Why it matters
Without AI, mispricing and catalogue misinterpretation lead to lost value, misaligned auction strategy, incorrect quality expectations, and rejected lots due to misread grades or misaligned broker valuations.
From design to safety
Design
AI informs design of catalogue schema
The system reads training catalogues, grade distributions, and historical price signals to define data fields, threshold rules, and validation checks for future auctions and private sales, ensuring metadata tagging and audit-friendly lineage.
- Sorting screen
- Broker valuation ledger
- Fermentation drum
- CTC roller segment
Build
Data pipelines and model validation
During build, data flows from auction catalogues, price histories, and grade records through ETL, with feature engineering for fine leaf count and made tea outturn indicators, plus validation dashboards for QA reviews.
- Rotorvane drum
- Withering trough
- Sorting screen
- CTC roller segment
Operation
Live pricing and catalogue scoring
The model ingests current catalogues, broker valuations, and lot metadata, producing predicted prices, confidence scores, and grade-consistency flags for each lot, with per-lot rationale embedded in metadata.
- Sorting screen
- Broker valuation ledger
- Fermentation drum
- Rotorvane drum
Maintenance
Model monitoring and recalibration
Regular retraining with new auction results, cross-validation checks, and drift alerts ensure predictions stay aligned with market dynamics and catalogue structure, with versioned deployments and rollback options.
- Withering trough
- CTC roller segment
- Sorting screen
- Fermentation drum
Safety
Audit trails and compliance checks
All predictions, catalogue reads, and approvals are logged with reviewer identity, timestamps, and justification, supporting certification audits and regulatory reporting, including traceable data provenance and access logs.
- Broker valuation ledger
- Sorting screen
- Fermentation drum
- Rotorvane drum
Part by part
- 1
CTC roller segment
- Watch for
- Worn roller surface causing inconsistent compression and leaf breakage.
- AI action
- Read roller wear sensors; trigger maintenance work order if wear threshold exceeded.
- 2
Rotorvane drum
- Watch for
- Uneven drum rotation or belt slippage affecting fermentation consistency.
- AI action
- Monitor motor current and belt tension; alert for calibration and service.
- 3
Withering trough
- Watch for
- Drying rate variance across beds leading to uneven wither percentage.
- AI action
- Compare moisture readings; adjust airflow and bed position recommendations.
- 4
Fermentation drum
- Watch for
- Temperature and humidity variance inside fermentation chamber.
- AI action
- Read sensors; propose run-time adjustments and pad changes to keep target fermentation profile.
- 5
Drying tunnel belt
- Watch for
- Inconsistent belt speed or heat profile affecting drying.
- AI action
- Monitor belt tachometer and heater logs; trigger maintenance if drift detected.
- 6
Sorting screen
- Watch for
- Grade misclassification due to screen wear or misalignment.
- AI action
- Detect out-of-spec leaf fractions; initiate re-sorting and quality review.
Worked example
Catching a seal leak before a charter
Illustrative scenario with sample data.
A potential seal leak in the storage bay could contaminate made tea weights if not addressed before shipment.
Signal in
estate: Rivercrest, hectares: 320, factory: 60 t/s; catalogue_id: 2026-11-01-AL; grades: B, F; broker_valuations: 42, 35; made_tea_outturn: 38.2
System out
alert: seal leak detected; work order created for bay inspection, seal replacement, and QA recheck of affected lots; revised lot valuations issued.
Human sign-off: Estate manager approves the maintenance and QA follow-up; factory manager validates the updated lot values for the auction catalogue.
How to roll it out
Phase 1
Weeks 1 to 4
Baseline data ingestion established; catalogue templates mapped to AI reading modules and validation rules.
Phase 2
Weeks 5 to 8
Model training with historical auction results; initial predictions tested against known outcomes; governance reviews completed.
Phase 3
Weeks 9 to 12
Live deployment in pilot auctions; monitoring, feedback loops, and rebalancing thresholds for precision.
Where people stay in charge
- All pricing predictions must be reviewed by a certified auction analyst before publication.
- Final bid strategy and lot-level recommendations must be approved by the estate manager with a full audit trail.
- Catalogue reads, AI outputs, and approvals must be archived per certification and regulatory requirements.
Questions owners and crews ask
What data feeds does the AI rely on for auction pricing and catalogue analysis?
The system reads auction catalogues, broker valuations by grade, historical prices, leaf quality and outturn data, and lot-specific metadata to compute predictions and confidence signals.
Who signs off on the AI-generated auction recommendations?
A qualified estate manager or senior procurement officer reviews and approves the final bid strategy and lot-level recommendations.
How does the system handle grade disputes in a catalogue?
Disputed grades trigger re-inspection flags, recalculated valuations, and a senior tasters’ panel review before final catalogue updates.
Can the tool aid in export pricing and private sales beyond auctions?
Yes, it aligns catalogue assessments with private terms, broker inputs, and international market signals to guide negotiation and risk planning.