In Development

We Gave Our Clients an AI Analyst. Nobody Used It.

Nachi Mehta·with Claude·February 2026

We've been running data stacks for five years. Fivetran, Snowflake, dbt, GoodData — the full pipeline from source systems to dashboards, managed end to end for our clients.

Last year, we attached an AI analyst to every client's data. Natural language queries, conversational interface, the whole thing. Ask questions about your data in plain English, get answers back instantly.

Nobody used it.

Not because it didn't work. It worked fine. The problem was more fundamental: business users don't want to have conversations with their data. They want dashboards that update every morning. Reports the board trusts. A metric that means the same thing in every meeting. Chat answers don't do that. They disappear the moment you close the window.

The Real Bottleneck

Here's what actually happens at our clients: someone needs a new report. They describe what they want. Our team builds a data model, wires it into the BI layer, creates the dashboard, and delivers it. Takes a few days, runs perfectly, updates automatically, persists forever.

The AI analyst couldn't do any of that. It could answer questions about existing data, but it couldn't build anything. It couldn't create a data model. It couldn't configure a metric in our BI tool. It couldn't set up a new data connector. The actual bottleneck was never “I have a question” — it was “I need a thing built.”

So We're Building Something Different

We're building an agent that doesn't answer questions — it builds things. A business user says “I need a report showing customer lifetime value by acquisition channel” and the agent:

  • Figures out what data already exists and what's missing
  • Creates the data model if one is needed
  • Wires it into the BI layer with the right metrics
  • Builds the dashboard
  • Delivers a link to a real, persistent, production report

Not a chat response. A real artifact that runs every day, updates automatically, and becomes part of how the company operates.

Why Nobody Else Is Doing This

There are over 30 companies putting AI on the data problem right now. Over a billion dollars raised in the last two years. But every one of them sits on a single layer of the stack:

  • BI-layer tools like ThoughtSpot and Omni build dashboards on models that already exist — they can't create the underlying data infrastructure
  • Query tools like WisdomAI and Seek AI answer questions about data — useful, but the answers don't persist or compound
  • Engineering tools like dbt Copilot and Hex help data engineers write code — wrong persona entirely
  • Platform tools like Snowflake Cortex and Databricks Genie only work within their own ecosystem

Nobody orchestrates across the full stack. And there's a reason for that — it's genuinely hard. You need to understand how data flows from source systems through transformations into the BI layer, when a change belongs in one layer vs. another, and how a schema change in the warehouse cascades through models into dashboards.

We've been solving these problems by hand for five years. That's not a pitch line — it's just literally what we do every day. Now we're codifying it.

Where We Are Today

Honest answer: early. We have the architecture designed, the individual pieces working (we already control the full stack for every client), and we're building the orchestration layer that connects them.

What works today:

  • The agent can see everything across a client's stack — what data sources are connected, what models exist, what metrics are available, what dashboards are running
  • It can determine where a change should happen — whether a request needs a new data model, a new BI metric, or just a new visualization on existing data
  • It can generate BI metrics and visualizations from natural language descriptions

What we're building next:

  • Cross-layer coordination — making changes across the data warehouse and BI tool in a single operation
  • Self-healing — detecting when something breaks upstream and fixing it before anyone notices
  • Approval workflows — preview what the agent will build before it ships to production

What This Means For Our Clients

The data team bottleneck is real. Even with Stackless managing the full stack, requests still go through a queue. A business user asks for a report, it takes a few days. That's fast by industry standards — most companies wait weeks — but it should be minutes.

The agent won't replace our team. It'll make them dramatically faster. And for our clients, it means the time between “I need this” and “here it is” collapses from days to minutes. The report the board needs by Friday? It's done Friday morning, not the following Wednesday.

We'll be rolling this out to clients over the coming months. If you want to see what it looks like when AI actually builds things instead of just talking about your data, we'd love to show you.

Stop fighting your data.
Start using it.

Tell us about your data challenges. We'll show you what's possible — no pressure, no pitch deck, just an honest conversation about whether we can help.

Typical response time: same business day