Count users by role
“How many users do we have in each role?”
The assistant finds where user roles are stored, groups users by role and counts each group. Four groups, so the answer comes back with totals and a pie chart.
Cognitive AI: ask a question, get the right view.
One of our clients had an admin area with a page for every table. Admins could browse one table at a time, but couldn’t ask a question that needed several. Now they ask in plain English, and the answer comes back as a table, a chart or a few totals.
See it answer
Questions in, the right view out
Two questions from the project, replayed with the same figures. Pick one to follow how it gets answered.
1 Question
2 How it answers
60–70 tables3 Result
ChartThere are 200 users across four roles.
Normal user150
Read-only user25
Admin20
Super admin5
The assistant picks the view that fits the answer.
- 01
The problem
The admin area had a page for every table, somewhere between 60 and 70 of them. Admins could browse and filter one table at a time. A question that needed two tables meant exporting lists into a spreadsheet, or waiting for someone to write a query.
- 02
Our approach
Let admins type the question the way they’d ask a colleague. We wrote the assistant a plain-words guide to the tables, their columns and how they connect, so it can work out where to look.
- 03
The solution
An AI assistant with a few tools to look up tables, fetch records and count them. It only reads data, through the same data service the application already used, and it fixes its own query when a field name is wrong.
Adaptive views
An interface that fits the answer
Each answer arrives as the right piece of screen. The way the assistant asks the data decides the view, and one answer can show more than one, so admins never wade through raw data.
- Find matching recordsA table with the relevant columns
- Compare groups, such as user rolesA bar or pie chart showing the breakdown
- See how activity changes over timeA chart showing the trend
- Get a count or a few key numbersClear totals with short labels
How it works
From question to answer in five steps
- 1
Understand the question
Work out what the admin is asking about: which records, which conditions and dates, and whether they want a list, a count or a comparison.
- 2
Find the right tables
Search a plain-words guide to the data and pick the few tables and columns that can answer it, along with how they connect.
- 3
Build the query
Choose the columns and filters, and what to count or group. If a field name is wrong, the tool says so and suggests the right one, and the assistant tries again.
- 4
Fetch the results
Run the query through the application’s existing data service. Nothing is copied and nothing is changed.
- 5
Show the right view
The kind of query decides the view: records become a table, groups become a chart, activity over time becomes a trend, with the key numbers on top.
In the product
Two examples from the project
Look across related records
“How many organizations do we have, how many are active, and how many created a report in the last 30 days?”
This one needs two tables. The assistant counts organizations, checks which are active, then looks at their reports from the last 30 days. Here that gives 24 organizations, 18 active, and 11 that created a report.
