Data Analytics in Chicago, IL

Manufacturing / Industrial Products

A Chicago-based industrial products manufacturer (4 plants, 2,600 employees, $580M revenue) was making production decisions based on monthly Excel reports compiled by 3 analysts from data extracted from SAP, the MES, and quality management systems. Problems: reporting lag (monthly reports arriving 2-3 weeks after month-end — managers making decisions based on data that was 3-7 weeks old. By the time a quality problem appeared in the monthly report, the production line had been producing affected product for weeks), inconsistent metrics (each plant calculating OEE differently — Plant A excluding planned downtime while Plant B included it, making cross-plant comparison meaningless. No standardised metric definitions across the organisation), limited drill-down (Excel reports showing summary-level data — when managers identified a concern, they requested ad-hoc analyses from the 3 analysts, who took 1-2 weeks to query systems and compile answers. The analysts: perpetually backlogged with ad-hoc requests, unable to proactively identify insights), no predictive capability (all analysis backward-looking — identifying problems after they occurred rather than predicting them. The VP of Operations: "I always know what happened last month. I never know what's happening today or what will happen next week"), and sensor data unused (8,200 sensors across 4 plants generating data continuously — stored in local historians but never analysed beyond basic alarming. The data: potentially revealing equipment degradation, process drift, and quality correlations — but inaccessible to anyone outside the automation engineering team).

We designed and deployed a manufacturing analytics platform over 14 weeks. Data platform (weeks 1-5). Data lakehouse: Databricks on Azure — ingesting data from SAP (production orders, inventory, costs), MES (production events, machine states, labour), quality systems (inspection results, SPC data, nonconformance reports), maintenance systems (work orders, equipment history), and sensor historians (8,200 sensors via edge gateways streaming to the platform). Data model: manufacturing-specific dimensional model — standardised metric definitions agreed by all 4 plant managers (OEE calculated identically, quality metrics defined consistently, cost allocation methodology unified). The standardisation: the most politically challenging aspect — requiring facilitated agreement between plant managers who had different interpretations of the same metrics. Data quality: automated quality rules catching 340+ data issues in the first month — missing production records, sensor data gaps, and inconsistent coding. Each issue: traced to its source system and resolved, improving data quality for all downstream analytics. Operational dashboards (weeks 4-9). Real-time production dashboard: replacing the monthly Excel report with live visibility — OEE by machine, line, and plant (updated every 5 minutes from MES data), production output vs. plan (real-time units produced vs. target, with projected end-of-shift completion), quality rates (real-time defect rates with drill-down to defect type, machine, and material batch), downtime analysis (current and recent downtime events with duration, reason code, and impact), and energy consumption (real-time energy usage by production area with cost calculation). Cross-plant comparison: standardised metrics enabling meaningful comparison — plant managers seeing their performance alongside other plants on identical calculations. The comparison: immediately revealing performance gaps that were invisible in the old reporting. Drill-down: every metric clickable — plant → line → machine → shift → individual production event. Managers: answering their own questions instantly rather than requesting analyst ad-hoc work. Mobile: dashboards accessible on tablets — plant managers and supervisors reviewing data on the production floor. Advanced analytics (weeks 7-12). Quality prediction: machine learning model correlating production variables (machine parameters, material batch, operator, environmental conditions) with quality outcomes — predicting quality risk before production rather than detecting defects after. The model: identifying 6 previously unknown quality correlations — including a relationship between ambient humidity and a specific adhesive bond defect that had caused recurring quality holds. Demand forecasting: AI-powered demand forecasting replacing the sales team's spreadsheet-based forecasting — analysing historical demand patterns, seasonal trends, customer ordering behaviour, and economic indicators to produce SKU-level demand forecasts with confidence intervals. Forecast accuracy: 91% at the monthly level (vs. 74% for the previous manual forecasts). Energy optimisation: analytics identifying energy waste patterns — equipment running during non-production periods ($180K/year waste identified), HVAC not optimised for production heat generation ($120K/year), and compressed air leaks ($95K/year). Total energy waste identified: $395K/year — with specific recommendations for correction. Self-service analytics (weeks 9-14). Analyst empowerment: the 3 analysts retrained from Excel report builders to self-service analytics users — using Databricks SQL notebooks and Power BI to conduct their own analyses. The analysts: now proactively identifying insights rather than reactively compiling reports. Knowledge transfer: analysts trained on the data model, query patterns, and analytical techniques — capable of creating new dashboards and analyses independently.

The analytics platform launched across all 4 plants within 16 weeks. Decision speed: operational decision-making: from weeks (waiting for monthly reports) to minutes (real-time dashboards). Ad-hoc analysis turnaround: from 1-2 weeks (analyst queue) to same-day (self-service or analyst-driven on the platform). The acceleration: enabling proactive management rather than reactive response to problems that were weeks old. OEE improvement: cross-plant comparison revealing performance gaps — Plant 3 operating at 72% OEE vs. Plant 1 at 84% on the same product. Investigation: Plant 3 changeover procedures taking 40% longer than Plant 1. Best practice sharing: Plant 1's changeover procedures documented and implemented at Plant 3 — OEE improving to 80% within 2 months. Similar cross-plant improvements across all 4 plants: average OEE improving from 76% to 83% (7-point improvement). Revenue impact: 7-point OEE improvement = $8.4M/year in additional production capacity (at existing prices). Quality improvement: quality prediction model preventing defects — the humidity-adhesive correlation alone eliminating a recurring defect that had caused $280K/year in scrap and rework. Total quality cost reduction from predictive analytics: $1.8M/year across all 4 plants. Demand forecast accuracy: forecast accuracy improving from 74% to 91% — reducing both overproduction (excess inventory) and underproduction (stockouts). Inventory reduction: $4.2M (14% lower average inventory from better forecasting). Stockout reduction: 62% fewer stockout events. Energy savings: $395K/year in identified energy waste — $280K addressed immediately (equipment scheduling changes, compressed air repairs), remainder planned for capital projects. Analyst productivity: analysts producing 5x more analytical output — from reactive report compilation to proactive insight generation. 3 analysts: now supporting the same analytical needs that previously required 3 analysts plus frequent outsourced analysis projects. Financial impact: OEE improvement (additional production capacity): $8.4M/year. Quality cost reduction: $1.8M/year. Inventory reduction (one-time working capital release): $4.2M. Energy savings: $280K/year (immediate corrections). Demand forecasting (reduced stockouts and overproduction): $960K/year. Total Year 1 impact: $11.4M recurring + $4.2M one-time. Analytics platform development: $620,000. Annual platform operations (Databricks, Azure, support): $180,000. First-year ROI: 1,938%.