Demonstration · Illustrative data · BI Consulting Services
JOST International × BI Consulting Services
Grand Haven, MI · Greeneville, TN · serving the US, Canada and Mexico
Prepared for Al September 2026 · Plant & Order Performance
Two plants. One lead-time promise.
Open orders against the published 3-week fifth-wheel and 5–6-week landing-gear targets, on-time delivery, scrap & rework and OEM customer mix — on one page that recalculates in milliseconds because the model is an Import star schema, not DirectQuery.1
Recalculated in — · Import model, no DirectQuery round-trips
Trend · lead time vs published target
Weeks of open orders on the books, by month
Backlog ÷ shipping rate, in weeks. Dashed line is the published target for the selected mix (3 wks fifth wheels · 5–6 wks landing gear).
Breakdown · backlog vs target by line & plant
Where the lead-time promise is at risk
Bar = current weeks of backlog. Tick = published target. Colour = status.
Signature · plant-to-OEM supply map
Shipments from two plants to OEM truck & trailer plants in three countries
Arc width = units shipped in the period. Hover a destination or a plant.
Plain-English questions, answered from the same model
Illustration of a private AI layer over the star schema. Pick a question — the answer recalculates with your filters.
Why this page loads in seconds when the source SQL is slow
Import star schema with incremental refresh — not DirectQuery
The pilot ran DirectQuery against the production SQL Server, so every click waited on the database. Here the heavy work happens once per refresh, off-hours; the report answers from an in-memory model.
01Star schema. One fact table (order lines: ordered, shipped, on-time, scrap, rework) and five dimensions — Date, Plant, Product line, Customer, Destination — instead of wide ad-hoc views.
02Incremental refresh. Only the last 3 days of order lines are re-read each night; 24 months of history stay as compressed partitions. Refresh runs in minutes, not hours.
03Manual data in the same model. Scrap tallies and OEM forecasts from Excel land in a staging table with the same keys, so they join to the fact table like any other source.
04Measured, not assumed. The timer in the header is this page recalculating — data cells across every panel after your last click.
Also from BI Consulting Services
Private AI on your own server
The "ask the data" ideas on this board are designed to run on a GPU server inside your own network — an open-weight model over your own documents, cited answers, no per-seat licence, nothing leaving the building. The short deck below explains how it works. Scroll through, or download it as a PDF.