Inovai / BI and data
Three reports, three different numbers.
One data model, forecasts and AI on the right numbers. Our BI is led by a Microsoft MVP.
Margin by month
48.2mrevenue, October
18.4%margin, -1.2 pp
AI commentary on the model
Why did the margin drop in October?
The purchase price of steel from one supplier rose by 6%, the sales price list stayed the same. Source: 12 purchase documents.
Problem
Every department calculates margin differently, and then AI answers wrongly.
What we’ll do
Data platform, semantic model and reporting for management.
Result
One version of the numbers for management and for AI.
What exactly we do
Reporting in Power BI and Qlik Sense that management opens and finds the answer in.
PerformanceNobody opens a report that takes a minute to load. We optimise the model, queries and data refresh.
GovernanceWho sees what, who publishes what, how changes are versioned. Rules for workspaces, certified datasets and sharing.
Reporting for managementFewer charts, a clear question and the answer on one page. A deviation leads to its cause, not to another dashboard.
Microsoft Fabric, Azure Databricks and data warehouses that can carry both reporting and AI.
Microsoft FabricLakehouse, dataflows and semantic models in one environment. Design of capacities, workspaces and administration.
Azure Databricks and data warehouseLarge volumes, machine learning and data preparation for AI. ERP, CRM, warehouse and production in one model.
Versioning and deployment for BI teamsVersioning of models and reports, deployment through development, test and production. Any change can be rolled back.
Copilot and data agents answer only as well as the model underneath them.
Copilot in Power BIQuestions about a report in natural language. We prepare the model, descriptions and synonyms so it understands your terms.
Data agents in FabricAn agent on the lakehouse and semantic model answers beyond the ready-made reports. We set up the sources, instructions and permissions.
Explaining deviationsThe report shows the change, AI traces the cause in the documents and prepares a commentary with the source. Management gets an answer, not a chart.
Machine learning on your data, with an explanation of why the model recommends exactly this.
Demand and revenue forecastingA weekly or monthly forecast with an uncertainty band, based on orders, seasonality and prices.
Recommended actionWhat to order, when to restock, where to adjust a price or schedule maintenance. A human decides.
Anomalies and deviationsA customer, line or item off trend before it shows up in a report.
How it works
From the current state through the model and platform to AI on reports.
Step 1Current stateWe review the models, reports and metrics the company can’t agree on.
Margin1 definitionRevenueowner: financeInventorysource: ERP
Step 2Semantic modelOne definition of each metric, with a source and an owner. Model and data refresh performance.Fabric
Step 3Platform and deploymentFabric, Databricks or a data warehouse, depending on volume and team. Versioning, development, test, production.Why did the margin drop in October?Steel purchase price +6%. Source: 12 documents.
Step 4AI on top of reportsCopilot in Power BI and data agents in Fabric. Descriptions, permissions, testing.BI is led by Jiří Neoral, Microsoft MVP. We test a data agent on a report the same way as a chatbot.
What a recommendation looks like in practice
Demand, stock and anomalies on one screen, with an explanation.
Weekly demand, actuals and forecast
actualsforecast with uncertainty band
Recommendations for next week
Steel plate S235JRorder 18 t, stock runs out in 9 days
HEA sectionsstock is sufficient
Anomalycustomer B: -40% against trend
Explanation: seasonality, open orders, steel prices
Do you also do Qlik Sense, or only Microsoft?
Both. Jan Koňařík builds analytics in Qlik Sense on Oracle and MSSQL. We choose the platform based on what you have, not on what we want to sell.
Is our Power BI ready for Copilot?
Copilot answers based on the names, descriptions and relationships in the semantic model, and when they are unclear, it answers wrongly. Preparing the model is usually the first step.
Is BI related to AI security?
Yes. A data agent on a report is AI with access to company figures. Permissions in the model and a test of what can be extracted from it are part of the deployment.