Data to Prepare Before Using AI in the Food Industry: Raw Materials, Recipes, Lots, Inspections, and Shipments

Target Readers: Executives, site managers, plant managers, quality assurance managers, and production control staff at Japanese food manufacturers and food processing plants operating in Thailand and across ASEAN. This article is written for those who are interested in “AI adoption” or “DX” but are unsure where to start, or those whose organizations have begun evaluating implementation but feel that on-site data is not yet in order.

“Implementing AI will give you a competitive edge.” “DX will dramatically boost productivity.” Such phrases fill industry magazines and seminars. Yet when you step onto the factory floor, a very different reality comes into view. Raw material receipt records are kept in paper ledgers, recipe changes exist only in the heads of veteran workers, lot traceability relies on handwritten notes combining production dates and product names, and quality inspection data is entered into scattered Excel sheets that cannot be aggregated. This is the situation at more than a few Japanese-owned food companies in Thailand and across the ASEAN region.

AI and machine learning are powerful tools. However, if the data fed into AI is not in order, neither accurate predictions nor anomaly detection will emerge. “Garbage in, garbage out” — the iron rule of the data world — applies equally to food manufacturing. Before implementing AI, bringing the on-site “data foundation” into order is the single most critical step that determines success or failure.

This article breaks down the types of data and key management points that food factories and food processing facilities in Thailand should prepare before leveraging AI and advanced analytics tools. It covers five domains — raw materials, recipes, lots, inspections, and shipments — and draws on TOMAS TECH’s hands-on field experience to explain how visualizing quality, temperature, lots, and yield can reduce food loss and risk, and how to approach investment decisions on the plant floor.

Why Data Readiness Is Now the Top Priority for Food Factories

The Thai business environment in 2026 is no longer one of straightforward growth. The World Bank has issued a cautious outlook for Thailand’s economic expansion, and the OECD has flagged ongoing risks from external uncertainty and persistently high energy and logistics costs. The food industry occupies a particularly distinctive position within this landscape. Raw materials — agricultural produce, seafood, dairy products, and more — are highly susceptible to climate change impacts, with prices that fluctuate sharply. Finished products have best-before and use-by dates, and unsold goods generate disposal costs. Hygiene and safety regulations must comply not only with Japanese headquarters standards but also with Thai authority requirements such as FDA Thailand.

In this environment, running a business that relies solely on revenue growth has become increasingly difficult. To achieve cost reduction, waste elimination, quality stability, and stronger traceability, it is essential that on-site “data” flows in a way that connects directly to management decision-making. AI is a tool that accelerates this flow, but if the data that should flow does not exist, AI cannot function.

Furthermore, the BOI (Thailand Board of Investment) is actively promoting incentives for investments in automation, AI, data analytics, and enterprise management IT. To take full advantage of these incentives, it is necessary — from the investment planning stage — to clearly define what data will be prepared, how it will be managed, and what AI or analytics tools will be used for what purpose. When companies attempt to apply for subsidies or tax incentives retroactively after system deployment, there are frequent cases where the application fails because the documentary evidence is internally inconsistent.

Domain 1: Raw Material Data — A Three-Way Integration of Receiving, Inventory, and Cost

In food manufacturing, raw materials represent a concentration of “risk before it enters the process.” For all inputs — agricultural products, livestock products, seasonings, food additives, packaging materials — the following information must be managed electronically as a minimum prerequisite.

• Supplier name and supplier code (distinguishing domestic vs. imported)
• Raw material name, raw material code, and lot number (or supplier’s lot number)
• Receipt date, quantity, unit price, and storage location
• Best-before date, use-by date, and storage conditions (temperature, humidity)
• Quality inspection results at receipt (sensory, physicochemical, microbiological, etc.)
• Usage records (which manufacturing lot each item was consumed into)

When this information is managed in paper ledgers or Excel spreadsheets, tracing the complete journey from receipt through use and disposal becomes extremely difficult. In Thai food factories in particular, challenges tend to compound: the same raw material is used across multiple product lines; imported materials and domestically sourced materials coexist; and staff literacy and data-entry skill levels vary — all of which contribute to declining raw material data quality.

When attempting to apply AI or predictive analytics to raw material procurement — for example, optimizing order quantities based on demand forecasting or predicting waste — a minimum of 12 to 24 months of digitized and structured raw material receipt, consumption, and disposal data is required. Without this, procurement optimization models are either impossible to build or too unreliable to trust.

Domain 2: Recipe Data — Managing Change History and Cost

One area frequently overlooked on food manufacturing floors is “change management for recipes (manufacturing formulas).” A portion of a raw material is swapped to reduce costs; a blending ratio is adjusted to improve quality; a recipe is revised for allergen compliance — these kinds of changes often exist only on shop-floor work sheets or in workers’ heads and are never reflected in any system.

The minimum elements that must be established as recipe data are as follows.

• Product code, product name, and version number
• List of raw materials used (raw material code, input quantity, unit)
• Manufacturing process and conditions (heating temperature, time, mixing sequence, etc.)
• Change history (change date, change details, person who made the change, approver)
• Standard cost (cost calculation linked to the recipe version)
• Allergen information and nutritional composition (for compliance with export regulations and labeling laws)

When recipes are accurately managed, comparing actual production results against the recipe automatically surfaces deviations such as “yield is below standard,” “raw material consumption exceeds the theoretical value,” or “actual cost exceeds standard cost.” This variance analysis is the foundation on which AI-driven anomaly detection and improvement recommendations are built.

Recipe version control is also indispensable for tracing the impact of recipe changes on product quality. Only a factory with well-organized data can transform a vague feeling like “returns started increasing around that time” into an objective verification — “a comparison of inspection data before and after the recipe change.”

Domain 3: Lot Data — Traceability and Results Linkage Mechanisms

Lot management in a food factory is directly linked to “containing the damage” and “identifying the root cause” when a quality problem occurs. If a foreign object contamination, an incorrect best-before date, or allergen contamination is discovered after shipment, the inability to immediately identify which lots reached which customers or stores means that a voluntary recall will demand enormous amounts of time and cost.

The minimum required data for lot management is as follows.

• Manufacturing lot number (internal management number)
• Manufacturing date and time, manufacturing line, and operator
• Raw material lot numbers used (bidirectional traceability between raw materials and finished products)
• In-process records (temperature logs, CCP records, etc.)
• Quality inspection results (pass/fail, measured values, inspector)
• Shipment destination, shipment quantity, and shipment date (linking product lots to shipment destinations)

At Japanese-owned food factories in Thailand, it is not uncommon for raw material lots and product lots to be managed in separate Excel files — meaning that when a problem occurs, simply cross-referencing the two can take half a day. This “lot traceability lead time” represents a significant practical risk in quality management operations.

Furthermore, enabling AI-driven quality prediction (e.g., “using this raw material lot tends to cause color unevenness in the finished product”) or process optimization (e.g., “there is a high probability of meeting quality standards even if heating time is reduced by five minutes”) requires raw material lots, process conditions, and quality results to be integrated into a single table. Building such an integrated database starts with individual lot data being digitized and structured.

Domain 4: Quality Inspection Data — Electronic Records of Measurements and Judgments

Quality inspection is the cornerstone of food manufacturing, but from a data utilization perspective, many factories are sitting on an untapped resource. Inspection results are recorded — but only on paper inspection sheets or Excel sheets that inspectors fill in manually. In this state, using inspection data for analysis requires manual data collection and conversion each time, making real-time analysis and Statistical Process Control (SPC) virtually impossible.

The content that must be established as quality inspection data is as follows.

• Inspection date and time, inspector, and inspection location (receiving / in-process / pre-shipment)
• Lot number and product code of the item inspected
• Measured value for each inspection item (recorded as a number, not just “OK/NG”)
• Pass/fail judgment, basis for the judgment, and approver
• Disposition when failing (disposal, re-inspection, downgrade, return, etc.)
• Inspection equipment used and calibration records

Particularly important is “recording measured values as numbers.” With only a binary pass/fail, trend analysis over time and outlier detection are both impossible. Preventive quality control — such as issuing an alert when a pH reading reaches 90% of the upper specification limit — can only be realized once numerical data has been accumulated.

Temperature management is especially critical in food manufacturing. Continuous electronic storage of temperature records is required — from a HACCP perspective and a quality assurance perspective — at all of the following points: raw material receipt, refrigerated and frozen storage, thawing, heat treatment, rapid cooling, and pre-shipment inspection. Automated temperature recording via IoT sensors is a realistic, low-cost option for achieving this “continuous temperature record.”

Domain 5: Shipment Data — Integration with Inventory, Billing, and Traceability

Shipment data is the “last line of defense” in food manufacturing. It is the only data that confirms whether manufactured goods have reached customers at the right quality, quantity, and delivery time, and it also serves as the starting point for customer billing, inventory drawdown, and product lot traceability.

The minimum elements that must be managed as shipment data are as follows.

• Shipment date and time, and destination (customer code, delivery address)
• Shipped product code, product name, lot number, and quantity
• Carrier and waybill number (for logistics tracing)
• Quality check results at shipment (temperature check, etc.)
• Invoice number and linkage to billed amount

One of the most common problems observed at Japanese-owned food factories in Thailand is a “mismatch between actual shipments and invoices.” Because shipping staff and back-office staff manage records in separate systems — or with paper and Excel — a reconciliation exercise arises at month-end, uncovering billing omissions and quantity errors. This is not merely a clerical mistake; the root cause is that data systems are not connected.

Moreover, when a food recall occurs, having well-organized shipment lot data allows you to identify “which lots were delivered to which customers” within minutes. Without that data, the process of working through phone calls, emails, and paper delivery slips begins — leading to an expansion of the damage and a loss of customer trust.

Numerical Management for Visualizing Yield and Food Loss

In food manufacturing, yield is the proportion of input raw materials that can be shipped as finished product. Low yield means raw material costs are being wasted, and disposal costs are also incurred. Yet many factories have no accurate grasp of their “yield.”

To accurately track yield, at a minimum the following figures must be recorded on a per-manufacturing-lot basis.

• Raw material input quantity (theoretical value per recipe vs. actual input quantity)
• Waste and returned quantities during manufacturing (by process)
• Work-in-progress and semi-finished goods inventory quantities
• Completed quantity of shippable product
• Disposal quantity due to quality failure
• Quantity disposed of due to expiration or warehouse waste

When this data is in place, analysis becomes possible: “Which process generates the most yield loss?” “Which raw materials or supplier lots have the highest disposal rates?” “Is there a correlation with season or temperature?” The results of this analysis are the very first step toward AI-driven process optimization and predictive maintenance.

Reducing food loss is becoming increasingly important not only as a cost reduction measure but also from ESG and SDG perspectives. At Japanese companies with operations in Thailand, sustainability reporting requirements from Japanese headquarters grow stronger each year, and “waste reduction track records” are increasingly expected to be reported in specific numerical terms. Factories that lack waste volume data cannot meet this requirement.

Temperature Management Data: The Intersection of Food Safety and AI

Temperature management is “the foundation of quality assurance” in food manufacturing, and it is also one of the most realistic entry points for AI adoption. Temperature is numerical data that IoT sensors can capture inexpensively, continuously, and automatically — and it has strong compatibility with existing HACCP management practices.

Common temperature management challenges seen at food factories in Thailand are as follows.

• Refrigerator and freezer temperature records consist only of a single manual reading per day (no records at night or on holidays)
• Temperature alerts sound but leave no record, or records are modified after the fact
• Product temperature checks at raw material receipt are visual only, with no numerical records
• Temperature logs for heat treatment (sterilization, cooking) exist only as paper records
• Product temperature checks at shipment are rough and not managed numerically

Introducing continuous automated temperature recording via IoT sensors resolves many of these challenges. Furthermore, combining historical temperature logs with quality inspection results enables correlation analysis such as: “Lots where the refrigerator temperature rose by 2°C during a specific time window have a higher probability of failing inspection later.” This is the foundation for AI-based quality prediction.

From an initial investment standpoint, IoT temperature sensors can be introduced at a relatively low cost. The specifics vary depending on whether existing Wi-Fi infrastructure can be leveraged or wired connections are required, but cloud-based sensor services make numerical temperature management achievable from the first year. This investment is also connected to energy savings — by detecting refrigeration equipment anomalies early to prevent wasted power consumption — and may therefore qualify for BOI incentives targeting energy efficiency investment.

Prioritizing Investment: What to Pause and What to Continue

In an environment of increasing economic uncertainty, neither pausing all investment nor continuing all investment is the right answer. The following framework provides a perspective for prioritizing DX investment in food factories.

What matters is “tying investment objectives to specific numbers.” Rather than a vague goal like “efficiency through DX,” targets should be set in concrete terms — for example, “by tracking raw material disposal at the lot level, we project a 30% reduction in a monthly disposal cost of X ten-thousand baht” or “by digitizing inspection records, we will reduce time spent on quality audit response by X hours per week.” This specificity is essential both for explaining proposals to Japanese headquarters and for maintaining on-site team motivation.

Using BOI to Plan Investment in Data Infrastructure

Thailand’s BOI (Board of Investment) provides incentives such as corporate income tax exemptions and import duty exemptions on equipment for investments related to automation, AI, data analytics, and enterprise management IT. Investment in data infrastructure for food factories can qualify for these incentives.

Key points for maximizing BOI incentives are as follows.

• Prepare applications at the investment planning stage: BOI incentives are in principle applied for before the investment is made. Because applications submitted after system deployment may not be approved, BOI application must be kept in view from the earliest stage of the investment plan.
• Clarify eligible investments: Confirm in advance which of the following fall under BOI-eligible categories: IoT sensors, quality management software, inventory management systems, paperless apps, and AI tools.
• Prepare investment plans in Thai and English: BOI applications require a detailed investment plan in Thai or English. Documentation prepared for BOI submission must be prepared separately from internal materials written in Japanese.
• Verify the eligibility of the local entity: To benefit from BOI incentives, the local legal entity must either already be a BOI-approved company or be in a position to apply for new BOI certification.

In practice, it is most efficient to proceed in coordination with BOI application specialists — accounting firms, law firms, or BOI-registered consultants. TOMAS TECH shares foundational BOI-related information at the system proposal stage and supports coordination with specialists.

Common Failure Patterns in Data Readiness — and How to Avoid Them

Data infrastructure initiatives in food factories more often stall due to “people and operations” problems than technical ones. Below is a summary of common failure patterns and how to avoid them.

Failure Pattern 1: Trying to do everything at once

Attempting to digitize raw materials, recipes, lots, inspections, shipments, and inventory all at the same time causes the project to grow complex, resistance on the factory floor intensifies, and ultimately nothing becomes established practice. The way to avoid this is to start with “the most painful problem.” If quality complaints are frequent, begin with digitizing inspection data. If disposal losses are large, start with inventory and lot management. Set priorities accordingly.

Failure Pattern 2: Only Japanese staff can use the system

A system is introduced, but Thai staff responsible for data entry cannot operate it effectively, and Japanese staff end up entering data on their behalf. The way to avoid this is to select systems and plan implementations with “Thai-language UI, Thai-language manuals, and Thai-language OJT” as baseline requirements. Prioritizing simple input screens — selection menus, barcode scanning, photo capture — and minimizing keyboard entry is also effective.

Failure Pattern 3: Data is entered but nobody looks at it

Data is being entered, but no one is using the reports or dashboards — resulting in a situation where “only the data entry workload has increased.” The way to avoid this is to design reports from the outset that enable on-site managers (including Thai team leaders) to clearly see “what this data tells us and what decisions it informs.”

Failure Pattern 4: Data is collected solely for headquarters reporting

Data is gathered only for KPI reporting to Japanese headquarters and is not utilized at the local factory floor level. This fails to motivate the on-site team, and data quality deteriorates. The way to avoid this is to build a structure where data is also used for local problem-solving — waste reduction, complaint response, shift scheduling, and so on. Report design that serves both local and headquarters needs is key.

Failure Pattern 5: Leaving operational design entirely to the vendor

System implementation is left to the vendor, but “who enters which data, when,” “the response flow when data problems arise,” and “who is responsible for periodic data quality checks” are never defined — so data quality deteriorates shortly after go-live. The way to avoid this is to design “data governance” operations in parallel with system implementation. This typically does not require hiring a consultant at additional cost; clearly designating a factory manager as the responsible owner is usually sufficient.

Phased Implementation: The “Start with One Process” Approach

What TOMAS TECH recommends for food factory data infrastructure development is a phased implementation approach that starts with the unit of “one process, one warehouse, one form.” This is not a compromise born of small scale — it is a strategy that prioritizes certainty: establish the practice, then expand horizontally.

A typical phased implementation flow looks like this.

• Phase 1 (Months 1–3): Select the one process with the greatest challenges and design and implement a data collection method. Example: digitizing quality inspection data at raw material receipt (tablet entry + barcode scanning). The goal at this stage is that data “gets entered” — in other words, operational adoption.
• Phase 2 (Months 3–6): Begin using the data collected in Phase 1 for analysis. Conduct analysis such as “which suppliers have the highest rejection rates at incoming inspection?” and “does the NG trend correlate with season or lot?” and connect results to concrete improvement actions. Demonstrating Phase 2 results in numbers is critical for building internal consensus to move forward to the next phase.
• Phase 3 (Months 6–12): Expand the scope of data management to processes adjacent to Phase 1 (manufacturing lot management, in-process inspection, etc.). By applying the tools, formats, and operating rules used in Phase 1 horizontally, adoption can be achieved faster and at lower cost than Phase 1.
• Phase 4 (Month 12 onward): Aim for a state where the entire flow — raw materials → manufacturing → inspection → shipment — is connected through data. Only at this stage does investment in AI analytics tools and predictive models become meaningful.

The advantages of this phased implementation approach are three: low initial investment (Phase 1 alone can often be completed within a budget of several hundred thousand baht), results are confirmed early, and the adaptation burden on factory floor staff is low. Additionally, by reporting Phase 1 results to Japanese headquarters, it becomes easier to obtain investment approval for Phase 2 and beyond.

Explaining to Japanese Headquarters: The “3-Year Investment Plan in Numbers”

To gain approval from Japanese headquarters for DX investment at a Thailand site, qualitative explanations such as “it will be more convenient” or “we can improve efficiency” are not sufficient. What headquarters demands is quantitative evidence of “return on investment” and “risk reduction.”

For food factory data infrastructure investment, building the case around numbers in the following areas forms the backbone of the explanation.

• Disposal cost reduction: Identify the current monthly cost of raw material and finished product disposal, and set a target reduction percentage achievable through improved data management (e.g., “of a monthly disposal cost of XX ten-thousand baht, we project a 30% reduction through strengthened lot management”).
• Quality complaint response cost: Convert the man-hours required to handle each complaint (investigation, reporting, corrective action) into a monetary value, and estimate the impact of reducing complaint volume.
• Management labor reduction: Assess the current man-hours spent on daily report preparation, aggregation, and report writing, then estimate post-digitization savings (e.g., “XX hours per month × hourly labor cost = equivalent of XX ten-thousand baht per month”).
• Quality audit response cost: Estimate the time currently spent responding to quality audits by customers and certification bodies, and the reduction achievable through enhanced traceability.
• Risk cost (estimated recall scenario): Present both qualitative and quantitative assessments of the projected cost of a recall (product retrieval, disposal, customer compensation, reputational damage) and the risk reduction achievable through stronger traceability.

By building these numbers into a case and demonstrating that “payback within 3 years is achievable,” a sufficiently compelling argument for headquarters approval is created. Accurately performing this calculation itself requires data on current disposal volumes, complaint counts, and management man-hours — which is itself a paradoxical demonstration of why data infrastructure matters.

The TOMAS TECH Perspective

TOMAS TECH supports Japanese manufacturers and food processors in Thailand and across ASEAN in building on-site data foundations, centered on the inventory management system “PEGASUS,” the paperless app “i-Reporter,” the operation management system, and the smart watch system.

The inventory management system “PEGASUS” provides real-time visibility into raw material, WIP, and finished goods inventory and enables lot-level traceability. It directly contributes to reducing raw material disposal in food factories, improving the accuracy of best-before date management, and reducing the risk of excess inventory and stockouts. Data linkage from raw material receipt through shipment can also be built on the PEGASUS platform.

The paperless app “i-Reporter” digitizes forms, daily reports, inspection records, and work checklists on the manufacturing floor. It eliminates paper recording and transcription work, enabling managers to review data recorded on the floor in real time. With Thai-language support, tablet operation, and photo attachment functionality, it has a track record of adoption by Thai staff. Digitization of quality inspection data is one of the areas where i-Reporter delivers implementable results most quickly.

The operation management system provides real-time visibility into the operating status of manufacturing lines and equipment. At food factories, combining utilization data with quality data enables correlation analysis such as “NG rates are higher immediately after line startup” or “yield declines during specific time windows.” This is the first step toward AI-driven predictive maintenance and process optimization.

The smart watch system delivers alert notifications and work instructions to on-site operators via watch devices. By immediately notifying floor staff of temperature anomalies and quality NG alerts, it contributes to early problem response and prevention of recording omissions.

TOMAS TECH’s foundational approach is “start with one process, confirm the results, and then expand horizontally.” Rather than a large-scale system overhaul, the priority is to begin with the most pressing problem and establish the system in a form that the local team can operate effectively. Consultation on BOI incentive opportunities is also available from the earliest stage of investment planning.

For inquiries and consultations, please visit https://tomastc.com/contact.

Summary

The theme of “implementing AI in food factories” remains a critical challenge in the Thai business environment of 2026. However, AI is “something that works on a foundation of well-organized data,” and deploying AI before that data is in order will not deliver the expected return on investment.

The five domains organized in this article — raw materials, recipes, lots, quality inspections, and shipments — being digitized, structured, and interconnected is the “prerequisite” for AI adoption. Establishing this prerequisite is both preparation for AI implementation and, in its own right, a direct source of improvement in on-site quality, cost, and management efficiency.

Visualizing quality, temperature, lots, and yield is the foundation for reducing food loss and risk, fulfilling accountability to Japanese headquarters, and meeting the quality demands of Thai authorities and customers. Only with this foundation in place does AI function as “a tool that changes the numbers on the factory floor.”

What matters is not attempting to build the perfect foundation all at once. Start with the most painful problem, confirm results at the unit of one process, one form, one warehouse, and build from there. That accumulation of progress leads, three years from now, to “a food factory where AI is used as a matter of course.”

Building a data foundation is a source of long-term competitiveness, regardless of the economic climate. Now is precisely the right moment to take the first step toward “DX that changes the numbers on the factory floor” — not “DX as a trend to follow.”

References

• World Bank Thailand
• Thailand BOI (Board of Investment)
• JETRO Thailand
• S&P Global PMI
• Ministry of Economy, Trade and Industry — Manufacturing White Paper 2025