How Nivii built a conversational platform for enterprise clients, with Marvik as implementation partner and Anthropic as the reasoning layer to translate natural language into structured data queries.

Business Intelligence Is Moving Beyond Static Dashboards

Business intelligence has operated on the same assumption for decades: if analysts build the right dashboard, business users will find the answers they need.

That model works until the question changes.

Dashboards are static by design. They answer the questions their creators anticipated. Every question outside that scope requires a new chart, a new filter, a new request to the data team, or a manual export into a spreadsheet.

For medium and large B2B organizations, the gap between the question a business user has and the data infrastructure that can answer it creates real operational friction. It slows down decisions, overloads analytics teams, and limits access to KPIs that should be available in real time.

Nivii was built around a different premise: business users should be able to ask questions in plain language and get structured, accurate answers directly from their own data.

Marvik worked with Nivii to design and build the engine that makes that possible.

The Challenge: Turning Natural Language Into Correct SQL

Replacing dashboards with a conversational interface sounds simple until the system has to translate open-ended business questions into correct accurate data queries.

A dashboard works because someone already wrote the query, mapped the joins, handled edge cases, and turned the result into a visualization. In a conversational BI system, that translation has to happen dynamically, against each client’s database structure.

For B2B companies, this creates several challenges:

Business context matters

Enterprise data is rarely self-explanatory. Tables, columns, and internal definitions often require business knowledge to interpret correctly

Language is ambiguous

A question like “revenue this quarter” can mean different things depending on how each company defines revenue, time periods, customers, or reporting rules. The same question at two different companies can require structurally different queries.

Accuracy is critical

A dashboard fails visibly when its underlying query breaks. A conversational system needs to return answers that are not only technically valid, but also aligned with the way the business understands its data.

Data environments evolve

Client databases change over time, so the system needs to detect updates and remain reliable as tables, columns, and data sources evolve. These constraints shaped the way Nivii’s intelligence layer was designed and implemented.

Why Anthropic Was the Right Fit

Nivii was designed as a modular system that can connect with each client’s data environment while supporting privacy, governance, and deployment requirements. Within that architecture, Anthropic’s models serve as the reasoning layer between natural language and structured data queries.

The system needed a model capable of interpreting business intent, working with structured schema context, following precise instructions, and generating SQL that aligns with each client’s specific data model.

That made Anthropic’s models a strong fit for Nivii's capabilities.

The architecture was also designed with flexibility in mind. The model layer is abstracted from the rest of the system, allowing Nivii to evolve as Anthropic releases new models and capabilities, without rebuilding the core query generation logic.

Anthropic is the current production choice. The architecture gives Nivii the ability to evolve with the model ecosystem over time.

How Nivii Works

Nivii connects business users with company data through a conversational AI experience powered by Anthropic’s models.

Instead of relying only on predefined dashboards, users can ask questions in natural language and receive answers grounded in their own data.

Behind the experience, the system combines three core capabilities:

Data context: Nivii builds the necessary context around each client’s data so the model can interpret business questions more accurately.

Anthropic-powered reasoning: Anthropic’s models help translate user questions into structured queries and summarize the results in plain language.

Secure access and delivery: The application layer manages authentication, data access, query execution, and response formatting so users interact only with the information they are authorized to see.

How Nivii Was Built

Marvik worked with Nivii following a structured implementation process focused on making conversational BI reliable across client data environments.

The process started with understanding how client data was structured, how business concepts were represented, and where additional context was needed for the model to reason correctly.

From there, both teams worked on building the metadata and configuration layers that translate technical data structures into business-readable context.

Prompt design and testing were then refined through representative business questions, helping the system improve how it interprets intent, selects the right data context, and generates useful answers.

Finally, the system was prepared for deployment and ongoing monitoring, with attention to data pipeline integrity, query quality, and the ability to keep improving as client data and business questions evolve.

Where the System Stands

Nivii represents a practical answer to a question many BI teams face: how do you give business users direct access to structured data without either simplifying the data or training everyone to write SQL?

The solution combines conversational interaction, structured data access, metadata, access control, and LLM-based reasoning into a single business intelligence experience.

Anthropic’s models play a central role in that architecture by helping translate business questions into structured queries and plain-language insights.

For business users, that means asking questions in natural language and receiving answers grounded in their own databases.

For data teams, it reduces the number of repetitive requests that require manual SQL writing, dashboard changes, or spreadsheet exports.

For Nivii, it creates a scalable foundation for conversational BI across clients, data structures, and access patterns.

From Dashboards to Conversational Business Intelligence

Nivii shows how business intelligence can move beyond predefined dashboards toward direct dialogue with company data.

The broader lesson is clear: text-to-SQL systems do not become reliable by connecting a language model directly to a database. They require data understanding, business context, access control, prompt design, monitoring, and a model capable of reasoning across natural language and structured information.

That is where Anthropic’s models play a central role inside Nivii: as the reasoning layer that helps turn business questions into structured answers and usable insights.