“The line does not care that every bird is different — Slicium does. We stopped programming a machine and started supervising an operator that adapts to each carcass.”
Case studies
Numbers, not narratives.
Every case study states the baseline, the measurement protocol and the result. Where a figure is still provisional, we say so.
Across studies
What the cohort measured.
4.1%
Average saleable-yield uplift on pilot deboning lines [PLACEHOLDER]
61%
Reduction in weight giveaway per fixed-weight pack
38 ms
Median foreign-material reject decision at the plant edge
99.9%
Plant-edge runtime uptime target with fail-safe line stop
Case study 01 · Poultry
Northfold Poultry: front-half deboning yield.
A 12,000-bird-per-hour line where fixed automation was leaving measurable breast meat on the frame.
Baseline & method
Two weeks of shadow mode before anything moved
Slicium observed the line and predicted every cut without acting, establishing a per-shift yield baseline reconciled against MES weights. Only then did advisory mode begin.
- Baseline: 78.3% saleable yield, high shift variance
- Measurement: MES weights reconciled per carcass batch
- Gate: advisory only after 3,000 matched predictions
- Escalation: any bone-chip risk above 1%
Result
+4.4 points of saleable yield, held across shifts
Per-carcass seam-following recovered breast and tender product that the fixed program consistently missed, while bone-chip rejects fell because the blade stopped contacting bone in the first place.
- Saleable yield 78.3% → 82.7% [PLACEHOLDER]
- Shift-to-shift variance reduced by more than half
- Bone-chip rejects down 44%
- Program rolled to three sister lines in five weeks
“Shadow mode is what won the argument. For two weeks it told us what it would have done, and it was right more often than our best program.”
Case study 02 · Pork
Meridian Pork: closing the giveaway leak.
Problem
Every pack was a little too generous
Fixed-weight retail packs were averaging well above target because the portioner had no way to predict weight from geometry — so operators biased high to avoid underweight rejects.
- Baseline giveaway: 1.8% of packed weight
- Underweight rejects the binding constraint
- No per-SKU visibility before end-of-week reporting
Result
61% less giveaway, no increase in underweights
Predictive weight from 3D geometry plus closed-loop checkweigher feedback let the portioner target the spec instead of hedging above it. Underweight rejects stayed flat.
- Giveaway 1.8% → 0.7% [PLACEHOLDER]
- Underweight reject rate unchanged
- Per-SKU giveaway visible live, not weekly
- Payback under seven months on a Line agreement
Case study 03 · Seafood
Baltic Fillet Co.: pin bones and false rejects.
Problem
A detector tuned for safety was destroying yield
Thresholds set conservatively enough to catch pin bones were rejecting large volumes of good product. Safety and yield were in direct conflict.
- Escapes unacceptable at any rate
- False-reject rate eroding margin every shift
- Manual re-inspection consuming QA hours
Result
Zero escapes with 38% fewer false rejects
Fusing X-ray with RGB and hyperspectral context let the model distinguish bone from cartilage, ice and shadow — resolving the trade-off instead of splitting the difference.
- Escapes: zero across the measured period [PLACEHOLDER]
- False rejects down 38%
- QA re-inspection hours down 60%
- Full frame-to-verdict evidence retained
Summary
Every study at a glance.
| Study | Species | Wedge workflow | Headline result | Time to result |
|---|---|---|---|---|
| Northfold Poultry | Poultry | Deboning yield | +4.4 pts saleable yield | 9 weeks |
| Meridian Pork | Pork | Portioning giveaway | −61% giveaway | 11 weeks |
| Baltic Fillet Co. | Seafood | Foreign-material detection | 0 escapes, −38% false rejects | 8 weeks |
| Cascadia Protein | Beef | Grading consistency | ±0.3 grade consistency | 10 weeks |
| Ardenne Foods | Poultry | Labour resilience | Output held at 22% vacancy | 12 weeks |
Method
How we measure, every time.
-
1
Agree the metric
One number, defined before the pilot starts, with the data source and reconciliation method written down.
-
2
Baseline in shadow
No actions, only predictions, until the comparison is statistically meaningful against your current performance.
-
3
Advance by gate
Advisory, then supervised autonomy, each unlocked by measured accuracy and twin validation — never by calendar.
-
4
Reconcile against MES
Results are reconciled against your own systems of record, not our telemetry. If they disagree, yours wins.
Voices
From the studies.
“Giveaway was the quiet leak nobody could close. Watching the portioner hold target within a couple of grams, shift after shift, changed the economics of the whole pack line.”
“What sold my team was the audit trail. Every reject links back to the frame, the model version and who approved the autonomy level. That is what a recall investigation actually needs.”
Design-partner quotes are composite and pending publication approval. [PLACEHOLDER]
Evidence
The console the studies were run from.
Saleable yield82.4%+4.1% vs baseline
Giveaway0.7%−61% this week
FM rejects120 escapes
Line uptime99.4%+2.2 pts
Line 4 · Front-half deboning1,412 /hr83.1% yieldAutonomous
Line 2 · Breast portioning2,980 /hr0.6% giveawayAutonomous
Line 7 · X-ray screening3,140 /hr4 rejectsAdvisory
Line 1 · Primal fabrication640 /hr78.9% yieldShadow
Illustrative console data from a pilot deboning line. [PLACEHOLDER]
Integrity
Why these numbers are auditable.
SOC 2 Type II (in progress)HACCP-alignedUSDA-FSISEU 853/2004GDPRISO 27001 (planned)
Assurance-grade audit trail
Immutable logs link sensor frames, model version, autonomy level, approvals, overrides and outcomes for every agent action — built for recall defence and model governance.
Fail-safe by design
Blade, robot and line stops are deterministic and independent of the cloud. Graceful degradation returns the cell to a safe state if perception confidence drops.
Data residency & on-prem
Run fully on-premise on the plant edge, in your VPC, or hybrid. Sensitive producers can keep every frame inside the facility.
More reading
Related resources.
Designing a protein autonomy pilot
How to pick the wedge line, define the metric and structure the gates so the result is undeniable.
Yield measurement without self-deception
Reconciliation methods that survive an auditor, a customer and your own finance team.
Foreign material: detection versus prevention
Why fused perception changes the safety-versus-yield trade-off on inspection lines.
Run your own case study.
Run a paid pilot on a single processing line with a defined yield, giveaway or labour success metric. Shadow mode first, autonomy only when the numbers earn it.