Custom Vision System Saves Dairy Processor ₹2 Crore Annually in Quality Costs

Company Background

Punjab Dairy Products Pvt. Ltd. is a mid-sized dairy processing company based in Rajpura, Punjab — strategically located in the heart of Punjab's dairy belt. Established in 2008, the company processes approximately 1.5 lakh litres of milk daily, producing pasteurised milk, flavoured milk, dahi (curd), paneer, lassi, and milk-based sweets under the "Punjab Fresh" brand. With an annual turnover of INR 185 crore, Punjab Dairy supplies to modern trade chains, institutional buyers (hotels, restaurants, caterers), and a network of 650+ retail outlets across Punjab, Haryana, and Chandigarh.

The company operates a state-of-the-art processing plant with FSSAI, ISO 22000, and HACCP certifications. The facility includes automated pasteurisation lines, UHT processing for flavoured milk, a paneer production line with a capacity of 8 tonnes per day, and a cold chain network of 22 refrigerated vehicles.

Dairy processing is a business where quality failures are extraordinarily expensive. A single contaminated batch can trigger a product recall affecting thousands of retail outlets. A batch of paneer with inconsistent texture or colour gets rejected by modern trade quality auditors, resulting in returns, credit notes, and the risk of de-listing. The shelf life of most dairy products is measured in days, not months — which means that quality problems must be detected immediately or the product reaches consumers before the issue is identified.

The Challenge

Punjab Dairy's quality challenges centred on three product lines — paneer, flavoured milk, and dahi — where visual and physical quality parameters are critical to customer acceptance but were being inspected manually.

• Paneer quality inspection was entirely manual. The QC team visually inspected paneer blocks for colour consistency (too yellow indicates excess turmeric, too white indicates excess citric acid), surface texture (smooth versus crumbly), dimensional uniformity (block size and shape), and visible foreign particles. At a production rate of 8 tonnes per day (approximately 32,000 blocks of 250g), the 4-person QC team could inspect only a 2% sample. The remaining 98% passed uninspected. Despite sampling, defect escape rate was 4.8% — meaning approximately 1,500 blocks per day with quality issues reached the market
• Flavoured milk packaging inspection gaps. The PET bottle filling line runs at 6,000 bottles per hour. Manual inspection for fill level accuracy, cap seal integrity, label placement, and visible contamination was humanly impossible at this speed. The QC team relied on periodic checks (10 bottles every 30 minutes), which meant that a fill level drift or cap seal failure could produce thousands of defective units before being detected
• Dahi consistency variation. Dahi (curd) quality depends on culture activity, incubation temperature, and incubation time. Visual cues — surface appearance, whey separation, and gel firmness — are the primary indicators of a batch that will not meet specification. These were assessed by a single experienced operator whose judgement was subjective and varied by fatigue and shift timing
• Total quality-related losses were INR 2.4 crore annually, comprising:
  • Product returns from modern trade: INR 85 lakh (mostly paneer with colour/texture issues)
  • Product returns from institutional buyers: INR 42 lakh
  • Internal rejections and rework: INR 38 lakh
  • Consumer complaints and compensation: INR 12 lakh
  • Cost of QC team labour (4 full-time + 2 part-time inspectors): INR 28 lakh
  • Brand damage and customer attrition: estimated INR 35 lakh (based on lost accounts)
• Quality data was disconnected from the ERP. Punjab Dairy was already running SAP Business One (implemented by another partner for financials and distribution). However, QC data — inspection results, batch quality scores, defect classifications — was maintained in a standalone application that did not communicate with SAP B1. Production batch records in SAP B1 had no quality attributes, making it impossible to correlate quality outcomes with production parameters (supplier, shift, machine, operator)
• Modern trade customers were demanding higher quality standards. Two major retail chains had issued quality improvement notices, citing inconsistent paneer colour and texture as concerns. One had placed Punjab Dairy on a three-month probation with monthly quality audits. The risk of de-listing from these chains — which collectively represented INR 28 crore of annual revenue — was the immediate trigger for action

"We were inspecting 2% and hoping the remaining 98% was fine. That is not quality control — that is quality guessing. When our largest modern trade customer put us on probation, we knew that manual inspection at our production volumes was simply not viable. We needed a system that could inspect every single unit at production speed."

— Harjinder Pal Singh, Managing Director, Punjab Dairy Products

The Solution

Indivar Software Solutions proposed a two-part solution: a custom computer vision quality inspection system for the three product lines, fully integrated with their existing SAP Business One instance for end-to-end traceability and quality analytics.

Module 1: Paneer Vision Inspection System

A camera-based inspection station was installed at the paneer packaging line, positioned after the block cutting stage and before the vacuum sealing stage:

• Hardware: Two industrial-grade cameras (one top-view, one side-view) with controlled LED lighting to eliminate ambient light variation. A conveyor encoder synchronises image capture with block movement, ensuring every block is photographed regardless of conveyor speed
• AI Model: A convolutional neural network (CNN) trained on 45,000+ labelled images of paneer blocks — covering acceptable and defective samples across colour variation, surface crumbliness, dimensional non-uniformity, surface contamination, and packaging defects. The model was trained using images from Punjab Dairy's own production line over a 6-week data collection period
• Inspection at speed: The system inspects every block at a rate of 12 blocks per second — faster than the production line's maximum throughput. Each block receives a quality score (0-100) and a pass/fail classification with specific defect codes
• Automatic rejection: Blocks that fail inspection are automatically diverted to a rejection lane via a pneumatic pusher. Rejected blocks are segregated by defect type for analysis
• Continuous learning: Borderline cases (quality score between 40-60) are flagged for human review. The human decision feeds back into the model as additional training data, continuously improving accuracy

Module 2: Flavoured Milk Bottle Inspection

A high-speed inspection station was installed on the PET bottle filling line:

• Fill level verification: A camera with backlight illumination measures the fill level of every bottle to within 1mm accuracy. Under-filled or over-filled bottles are rejected automatically
• Cap seal inspection: A top-mounted camera verifies that the cap is properly seated and sealed. Tilted caps, missing caps, and damaged seals are detected and rejected
• Label inspection: A side-mounted camera verifies label presence, placement accuracy, and print quality (legibility of batch number, manufacturing date, expiry date, and MRP)
• Foreign particle detection: Transmitted light imaging detects visible foreign particles in the liquid — an inspection that is virtually impossible for human inspectors at 6,000 bottles per hour
• Throughput: The system operates at 6,500 bottles per hour — exceeding the line's maximum speed with margin. Every bottle is inspected

Module 3: Dahi Surface Analysis

A vision system was installed at the dahi incubation area to provide objective, consistent quality assessment:

• A camera captures standardised images of the dahi surface at the end of the incubation period
• The AI model analyses surface texture, whey separation patterns, and colour uniformity to predict batch quality
• Each batch receives an objective quality score, replacing the subjective assessment of a single operator
• Batches scoring below the threshold are flagged for laboratory testing before release

Module 4: SAP B1 Integration

The vision system was fully integrated with Punjab Dairy's existing SAP Business One instance:

• Batch-level quality records: Every production batch in SAP B1 is automatically enriched with quality data — inspection pass rate, defect distribution, average quality score, and sample images of detected defects
• Defect correlation analysis: SAP B1 dashboards correlate defect rates with production parameters — milk supplier, shift, machine, operator, ambient temperature, and raw material batch. This enables root cause analysis that was previously impossible
• Automatic batch hold: If the inspection defect rate for a production batch exceeds a configurable threshold (set at 5% for paneer), the batch is automatically placed on hold in SAP B1, preventing dispatch until QC manager review
• Quality trend dashboards: Real-time dashboards in SAP B1 show defect trends by hour, shift, product line, and defect type — enabling the production team to detect and respond to quality drifts in real time rather than at the end of a shift
• Customer quality reports: Modern trade customers receive automated monthly quality reports generated from SAP B1 data, showing inspection statistics, defect rates, and corrective actions — directly addressing the quality audit requirements that triggered the project

Results

Results were measured over the first 12 months of operation:

• Defect detection rate improved from 78% (manual sampling) to 99.2% (100% automated inspection) — the 21.2 percentage point improvement means that virtually no defective product reaches the market. The 0.8% escape rate is under continuous improvement through model retraining
• Quality-related losses reduced from INR 2.4 crore to INR 32 lakh annually — a net saving of INR 2.08 crore, which we round to INR 2.1 crore for reporting purposes. The remaining losses are primarily from inherent process variations that cannot be eliminated through inspection alone
• Modern trade product returns reduced by 91% — from INR 85 lakh to INR 7.5 lakh annually. The retail chain that had placed Punjab Dairy on probation not only removed the probation but cited their quality system as a benchmark for other suppliers
• Consumer complaints reduced by 82% — from an average of 35 per month to 6 per month. Most remaining complaints relate to cold chain issues during distribution, not manufacturing defects
• QC team redeployed to higher-value work — the 4 full-time inspectors who previously performed visual inspection are now focused on laboratory testing, process improvement analysis, and supplier quality audits. The automated system freed approximately INR 16 lakh annually in labour that was previously dedicated to repetitive visual inspection
• Root cause identification accelerated by 90% — the correlation of defect data with production parameters in SAP B1 enabled the quality team to identify that 65% of paneer colour defects originated from milk supplied by two specific collection centres. Corrective action at the supplier level reduced colour defects by 70%
• Packaging waste reduced by 35% — defective units are now rejected before packaging (at the vision inspection stage), whereas previously, defects were discovered after packaging (during sampling) or by the customer, resulting in wasted packaging material
• Modern trade revenue protected and grew — the INR 28 crore of revenue from the two at-risk retail chains was secured. Additionally, Punjab Dairy won a new listing with a third chain based on the strength of their quality inspection system, adding INR 8 crore in annual revenue
• ROI achieved in under 7 months — the total project investment (vision hardware, AI model development, SAP B1 integration, and training) was recovered through quality cost savings within the first seven months of operation

"The vision system sees things that no human inspector can see at production speed. But the real value is not just in catching defects — it is in the data. When SAP B1 shows me that defect rates spike on the night shift, or that milk from a particular supplier produces more colour variation, I can act on the root cause. Before this system, we were firefighting individual complaints. Now we are preventing them."

— Harjinder Pal Singh, Managing Director

Key Learnings

• 100% inspection is the only acceptable standard for perishable consumer products. Statistical sampling (2-5%) is a compromise born of human limitations, not a best practice. When technology enables affordable 100% inspection at production speed, there is no justification for sampling-based quality control on food products
• Vision system accuracy depends on training data quality. The 6-week data collection period — where we captured 45,000+ labelled images from Punjab Dairy's actual production line — was the most critical phase of the project. Generic models trained on internet images would not have achieved the same accuracy on this specific product line. The model must learn what "good paneer" and "bad paneer" look like at this factory, with this lighting, at this speed
• Integration with ERP transforms inspection data into business intelligence. A standalone vision system catches defects. An ERP-integrated vision system catches defects AND tells you why they are occurring. The correlation of defect data with production parameters — supplier, shift, machine, operator, ambient conditions — is where the transformative value lies
• Modern trade quality audits can be turned into competitive advantages. Punjab Dairy went from being on probation to being cited as a benchmark. The same quality infrastructure that was initially a defensive investment (protecting existing revenue) became an offensive weapon (winning new listings)
• Start with the highest-impact product line. We installed the paneer inspection system first because it had the highest defect rate and the highest financial impact. The success of the paneer system built confidence and internal support for extending the approach to flavoured milk and dahi. A phased approach reduces risk and generates early wins

Can Computer Vision Improve Your Quality Control?

If your food processing, dairy, or FMCG manufacturing operation relies on manual visual inspection for quality control, you are accepting defect escape rates that technology can virtually eliminate. Indivar Software Solutions builds custom computer vision systems integrated with SAP Business One for end-to-end quality management. Contact us for a confidential discussion about how AI-powered inspection can transform your quality outcomes and protect your customer relationships.