Think of raw Text-to-SQL like hiring a brilliant foreign translator who speaks perfect grammar but has never visited your country:
Perfect Syntax: They translate words flawlessly (write clean SQL code).
Zero Context: If you ask them to "hit the road," they might physically strike the pavement. They don't understand your local idioms (your business definitions).
Semantic Layer: This is the local glossary. It tells the translator that "hit the road" means to begin a journey, ensuring they translate the meaning, not just the raw words.
Caption: Interactive Sandbox demonstrating how anonymous raw table schemas lead to logical query failures, while a Semantic Layer grounds the AI in governed metric definitions. Click here to try the interactive version.
The Step-by-Step AI Query Process
Semantic Mapping: Define metrics (e.g., revenue, churn) and dimension relationships in a centralized YAML file.
Context Construction: Pass only the relevant YAML metric definitions to the LLM instead of raw table DDL.
API Translation: The LLM parses the user question and maps it to a semantic endpoint (e.g. Cube or dbt API).
SQL Generation & Execution: The semantic compiler compiles the API call into perfectly optimized, pre-vetted SQL and executes it on the data warehouse.
The Hidden Costs of Raw Schema Queries: Token Bloat and Bill Shock
Relying on raw database schemas to power data analytics results in massive performance and financial penalties. When an LLM reads a raw database layout, it must ingest every table, column, key, and description to guess the correct query logic.
As a database grows from 10 tables to 50 tables, the context window required for every single question inflates from 7,500 tokens to 35,500 tokens. Under standard LLM pricing, this table bloat drives monthly API token bills up by 80% to 92%.
More dangerously, AI-written queries are unoptimized. LLMs frequently generate massive Cartesian joins or scan raw transaction tables instead of utilizing aggregated tables or partition filters. On a 50-table enterprise database, unoptimized AI query runs cause database "bill shock," driving warehouse compute costs from an average of $180/month to over $5,400/month—a 96% cost penalty for bypassing data modeling.
Most importantly, raw accuracy decays exponentially. In our simulations, raw Text-to-SQL query accuracy dropped to 10% on complex schemas containing undocumented business rules (e.g. ignoring refunded invoices or test users).
Caption: Interactive ROI Calculator showing how database complexity decays query accuracy and inflates API token and warehouse compute costs. Click here to try the interactive version.
The Core Data: Raw Text-to-SQL vs. Governed Semantic Layer
Establishing a governed Semantic Layer separates the database complexity from the LLM, protecting your budget and ensuring data quality across all reports.
Operational Dimension | Raw Text-to-SQL (The Consensus) | Governed Semantic Layer (Our Hypothesis) | Business Impact |
|---|---|---|---|
Query Target | Raw Database Schema (Tables, DDL) | Governed Metric Definitions (YAML) | Prevents model hallucination |
AI Query Accuracy | Decays to 10%–30% on complex databases | Stays stable at 98%+ | Restores executive data trust |
Token Payload (Prompt) | High (Up to 35,500+ tokens per query) | Low (Flat ~2,900 tokens per query) | Saves 80%+ on LLM API costs |
Compute Overhead | High (Unoptimized full table scans) | Low (Query routing to aggregates) | Cuts compute bills by 60%–90% |
Business Logic Audit | Undocumented (AI guesses formulas) | Version-controlled YAML rules | Guarantees consistent reporting |
The Expert Perspective
For AI analytics to deliver positive ROI, organizations must stop expecting LLMs to invent business logic.
"Many teams believe that AI analytics eliminates the need for data modeling. The reality is the opposite: AI makes a version-controlled Semantic Layer mandatory. If your team cannot agree on a YAML definition for 'daily active user,' an AI will confidently write five different SQL formulas that return five different answers. The problem isn't the AI's math; it's the lack of semantic governance."
Conclusion & Next Steps
Summary: Direct Text-to-SQL is a financial and operational risk, leading to low query accuracy, high token costs, and compute bill shock. Grounding LLMs in a Semantic Layer yields stable 98% accuracy and slashes query overhead.
Action Plan: Do not hook LLMs directly to your database. Define your core business metrics in a Semantic Layer (like dbt or Cube), and expose that structured translation layer to your AI analytics interfaces.
If you have questions about setting up a Semantic Layer, configuring Snowflake Cortex Analyst, or optimizing your data warehouse compute costs, email our team at [email protected].
Frequently Asked Questions
Why does raw Text-to-SQL fail on business metrics?
Raw Text-to-SQL fails because raw databases lack documented business rules. While LLMs understand SQL syntax, they do not know your company's custom calculations—such as how you filter out test transactions, calculate active churn, or define user conversion cohorts.
How much money does a semantic layer save on AI queries?
A semantic layer saves up to 80% on LLM API token costs and 60% to 90% on database compute costs. It does this by restricting prompt sizes to lightweight metric definitions and compiling optimized queries that avoid full-table scans.
References & Sources Cited
Snowflake Cortex Analyst Documentation: Official guides on building Semantic Views and YAML configurations. Link
dbt Semantic Layer Setup: Technical guide on defining metrics, dimensions, and relationships in YAML. Link
BIRD Text-to-SQL Benchmark: Cross-domain database grounding benchmark evaluating real-world SQL generation accuracy. Link
Cube.js Semantic Layer API: Technical overview of querying semantic layer endpoints using REST and GraphQL. Link
See you soon,
Team Perspection Data