Data analytics assistant
Ask your data in Arabic or English — and see how it answered.
An assistant that answers business questions from your data warehouse in Arabic or English, using only the metrics and definitions your data team has approved. Every answer comes with a chart, a short narrative and the exact query that produced it, and runs with the asking user’s own access rights.
The problem
Why reporting queues never shrink
- 01Every new question becomes a request to a small BI team, and the answer arrives days after the decision was made.
- 02Two departments report different revenue figures because each one calculates it its own way.
- 03Generic “chat with your data” tools guess at table names and joins, and their wrong answers look as confident as their right ones.
- 04Opening the warehouse to more people raises a real question: who is allowed to see which rows?
How it is built
How the system is built
The model never writes SQL against raw tables. It maps each question to a governed semantic layer and drafts a query, which a validator checks before a read-only connection runs it — with the warehouse’s own row-level security applied to the person who asked.
Layer 1
People and channels
- Web assistant in Arabic and English
- Microsoft Teams
- Embedded in existing BI portals
- Review console for the data team
Layer 2
Assistant
- Question understanding and clarification
- Text-to-SQL over the semantic layer
- Charts and narrative summaries
- Certified-answers library
Layer 3
Governance
- Semantic layer of approved metrics
- Query validator and cost limits
- Read-only, per-user credentials
- Query log and answer feedback
Layer 4
Infrastructure
- Private models on-premises or in-Kingdom cloud
- Your warehouse, with row-level security
- Result cache
- Monitoring and evaluation sets
Capabilities
Data analytics assistant
Governed semantic layer
Questions resolve to metrics and definitions your data team has approved — what counts as an active customer, net revenue or an on-time delivery. When no approved definition exists, the assistant says so instead of inventing one.
Validated, read-only text-to-SQL
Generated queries are parsed and checked against allowed tables, joins and row limits before they run, over a connection that cannot write. Expensive or unbounded queries are stopped before they reach the warehouse.
Permissions inherited from the warehouse
Queries run under the asking user’s identity, so the row-level and column-level security you already maintain applies automatically. A regional manager sees their region, and the assistant cannot widen that.
The query is always shown
Every answer carries the SQL that produced it, the metrics used and how fresh the data is. Analysts can check the working, and sceptical readers can see exactly what was counted.
Charts and narrative in both languages
Results come back as a suitable chart and a short summary in Arabic or English, with figures formatted the way your reports already present them.
Certified answers and caching
Answers your data team has reviewed are saved as certified and reused for the same question, and frequent results are cached. People get consistent figures, and the warehouse is not queried again and again for the same thing.
How it works
How a question is answered
- Question in Arabic or English
- Mapped to approved metrics
- Query drafted and validated
- Run with user’s permissions
- Chart, summary and query
Connects to
Data sources and tools
- Snowflake
- Databricks
- Google BigQuery
- Microsoft Fabric
- Oracle Database
- SQL Server and PostgreSQL
- dbt semantic models and metrics
- Power BI
- Tableau
- SAP S/4HANA and SAP BW
Where it runs
On-premises
Models and data run on hardware in your own facilities. Nothing leaves your network unless you allow it.
Private cloud in the Kingdom
Dedicated, isolated capacity in a Saudi data-centre region of a provider you select — the elasticity of cloud without a shared public AI service.
Hybrid
Sensitive workloads stay on-premises; less sensitive ones use private cloud capacity. You set the policy for what goes where.
What we measure
What we measure
Agreed with you before we build and reported against once the assistant is in use — targets we track together, not results we promise.
01
Answer accuracy
Share of answers that match a data-team-reviewed result on an agreed evaluation set, checked every month.
02
Time to answer
Time from a business question to a usable answer, compared with today’s request queue.
03
Self-service share
Share of routine data questions answered without a ticket to the BI team.
04
Governed coverage
Share of questions that map to an approved metric — the gaps show where the semantic layer needs work next.
Indicative delivery path
Indicative delivery path
- 01
2–3 weeks
Metric inventory
We pick one domain — sales, finance or operations — and agree its core metrics, definitions, owners and access rules with your data team.
- 02
5–7 weeks
Semantic layer and assistant build
We model the approved metrics, connect the warehouse read-only, and test the assistant against a set of real questions with known answers.
- 03
4 weeks
Pilot with one department
A group of business users works with the assistant every day. The data team reviews answers, certifies the good ones and closes gaps in the definitions.
- 04
Ongoing
Extend and operate
We add domains, monitor accuracy and cost, and report against the agreed measures.
Durations are indicative and depend on scope, data and integrations.
Questions
Data analytics assistant
How do you stop it giving wrong numbers?
By narrowing what it can do. It answers only from approved metrics, every query is validated before it runs, and each answer shows its query so anyone can check it. We measure accuracy on a set of questions with known answers before launch and on a schedule afterwards — no text-to-SQL system is right every time, which is why the working is always visible.
Can it show data a user is not allowed to see?
No. Queries run with the asking user’s identity against your warehouse, so its existing row-level and column-level security decides what comes back. The assistant has no privileged account of its own.
Does our data leave our environment?
Not unless you decide it should. The model runs on your premises or in private cloud capacity in the Kingdom, and only query results — never the whole warehouse — pass through it. Local hosting supports your obligations under the Personal Data Protection Law, but compliance also depends on how the data is governed, and we work alongside your compliance team on that.
Do we need a perfect data warehouse first?
You need one domain with trusted tables and agreed definitions — not a perfect warehouse. Part of the first phase is finding where definitions conflict, work that fits with the data-ownership and metadata practices in the NDMO data management standards. The governed-coverage measure then shows where to invest next.
Other solutions
All solutionsPrivate AI platform
Open-weight and commercial models served inside your boundary, behind one governed gateway.
View solutionKnowledge assistant
Cited answers from your documents and systems, in Arabic or English — limited to what each user may open.
View solutionDocument intelligence
Invoices, contracts and applications read, checked against your records and posted — people review the rest.
View solutionCustomer service agent
Arabic and English service on web chat, WhatsApp and voice — approved answers, lookups, hand-over to people.
View solution
Have a process that should run itself?
Tell us about it. We will help you work out whether AI is the right answer, what it would take, and what it should return.

