Agentic Business Analytics

You no longer ask a consultant for a report: you ask the system

By Xavier Franco, CEO de Elternativa · Grupo Zemsania
· 6 min read

Written with the help of AI and reviewed by Xavier Franco

A huge wooden library card catalogue with hundreds of drawers, one of them open and lit among all the others.
In this article

For years, a business question has always followed the same path. Someone in management wants to know why the quarter’s margin is drifting. They ask their team. The team opens a request with IT or a consultant. It is prioritised, modelled, the report is built. And the answer arrives weeks later, when the question has already changed.

Conversational BI breaks that order: the question comes first, and the answer comes from the data.

What conversational BI is

It is business intelligence in which you ask your data questions in plain language, from where you already work, such as Teams or Workspace, and the answer comes from your systems, with its source. It does not replace the report or your BI tool: it complements them. The report is still there for what recurs; the conversation covers everything else.

And it is not a passing fad. Gartner predicts that by 2027, 75% of new analytics content will be contextualised for intelligent applications using generative AI, and that autonomous analytics platforms will fully manage and execute 20% of business processes.

In practice: write down the five business questions that took you longest to answer this year. They are your first test.

Why connecting a chat to the database is not enough

Here is the most common mistake. A language model is connected to the tables and is expected to answer well. And sometimes it does, but not always.

Snowflake’s own documentation on its natural-language question assistant explains it very well: generic AI solutions struggle to turn questions into SQL when they only have the database schema, because the schema does not include business knowledge: how a process is defined or how a metric is calculated. That is why they add a semantic layer with descriptions, synonyms, metric formulas and join paths between tables.

In other words: for the conversation to get it right, the system has to know what «margin» means in your company, not just where the column is.

In practice: before opening conversational BI to everyone, write the definition of your ten key indicators. Without that, every answer is an interpretation.

Data is not enough: knowledge too

There is a second layer that is often forgotten. Many answers are not only in the data, but in documents, criteria and decisions: the agreement that applies to a case, the terms agreed with a supplier, the exception approved last year. That knowledge today lives in folders and in the heads of a few people.

Good conversational BI combines both: the data from your systems and the knowledge of your department, and with every answer it says where it comes from. That is what makes it reliable and what lets someone in the business trust it without having to call anyone.

Each role, its own question

What I like most about this change is that it is not for analysts. It is for each person in their job:

  • Whoever leads asks why the margin is drifting or which line can be stopped this week.
  • Whoever coordinates wants to know which customers show signs of leaving or which complaint keeps repeating.
  • Whoever operates asks which orders are at risk of missing their deadline or what the agreement says for a specific case.

Each person sees their own, with the permissions of their role. And the time that used to go into preparing repetitive reports is left for thinking.

In practice: pick a department under a lot of data pressure, such as finance, sales or operations, and start there. It is where you notice it soonest.

Governed from the start

Opening data to conversation requires control. Three things I would not negotiate:

  1. Permissions by role. Each person only sees what they can already see today.
  2. The source of every answer. Without a source, there is no trust.
  3. Consumption with limits. AI is paid for by use: you need shared consumption for the department, with limits and alerts per user.

And one more: that the platform runs on your infrastructure and with your AI providers, so that your data does not leave home.

What changes for IT

Conversational BI does not leave IT without work: it changes its work. Until now, a good part of the data team was a report factory: requests, queues, versions and last-minute changes. With the conversation, that repetitive work goes down.

What goes up is the value of what lies underneath: the data model, the metric definitions and data quality. If margin is calculated in three different ways, the conversation will give three different answers. That is why the first conversational BI project is usually also a data governance project: putting the definitions in order and writing them down once.

And a new job appears: looking after knowledge. Which documents go in, with what permissions and who keeps them up to date. It is less visible work than building reports, but it is what makes the answers reliable.

In practice: ask your data team how many hours a week they spend on reports that are requested once and never looked at again. Those are the hours conversational BI gives back to you.

Where to start

  1. Choose a department under data pressure.
  2. Define its indicators and connect the sources it already uses.
  3. Gather its documents and criteria in a knowledge base with sources.
  4. Switch it on in parallel with its usual reports.
  5. Compare response time, reporting hours and usage, with your own data.

How we help

In Agentic Business Analytics we orchestrate your analytics AI platforms so that each role decides with today’s data, in Teams or Workspace: a digital coworker that knows the data and the knowledge of its department, answers with the source and works within the permissions of each role. It does not replace Power BI, Tableau or Qlik: it complements them.

We govern it with Datagentis Management Center: views by role, management of Coworkers, a knowledge base isolated for you and shared consumption with limits and alerts per user.

Frequently asked questions

What is conversational BI?

It is business intelligence in which you ask your data questions in plain language and the answer comes from your systems, with its source.

Does conversational BI replace Power BI or Tableau?

No. The conversation complements the report: recurring needs stay in your BI tool.

What does conversational BI need to get answers right?

A semantic layer with the definition of the metrics and the business knowledge. With only the database schema, the AI gets more wrong.

Does my data leave the company?

It doesn’t have to: the platform can run on your infrastructure, with your AI providers.