Skip to content
Neural Soft
Contact us Español (View this page in Spanish)

Use case

Natural language data queries with AI

Anyone on the team asks the company's data questions in their own words: “which customers have cut their purchases this quarter?”. The AI turns the question into a query, runs it and returns the answer.

  • The figures come from the data, not from the language model.
  • Every answer shows the query that calculated it.
  • Each person only sees the data they already have access to.
Request an AI assessment See examples
Laptop screen showing several bar and line charts.

Examples

Questions that today end up as a request to the data team

Sample data from a fictitious distributor.

Sales

Question from the sales director

“Which three customers have cut their purchases most this quarter compared with the last one?”

Answer

North Hardware
−32%
Lake Supplies
−18%
Vega Garages
−11%

Sales model · query visible

Margin

Question from the purchasing manager

“Which product families have the lowest margin this year, and how much have they sold?”

Answer

It comes with a monthly chart that can be copied into a presentation.

Lowest-margin family
Consumables (14%)
Sales this year
€182,000

Operations

Question from the warehouse manager

“How many orders went out late last week, and from which warehouse?”

Answer

Late orders
23 of 410
Warehouse with most delays
Warehouse 2 (15)

Saved as a frequent question for the team

The problem

The data is there, but getting to it is hard

Today

  • Every new question is a request to the data team or to whoever knows the dashboard.
  • The answer arrives days later, once the decision has already been made.
  • Dashboards answer the questions someone foresaw, not today's.
  • Data is exported to spreadsheets that nobody keeps up to date.

With AI

  • Questions are asked in plain language, in the chat or in the dashboard itself.
  • The answer arrives in seconds, with the query so it can be checked.
  • Recurring questions become fixed reports.
  • The data team spends its time improving the data model, not pulling lists.

How it works

From question to figure, without the model making up numbers

  1. Question

    The person types what they want to know, in their own words.

    “Sales by region in October?”

  2. Query

    The AI turns it into a query on the data model, using the agreed names and definitions.

    sales[region], month = 10

  3. Calculation

    The database or Power BI runs it with that person's permissions.

    4 regions · 1 table

  4. Human review

    Answer

    The AI explains the result and shows the query. If the question is ambiguous, it asks before answering.

    Table + chart + query

What it connects to

Built on the data you already have

If the company works with Microsoft, that is the most direct route. If not, it connects to your database just as well.

  • Power BI and Copilot
  • Microsoft Fabric
  • Azure OpenAI
  • SQL Server, PostgreSQL or MySQL
  • ERP and CRM
  • Excel and SharePoint

Down to earth

What it takes for the answers to be reliable

  • A tidy data model

    If "sale" means different things in each department, the AI will be confused too. Definitions are agreed first.

  • Permissions that already exist

    The AI queries with each person's identity. Anyone who cannot see margins in the dashboard cannot see them in the chat either.

  • Test questions

    Before opening it to the team, we check a bank of real questions with their correct answers.

Empty meeting room with a long table and a screen at the far end.

Meetings

The answer arrives during the meeting, not the week after

When a question about the figures comes up in a committee, it can be settled on the spot, with the query in full view.

  • Fewer one-off report requests to the data team.
  • Decisions made with the figure in front of you, not an estimate.
  • Frequent questions are saved and shared.

Method

We start with one area and a bank of questions

  1. Choose the area

    We pick an area with reliable data and plenty of questions (sales, purchasing or operations) and collect the questions the team actually asks.

    Bank of questions with their correct answers

  2. Prepare the data model

    We review tables, names and definitions so the AI understands them, and set up permissions.

    Documented data model

  3. Pilot and measurement

    A small group uses it for a few weeks. We measure accuracy, unanswered questions and the requests no longer sent to the data team.

    Accuracy and usage report

Related service

  • AI integration in apps and systems

    We add AI to the tools your business already uses to automate tasks, search information and connect data without having to change everything.

Related training

How many report requests does your data team get?

Tell us where your data lives (Power BI, a database, the ERP) and which questions keep coming up. We will tell you whether they can be answered reliably in plain language.

Request an AI assessment