Insights / Supply chain

Live demo · Power BI and AI

A Power BI supply chain dashboard you can ask questions

Demand, forecast accuracy and stock risk across six distribution centres, in one report. When you need a quick answer, you ask the built-in AI assistant instead of digging through pages.

By Zohal Zahir · November 2026 · 6 min read · Supply chain analytics

A familiar Monday

A big customer calls. Their order was short-shipped again. Someone checks and finds the product ran out at one warehouse two weeks ago. Nobody noticed.

Then finance asks why stock is so high when customers can't get what they ordered.

The answers exist, but they sit in ERP exports. Someone opens three spreadsheets, matches them up, and by the time there is an answer the meeting is over.

The questions a supply chain team asks every week

What is about to run out before the next delivery? Where is the forecast wrong, and in which direction? Where is cash sitting in stock that isn't moving?

Try the live report

This is the full report, running live. Click Ask AI in the top right corner and type a question, or pick one of the suggested questions. Hover over the map, the charts and the tables for detail cards.

Demo data: 30 products, 6 distribution centres in Australia and New Zealand, actuals to October 2026 and a 6-month forecast. Best viewed on a laptop or desktop screen.

Reports for the full picture

The report keeps the same layout every time, so people always know where to look.

PageThe question it answers
Demand overviewIs demand on track against forecast and last year? Which products and categories drive it? Three short insights are written from the data on every refresh.
Forecast accuracyHow accurate is the forecast, where is it biased, and which product and warehouse pairs miss the most?
Network and stockWhich distribution centre needs attention, which products will run out before new stock arrives, and how much to order.

An AI assistant for the quick questions

Dashboards are good for the full picture. They are slow for the one question someone asks in a meeting. So the report has an assistant panel built in. Ask it in plain English:

"What should we reorder now?"Products whose stock cover is shorter than the supplier lead time, with order quantities
"Where are we under-forecasting?"The regions and categories where demand is running ahead of plan
"Summarise the key issues"Demand, accuracy, the biggest forecast miss, fill rate and stock risk in one answer

Every answer comes from the same data model as the report pages, with the page filters applied. The assistant works out ratios such as forecast accuracy, fill rate and days of cover after adding up the data, so the numbers match the report at any level.

How the assistant is built

It is a Power BI custom visual. By default it runs a built-in question engine inside the report: no internet connection, no API costs, and no data leaves Power BI. For open-ended questions it can connect to your own Azure OpenAI model, which looks up the data through the report instead of guessing numbers.

How the assistant gets its numbers

An answer is only useful if it matches the report. The assistant follows five rules so that it does.

RuleWhat it means
1. Same data, same filtersThe assistant is a visual in the report. Power BI gives it the same rows as every other chart (one row per product, warehouse and month), already filtered by the page slicers.
2. Totals first, ratios afterForecast accuracy, bias, fill rate and days of cover are calculated after adding up the base numbers, with the same formulas as the report. For example, fill rate is units shipped divided by units ordered, not an average of percentages.
3. Stock is a snapshotSales and orders add up over time, but stock does not. Stock on hand and forward demand use the latest month only, and lead time is the product's own value, never a sum.
4. Same time windows"Last 3 months" means the three months up to the latest month with actuals, the same window as the "3 mo" cards. Every answer says which period it covers.
5. No guessingThe built-in engine runs a fixed calculation for each type of question. When connected to Azure OpenAI, the model can only get numbers by asking the report's data through a query tool, so it cannot invent figures.

Check it yourself

In the live report above, ask the assistant "forecast accuracy last 3 months" and compare the answer with the Forecast accuracy card at the top of the page. Then pick a category in the slicer and ask again: both numbers change together.

Built for people to use

  • Text and chart colours meet WCAG AA contrast on white.
  • Status is never shown by colour alone. Every warning has a symbol and a word, such as "At risk" or "Overstock".
  • Every chart and table has alt text, and the assistant works with a keyboard and screen readers.
  • Hover cards on the map, charts and tables explain each number in plain words.

Where this fits

This setup suits distributors, wholesalers and consumer brands that run several warehouses or sites and still plan in Excel. The same approach works for sales, inventory and operations reporting: the report keeps everything in its place, and the assistant handles the quick questions.

Want this on your own data?

Book a free 30-minute call. Bring your ERP or spreadsheet exports and the questions your team asks every week.

Book a free call

Found this useful? Share it.Tag Zohal Zahir on LinkedIn so she can say thanks.

Planners still spending Monday in Excel?

Book a free 30-minute call. We'll look at where your demand, forecast and stock data sits today, and what a report like this would take on your data.

Book a free call30 minutes. Reply within two business days