Introducing Spotonix: The Concierge Analyst for the Agentic Era

We’re not giving everyone a dashboard. We’re giving everyone an analyst — a concierge experience to data.

---

For forty years, Business Intelligence has forced an unacceptable tradeoff: semantic precision or self-service speed — pick one. Traditional BI chose precision and locked answers behind gatekeepers. The modern data stack chose access and gave everyone dashboards nobody could use. Now, text-to-SQL promises to dissolve the tradeoff with AI. It won’t.

Today we’re introducing Spotonix — an Agentic BI Analyst, built on the Architecture of Analytical Intent. We believe there is a right way to solve BI, and it starts with understanding how expert analysts actually think.

The Problem Everyone Knows But Nobody Has Solved

The BI industry made a twenty-year bet: give every business user a dashboard, a semantic layer, and a drag-and-drop interface, and data will be “democratized.” Billions were spent on Tableau, Power BI, Looker, and the analytics engineering stack underneath.

It didn’t work. Self-service adoption is stuck below 20% in most enterprises. Data teams are drowning — analysts spend 40% of their time answering ad-hoc questions. Business users wait days for answers. When they try to self-serve, they hit confusing dashboards, conflicting metrics, and tools that require SQL they don’t know.

So they download to Excel. They get numbers. But are they the right numbers? Defined the same way? From the right source?

Self-service provides access to data, but it does not make data accessible. There’s a profound difference. Access means you can see the data. Accessible means you can understand it, trust it, and act on it. Self-service delivered the first. It never delivered the second.

The result is what we call the illusion of competence — people who look data-driven but are often citing the wrong numbers, defined differently by different people, from dashboards that weren’t designed for their question.

The False Dawn of Text-to-SQL

LLMs arrived and the industry saw salvation. “Just let people talk to their data!” Text-to-SQL became the hottest category overnight. Every vendor bolted on a chatbot. Dozens of startups launched. The promise: natural language to answers in seconds.

Three macro trends are genuinely converging — agentic AI, natural language as the de facto interface to everything, and the specific application of both to BI. The natural language interface is inevitable. But text-to-SQL is the wrong implementation of it.

Here’s why: foundation model builders — Anthropic, OpenAI, Google — are making general-purpose coding agents that are extraordinarily good at generating SQL. If your entire value proposition is translating natural language to SQL, you’re competing with foundation models on their home turf. That’s a race to the bottom.

Worse, text-to-SQL repeats the illusion of competence at a higher level. Ask “show me habitual buying customers” and the model guesses what “habitual” means. Different SQL every time. No composability — every query generated from scratch, no reuse, no patterns. No explainability — you get a number and SQL you can’t validate. The illusion of competence just got faster and more confident.

How Expert Analysts Actually Think

To understand what’s needed, we studied how expert analysts actually work. They don’t think in terms of SQL generation. They think compositionally.

When faced with a question like “Which stores are losing habitual buying customers?”, an expert analyst’s mental process involves three steps:

1. Discovery: “Have we analyzed customer behavior patterns before? How was ‘habitual buying’ defined?”
2. Pattern matching: “This is a change-over-time analysis with customer segmentation.”
3. Composition: “I can reuse the ‘Habitual Customers’ definition and combine it with store-level trending.”

Analysts don’t start from scratch. They identify what’s similar to previous work, then compose answers from existing building blocks. They might swap “Sales” for “Returns,” change “Year-over-Year” to “Month-over-Month,” or reuse a customer segment in a different context. This compositional thinking is what enables rapid analysis with consistency and explainability.

We formalized this workflow into a system.

Introducing the Architecture of Analytical Intent

Spotonix is built on a simple but powerful insight: any business question, regardless of complexity, maps to three fundamental concepts — Segmentations (the scope: who, what, when, where), Calculations (the measurements: how much, how many), and Business Analyses (complete questions that combine scope with measurements).

These concepts reference each other, forming a compositional graph — a Knowledge Graph of Analytical Intent. In this graph, AI agents can:

• Search by semantics — find “habitual customers”
• Search by structure — query all growth calculations
• Search by pattern — discover customer segmentations based on ranking
• Compose algebraically — transform analyses by swapping components
• Explain their reasoning — trace back through the composition chain

This is fundamentally different from text-to-SQL. Instead of generating low-level code from scratch every time, Spotonix discovers what’s already known, composes from reusable building blocks, validates the interpretation in business terms, and learns — capturing the reasoning for next time. The system doesn’t just answer your question — it gets smarter from the conversation.

A Concierge Experience for Every Business Question

Today, only the VIP with a dedicated analyst gets concierge-grade service. Everyone else gets self-service — the map and compass. Spotonix changes that.

When you ask Spotonix a question, it works through four phases — the same four phases an expert analyst would:

Understand. Spotonix doesn’t guess. It searches the Knowledge Graph for known definitions, surfaces what exists, and asks you to confirm or refine. “By ‘habitual buying customers,’ do you mean the definition your team established — 3+ purchases per month, $500+ quarterly spend?”

Reuse and Compose. Answers are built from existing building blocks. The “Habitual Buyers” segmentation + a period-over-period change calculation + a declining-trend filter. Not SQL generation. Analytical composition. Deterministic, reusable, and explainable.

Validate. Before executing, Spotonix presents its interpretation in business terms. “Habitual buyers” could mean frequency-based, recency-based, or value-based. The system surfaces the ambiguity — you decide. No black-box guessing.

Learn. Every conversation enriches the system. New definitions get captured. Reasoning patterns crystallize. The data language grows. The next question is faster, more precise, more confident.

Why This Can’t Be Replicated Easily

Rich Sutton’s Bitter Lesson from AI research is instructive: general methods that scale with computation beat hand-crafted rules — but only when they operate on the right representation. Text-to-SQL has no representation. It throws raw compute at an unstructured problem — translating natural language directly to SQL with no intermediate structure. The history of AI tells us this hits a wall.

The moat isn’t the LLM — that’s commoditized. It’s not the natural language interface — that’s table stakes. The moat is the representation — the Knowledge Graph that gives compute something meaningful to operate on.

Building it requires a formal ontology, a composition algebra, and an interpretation loop that mirrors how expert analysts work. You can’t prompt your way to a Knowledge Graph. Foundation model builders don’t build enterprise-specific knowledge representations. They build general-purpose engines. Spotonix provides the representation that makes those engines productive for BI.

And every conversation makes the moat deeper. Every interaction enriches the graph. Business lexicon gets crowd-sourced across the organization. Analytical patterns crystallize. Knowledge compounds. The Knowledge Graph becomes an institutional asset that appreciates with use — the representation that gets richer every time it’s used.

Institutionalize your knowledge, not your people.

Our Story

We didn’t read about BI’s problems in a blog. We lived them.

Harish Butani and Venkatesh Seetharam have built three generations of BI engines, relational databases, and data platforms. Venkatesh and Harish created Apache Atlas — the industry’s first open-source data catalog, now an Apache Foundation top-level project — and Apache Falcon for data lifecycle management at Hortonworks. Harish was founding CTO of SparklineData (acquired by Oracle in 2015), built Oracle IFS, and contributed to SAP HANA and Apache Hive.

Between us, we’ve worked at every layer of the BI stack — from query engines to semantic models to user-facing tools. We know exactly where it breaks. And we know what’s needed to fix it.

Spotonix is backed by 8VC (led by Joe Lonsdale), Bob Muglia (former CEO of Snowflake, former President of Microsoft Server & Tools Division), Keenan Rice (founding GTM leader at Looker), and Charlie Songhurst (former Head of Strategy at Microsoft).

What’s Next

We’re working with design partners across fintech, adtech, and PaaS to refine and deploy the Spotonix Analyst. If any of these ring true for your organization — your data team is drowning in ad-hoc questions, your business users are stuck in long breadlines for answers, or you’ve invested heavily in BI tools but adoption remains stubbornly low — we’d love to talk.

Every question makes your organization smarter. That’s the promise. And we’re building it.

---

Venkatesh Seetharam and Harish Butani Co-founders, Spotonix

Schedule a Conversation →