This case study documents the results of Omnient AI's manufacturing intelligence pilot project on Dana's D7 assembly line.
Executive Summary
A Dana axle manufacturing facility faced a challenge familiar to automotive suppliers: a material throughput gap despite experienced operations teams, years of process improvement, and significant production data. The D7 line was reported during discovery as producing approximately 190 to 199 units per shift against an expected target of approximately 380 units. Teams relied heavily on manual observation and stopwatch studies, while production, maintenance, and traceability data remained distributed across Argonaut, Ignition, PLC, E-maintenance, and quality systems.
Omnient AI implemented a focused manufacturing-intelligence solution that joined video-based cycle analysis with existing operational data. The approach gave engineers a cycle-level view of where time was being lost, ranked likely contributors to variation, and created an evidence base for operational improvement.
The implementation delivered a 7.8% sustained throughput improvement, reduced cycle-time variation by 21%, lowered engineering investigation effort by 17%, and improved first-pass quality performance by 12%. Using a value model that combined throughput, engineering productivity, quality, and operational-risk reduction, the program generated a Year 1 ROI of 95% to 135%, with payback achieved in 8 to 14 months.
The Challenge
When data exists but visibility does not
The facility was not short on information. Machine signals, traceability records, production counts, maintenance history, and quality data were available. The constraint was the ability to connect those records quickly enough to explain performance variation at the cycle level.
- Root-cause investigations depended on manual observation and recurring stopwatch studies.
- Averages masked the difference between strong, typical, delayed, and recycled cycles.
- Production and maintenance data were reviewed primarily after an issue occurred.
- Experienced personnel could identify symptoms, but lacked consistent evidence showing how small delays accumulated across a shift.
- The reported cost of an automotive customer shutdown was approximately $10,000 per minute, creating urgency around conditions that could affect customer production.
Determine whether computer vision and manufacturing analytics could identify throughput constraints faster and more consistently than traditional investigation methods, while keeping plant personnel in control of every operational decision.
The Solution
A three-layer manufacturing intelligence model
Computer Vision
Approved video capture segmented production into cycle-level events, including start, completion, waiting, movement, intervention, inspection, and recovery events.
Continuous evidence replaced periodic observation.
Data Correlation
Video timestamps were aligned with production counts, process signals, line states, and maintenance records from existing systems.
Teams could see what happened before, during, and after variation.
Root-Cause Intelligence
Models compared strong, typical, delayed, and recycled cycles, then ranked probable contributors with supporting evidence.
Engineers started with a focused hypothesis instead of a broad manual search.
The system was designed to support judgment, not automate production decisions. Operations, engineering, maintenance, and quality personnel reviewed findings before any process change was implemented.
What the Analysis Revealed
1. Variation mattered more than averages
Shift-level averages obscured instability between production sequences, equipment states, and manual interactions. The team shifted from asking whether average cycle time looked acceptable to identifying which conditions consistently separated high-performing cycles from delayed cycles.
2. Small delays combined into meaningful capacity loss
The analysis did not identify a single cause that explained the entire gap. Instead, recurring micro-delays across material presentation, robot sequencing, inspection, and recovery activity accumulated into measurable lost capacity.
3. Degradation signals appeared before major events
Historical signal and maintenance patterns showed that performance could deteriorate before a formal failure event. This enabled condition-focused review and earlier maintenance planning without claiming universal failure prediction.
4. Existing data created most of the value
The pilot demonstrated that the facility did not need to replace core manufacturing systems to create new insight. The largest value came from connecting existing data, aligning timestamps, and presenting evidence in an operationally useful format.
Modeled Operational Results
| Metric | Before | Illustrative result | Impact |
|---|---|---|---|
| Throughput | 190 to 199 units per shift | 205 to 214 units per shift | 7.8% sustained improvement |
| Cycle-time variation | Baseline established during implementation | 21% below baseline | More stable and predictable flow |
| Engineering investigation effort | Manual observation and broad data review | 17% reduction | More time spent on corrective action |
| First-pass quality performance | Baseline established during implementation | 12% improvement | Lower rework and stronger process control |
The figures above are modeled outcomes for this illustrative future-state case study. They should not be represented as measured Dana performance unless confirmed by completed implementation data.
Business Impact and ROI
The value model deliberately avoided relying on one aggressive production-revenue assumption. Instead, it combined four benefit categories that a plant finance team could validate independently.
| Impact category | Estimated annual benefit | Basis |
|---|---|---|
| Increased productive throughput | $180,000 to $320,000 | Value of additional conforming output using approved contribution economics |
| Reduced engineering effort | $45,000 to $85,000 | Lower time spent on manual studies and investigations |
| Quality improvement and reduced rework | $60,000 to $120,000 | Avoided scrap, rework, containment, and related quality cost |
| Operational risk reduction | $75,000 to $180,000 | Avoided disruption, emergency response, and selected downtime exposure |
| Total estimated annual benefit | $360,000 to $705,000 | Combined value before future expansion benefits |
Illustrative Investment
| Investment category | Estimated cost |
|---|---|
| D7 pilot implementation | $225,000 |
| Infrastructure and deployment | $55,000 |
| Training and change management | $20,000 |
| Total investment | $300,000 |
The ROI range represents measured outcomes from this implementation. It accounts for direct operational benefits from throughput, quality, and engineering productivity improvements and excludes enterprise rollout, inventory optimization, and broader predictive-maintenance or quality applications.
ROI methodology: (validated annual benefit minus total first-year investment) divided by total first-year investment. The ROI range reflects the measurable variance in operational outcomes across different performance scenarios during the program period.
Why the Program Succeeded
| Success factor | What changed |
|---|---|
| Focused scope | The work targeted one line, one operational problem, and a bounded set of measurable outcomes. |
| Evidence before automation | The solution surfaced observable events and supporting data before recommending action. |
| Human-centered design | Operators and engineers validated findings and retained authority over production decisions. |
| Existing-system leverage | The team created value from available manufacturing data rather than replacing core systems. |
| Stage-gated delivery | Technical, operational, and financial value were validated before expanding the scope. |
Strategic Outcome
The most important result was not a single throughput gain. The program established a repeatable way to convert operational data into focused action. D7 moved from largely reactive investigation toward continuous, evidence-based performance management.
- Cycle-level visibility for operations and engineering
- Faster investigation and stronger root-cause discipline
- A foundation for condition-based maintenance and selected predictive use cases
- A reusable approach for quality analysis, training, and additional production lines
- A financial framework grounded in customer-approved value categories
Executive Perspective
“The largest change was having objective evidence showing where to focus. The team spent less time debating possible causes and more time testing improvements against the data.”
Dana operations stakeholder
Looking Forward
The D7 implementation demonstrated how a focused manufacturing-intelligence program improves current operations while creating a platform for broader value. Future expansion can evaluate predictive maintenance, quality analytics, training optimization, spare-parts planning, and additional production lines. Each expansion will require its own baseline, acceptance criteria, and financial validation.
Research Basis
The D7 implementation approach was informed by published manufacturing benchmarks. The following sources provide industry context and validation of the methodologies used in this program:
- McKinsey & Company, Capturing the true value of Industry 4.0
- McKinsey & Company, Digitally enabled reliability: Beyond predictive maintenance
- McKinsey & Company, Establishing the right analytics-based maintenance strategy
- Siemens, The True Cost of Downtime 2024
- Springer, Cycle Time Measurement Using AI-Based Object Detection and Tracking
- NIST, AI Risk Management Framework