Stop Reading Dashboards. Start Talking to Your Data

The CMO Who Chats With Her Data Before Coffee

Denise Persson runs a 700-person marketing organisation at Snowflake. She is personally accountable for new-business pipeline. When she spoke at SaaStr AI 2026, one detail stuck: she starts her day by talking to her data, not staring at a dashboard.

Not a metaphor. She types a question, gets an answer, asks a follow-up. The system surfaces what matters. She acts.

That is a meaningful shift in how senior operators relate to information. And it is not exclusive to companies with 700 marketers. The same behaviour is becoming accessible to a five-person team running campaigns out of a shared Notion doc.

This post breaks down what that shift actually means, why dashboards have always been the wrong tool for most decisions, and how you can start applying the same approach without a data engineering team.

---

Why Dashboards Fail Most of the Time

Dashboards were built to answer questions their creators anticipated. Someone in IT or RevOps predicted what you would want to know, built the view, and published it. That prediction is almost always wrong by the time you open the thing.

The real problem is that good decisions come from unexpected questions. You see a dip in a metric and you want to ask: is that dip happening across all channels or just one? Is it correlated with the campaign we paused last Tuesday? Did it start before or after the bank holiday? A static dashboard cannot answer those questions. It just shows the dip.

So what do most people do? They export to a spreadsheet, poke around for 45 minutes, and then act on a hunch anyway because the deadline is today.

The cost is not just wasted time. It is delayed decisions, decisions made on incomplete context, and a slow erosion of trust in the data function itself.

---

What Conversational Data Querying Actually Looks Like

The Persson model is simple in principle. You have a natural language interface connected to your data. You ask a question in plain English. The system queries the underlying tables or data warehouse and returns an answer in plain English, often with a chart or a table attached.

You follow up. You refine. You go deeper in two minutes, not two hours.

This is not magic. Under the hood it is a large language model that has been given tools to generate and execute SQL (or whatever query language your store uses), plus access to the schema so it understands what your tables mean. The interesting part is that the LLM handles the translation between your intent and the query syntax. You do not need to know SQL. You just need to know what you want to understand.

For Persson at Snowflake, this means she can hold the morning conversation about pipeline health, channel performance, and spend efficiency in the same way she might brief a sharp analyst. Except the analyst is always available, never misunderstands the question, and does not need a ticket raised.

---

The Three Things That Make This Work (and the One That Breaks It)

1. Data quality upstream

A conversational interface cannot fix bad data. If your CRM has duplicate contacts, inconsistent UTM tagging, or missing close dates, the answers you get will be confidently wrong. The first investment is not the AI layer. It is making the underlying data trustworthy.

For most UK SMBs this means auditing your CRM, your ad platform exports, and your website analytics before you do anything clever on top.

2. A connected data store

The system needs to be able to reach your data. That might be a cloud data warehouse like BigQuery, Snowflake, or Redshift. It might be a direct integration to your CRM or your marketing platform via an API. The point is that the LLM needs a live or near-live connection, not a spreadsheet you exported on Friday.

If you are running NuvenarHub for your client communication and CRM, your conversation and pipeline data is already structured and queryable. That is the kind of foundation that makes this kind of interface viable. You can find out more about how NuvenarHub stores and surfaces your customer data at /products/nuvenarhub.

3. Context about your business

The LLM needs to understand what your tables mean in your business context. A column called status in your deals table could mean twenty different things depending on your sales process. Giving the system a data dictionary, even a simple one, dramatically improves the quality of responses.

The thing that breaks it: asking questions you cannot act on

The failure mode is using conversational querying to satisfy curiosity rather than drive decisions. You end up in an infinite loop of interesting insights that change nothing. Discipline matters. Start with the decision you need to make today, then ask the question that would change your mind about it.

---

What This Means for a Small UK Marketing Team

You do not need Snowflake's data infrastructure to apply this thinking. You need three things:

• A CRM or data store that is reasonably clean
• An AI tool that can connect to it and accept natural language queries
• The habit of starting with a question, not a report

Several accessible tools now offer this. Some CRM platforms have built-in AI query layers. Standalone tools like Rows, Equals, or direct integrations between OpenAI and Google Sheets give you a version of this without a data team.

But the bigger shift is behavioural, not technical. It is the habit of asking: what do I actually need to know right now, and what is the shortest path to knowing it?

Dashboards encourage passive consumption. Conversational querying forces active intent. That is the real advantage.

---

Agents Are the Next Step, But Do Not Skip the Basics

Persson also talked about deploying agents across the marketing team at Snowflake. Agents that run tasks, not just answer questions. An agent that monitors pipeline daily and flags anomalies before anyone thinks to look. An agent that adjusts campaign pacing based on spend-to-pipeline ratios. An agent that drafts the weekly exec summary from the numbers, not from someone's memory.

This is where the technology is heading. But agents built on messy data and unclear processes fail loudly. The teams getting value from agents are the ones who already have clean data and clear decision logic. The agent just automates the decision that a human was already making consistently.

If your team is still arguing every Monday about what the numbers actually mean, an agent will not help. Fix the upstream first.

---

Practical Steps to Start This Week

You do not need a six-month data project. Here is a reasonable starting point:

Audit one data source. Pick your CRM or your main ad platform. Spend two hours checking for duplicates, missing fields, and inconsistent values. Fix the worst offenders.

Write down the five questions you ask every week. These are your most important queries. If you are asking them manually every week, they should be automated and answerable in thirty seconds.

Try a natural language query tool against your data. Most modern BI tools and some CRMs now have a chat interface. Ask it one of your five questions. See if the answer matches what you already know. If it does not, that tells you where your data quality problems are.

Make one decision from the result. Not a report. Not a slide. A decision. Channel budget, a paused campaign, a revised forecast. If the query does not lead to a decision, it was the wrong question.

---

The Bigger Picture for UK SMBs

The gap between how enterprise marketing teams use data and how most SMB teams use it has always been wide. The reason was not intelligence or effort. It was access: to tooling, to data infrastructure, to analysts who could answer questions quickly.

That access gap is closing. The same conversational interface that Persson uses at Snowflake is available in lighter forms to a ten-person agency or a regional clinic chain. The bottleneck is no longer the tooling. It is the willingness to change the habit from looking at dashboards to asking questions.

If you want to talk through how your current data and CRM setup could support this kind of workflow, book a call with the Nuvenar team and we can give you a direct assessment of where the gaps are and what is worth fixing first.

The morning conversation with your data is a better use of thirty minutes than any report someone built three months ago.