Conversational AI is moving closer to enterprise data.

Until recently, "chat with your data" often meant connecting a general-purpose language model to a separate application, retrieving information from several systems and recreating enough business logic for the answer to make sense.

Databricks and Snowflake are taking a different route. Both are bringing conversational capabilities into the environments where enterprise data is already stored, processed, governed and queried.

That changes the question considerably. It is no longer simply whether a language model can translate everyday business questions into database queries, but whether the data platform can turn an open-ended question into an answer that respects established definitions, permissions and operating constraints.

Databricks approaches that challenge from the lakehouse, where data engineering, analytics and AI increasingly operate together. Snowflake approaches it from the cloud data platform and warehouse, with a particular focus on governed structured analysis and clearly defined business metrics.

For Ardent, neither approach begins with the chat box. It begins with understanding the business questions that matter, the data needed to answer them and the controls required for those answers to be trusted.

Databricks and Snowflake are making conversation a platform capability

The most significant development is not simply that both platforms now offer conversational interfaces. It is that conversation is becoming part of the platform itself.

Databricks Genie Agents allow business users to ask questions in everyday language against carefully prepared enterprise data.Data teams configure those agents using approved data and resources managed through Unity Catalog, Databricks’ central system for organising data and controlling who can access it. They also provide instructions, example SQL queries and trusted reference material to help the agent apply the organisation’s business logic correctly.

The platform then grounds its answers in queries against the organisation's governed data, rather than relying on the model's general knowledge. Databricks also provides APIs for embedding conversations that remember earlier questions and follow-up context into other applications and workflows.

Snowflake's Cortex Analyst follows a related but distinct pattern. It uses semantic views - a governed layer that defines business metrics, relationships and calculations - to translate business questions into SQL against physical Snowflake tables.

Those views define business entities, dimensions, facts, metrics and relationships. This gives the system a structured interpretation of what terms such as revenue, customer or average order value mean within that organisation.

Both platforms are solving the same underlying problem: business language is flexible, but enterprise data logic cannot be. Where they differ is in the platform environment surrounding that conversation.

Databricks starts with the convergence of data and AI

Databricks' conversational proposition reflects the broader lakehouse model.

The platform brings data engineering, SQL analytics, machine learning, unstructured data and governance into a shared environment. Genie therefore sits within a wider data and AI ecosystem rather than operating only as a conversational route into warehouse tables.

That breadth matters when the question crosses formats. A commercial user might begin by asking what happened to sales in a particular region, then move into customer comments, campaign material or another unstructured source. An operational user might want to connect a performance metric with incident history or technical documentation.

Genie Agents can be curated for specific domains, embedded into business applications and connected to governed platform assets. The wider product direction also moves beyond answering questions, towards agents that can reason across structured and unstructured information and participate in business workflows.

The platform still needs business context. Data teams must define the trusted assets, joins, terminology and instructions that shape each agent. Databricks supports example queries, Unity Catalog functions and business definitions to help the system interpret questions more accurately.

Ardent's role in this kind of environment is to connect the conversational use case to the underlying engineering. That may mean restructuring data products, strengthening catalogue information, defining reliable SQL examples or deciding which questions belong within one Genie Agent rather than attempting a single interface for the entire organisation.

Snowflake starts with governed business meaning

Snowflake’s Cortex Analyst lets users ask questions in everyday language and turns them into database queries against data already held in Snowflake.

It relies on semantic views: governed definitions that explain what the data means to the business, including how tables relate, which metrics are approved and how calculations should be performed. This gives Cortex Analyst more context than table and column names alone, helping it interpret questions consistently and generate queries within Snowflake’s existing access controls.

For organisations with established reporting and analytical data in Snowflake, that is a significant advantage. Instead of copying data into a separate conversational system, Cortex Analyst can query it where it already lives and apply the permissions already in place.

Snowflake is also building continuous improvement into the operating model. Evaluations compare answers with trusted queries, while monitoring helps teams identify unclear definitions, missing terms and gaps in how data is connected. Suggestions can be generated from real usage, but still require human review before being added.

That matters because users will ask questions in ways the implementation team did not anticipate. Over time, those questions expose weaknesses in the definitions and relationships behind the data.

For Ardent, the value of Cortex Analyst depends on more than the initial deployment. The semantic views, business metrics and access rules need to continue reflecting how the organisation actually operates.

That means monitoring the questions users ask, identifying where answers are unclear or incomplete, and refining the data relationships and definitions behind them over time. The platform becomes more useful as those foundations improve through real use.

Where each platform is likely to fit best

Taken together, the two approaches reveal an important distinction. Databricks and Snowflake can both support conversational AI, but they tend to suit different starting points.

From Ardent’s delivery perspective, Databricks is often the stronger fit where the use case spans data engineering, AI development and a mix of structured and unstructured information. Its lakehouse model is useful when conversational tools need to draw on tables, documents, machine learning workflows and wider operational data within the same environment.

Snowflake is often the stronger fit where an organisation already relies heavily on governed warehouse data and clearly defined business metrics. Cortex Analyst is particularly well suited to questions that can be answered through trusted structured data and an established semantic layer.

The trade-off is not simply about which platform has more capability. It is about where the organisation already has mature data, how broad the conversational use case needs to be and how much complexity the business is prepared to operate. Databricks may provide greaterflexibility across wider AI use cases, while Snowflake may offer a more direct route for organisations with a strong analytical warehouse foundation.

In Ardent’s experience, the best outcome usually comes from building on the environment the organisation already trusts, rather than introducing a new platform solely for the conversational layer.

Native capability does not remove the engineering challenge

It would be easy to assume that conversational analytics is now largely a configuration exercise, but neither Genie nor Cortex Analyst can resolve business ambiguity that the organisation itself has not addressed.

If two departments calculate revenue differently, the platform still needs an agreed definition. If ownership is unclear, a catalogue entry does not make the data authoritative, and if critical data arrives late, conversational access does not make the answer current.

Native tools make the interaction easier to build, but they do not remove the need to design what sits behind it.

Ardent’s team of data engineers works backwards from the business conversation, asking:

• Which decisions should the system help users make?

• Which data and documents provide reliable evidence?

• Which definitions and access rules must remain fixed?

• How will answer quality, latency, adoption and cost be measured?

Those answers determine how the platform needs to be shaped.

In Databricks, that may mean curating domain-specific Genie Agents, improving the descriptions, ownership and business context held in Unity Catalog, establishing trusted SQL examples or connecting structured analysis with relevant unstructured material.

In Snowflake, it may mean designing semantic views, clarifying metrics and relationships, creating verified queries and establishing a process for reviewing evaluations and user behaviour.

In both cases, the quality of the conversation reflects the quality of the preparation behind it.

Scale changes what a good conversation requires

Conversational AI creates a different operational workload from conventional reporting. A dashboard runs a relatively stable set of queries, while a conversation can move through several follow-up questions, changing timeframes, filters and levels of detail as it develops.

Hundreds of simultaneous users can create a long tail of small, difficult-to-predict workloads, so the platform has to balance responsiveness with cost.

Ardent already works with the kind of high-volume environments where those trade-offs matter. In one media engagement, Ardent used Databricks to support the processing of approximately 215 million commercial records and 9 million content records per hour, while producing around 80 reports and improving data turnaround by 80%.

Conversational AI adds another demand to environments of this kind. Users expect an answer quickly, but the platform cannot afford to scan the most granular data or invoke the heaviest processing route for every question.

Good architecture involves choices: using aggregates where appropriate, isolating conversational compute, caching repeated results, limiting the context passed between turns and reserving more intensive processing for questions that justify it.

The experience may look simple to the user, but making it responsive and economical is a platform-engineering problem.

The platform is only as conversational as the business is clear

The real value of conversational AI does not come from giving users another way to query data. It comes from creating a reliable path from a business question to evidence they can understand and trust.

That requires more than a capable platform. Business definitions, access rules, data quality and the operating model around the system all need to work together.

Ardent helps organisations bring those elements together, whether the solution is built around Databricks Genie, Snowflake Cortex Analyst or a wider enterprise data architecture.

Conversational AI becomes valuable not simply when it can answer, but when the organisation understands why the answer is right.

Exploring conversational AI with Databricks, Snowflake or your wider data estate? Speak to Ardent about building the foundations for trusted production use.

Sources and further reading

Ardent,Improving Data Turnaround by 80% with Databricks; Databricks documentation onGenie Agentsandtuning Genie Agent quality; Snowflake documentation onCortex Analyst,Cortex Analyst evaluationsandadministrator monitoring.