TİCİS TÜRKİYE · 14.09.2026
14 September 2026 · Weekly publication · TİCİS assessment
Investing in AI means more than purchasing software. It changes how decisions are prepared, production is monitored and responsibility is assigned when something goes wrong. TİCİS recommends judging success by measured business outcomes, reliability and sustainable operating costs.
Where can management benefit?
Our suggested starting points include draft management reports, combined sales and inventory analysis, supplier-quote comparisons and access to internal knowledge. Managers can compare demand scenarios more quickly, but forecasts must remain clearly distinguishable from recorded facts.
Use verified data and calculation tools for financial calculations, with language models supporting explanations and summaries. Define authority and human review for consequential payments, contracts, hiring and dismissal decisions. Decision support does not remove management accountability.
AI and production automation are different
The OECD manufacturing review discusses maintenance and defect-detection applications and highlights the importance of data preparation and human effort.
OECD — AI in manufacturing, 2026
A production line can operate automatically without AI. AI may contribute to maintenance, quality or planning by identifying patterns in sensor and image data. Specify whether material feeding, inspection, packaging, maintenance and fault response fall within the proposed automation scope.
Begin with a narrow comparison against the current method, such as visual inspection for one product group. Measure missed defects and sound products wrongly rejected, as well as detected faults. Validate the engineering before connecting model outputs to physical control systems.
Full automation still needs exception handling
Defining full automation as requiring no people hides maintenance and exceptions. Plan safe shutdown and controlled restart for sensor failures, loss of power or connectivity, changed materials and unexpected product variation.
TİCİS advises against making a generative model the sole safeguard for machinery. Qualified engineers should assess independent safety protections and validate automation changes against applicable machinery and workplace-safety requirements.
Data security and unreliable outputs
NIST’s voluntary framework addresses trustworthiness across system design and evaluation and provides a separate generative-AI risk profile. It is not product certification or a guarantee of error-free operation.
NIST — AI Risk Management Framework
Specify which services may receive confidential contracts, employee information and manufacturing recipes. Review provider terms for retention, training use, access and deletion. Plan least-privilege access, activity logs, backups and controlled connections between office and production networks.
Test false information, bias, leakage and attempts to redirect the system through external documents. Establish transaction limits and approval rules before allowing an AI-generated email or purchasing proposal to trigger an external action.
People and changing responsibilities
The OECD identifies opportunities for productivity and job quality alongside risks that need addressing.
Involve operators, maintenance specialists, quality teams and management. Training should cover recognising incorrect outputs and stopping a process, not just operating the tool. Review worker monitoring and performance assessment separately; productivity targets should respect fair evaluation and privacy.
Build a complete investment case
Include integration, data cleaning, sensors, hardware, training, maintenance, energy, security and downtime as well as licence fees. Examine data portability and operational continuity if the supplier changes.
Measure processing time, defects, scrap, unplanned downtime, rework and human review time together. Hours released are not automatically cash savings: explain how the capacity will be used. Compare similar product mixes and workloads rather than assuming a guaranteed return.
TİCİS recommends a first 90-day plan
Days 1–30: inventory processes and data, assign an owner and measure baseline performance. Days 31–60: run a limited pilot under human supervision and test failures and interruptions. Days 61–90: compare results with total costs and document whether to continue, revise or stop. This is a suggested timetable; safety-critical validation may take longer.
TİCİS assessment: competitive advantage may come from choosing the right process, measuring results and preserving responsibility. Establish controlled success before scaling. This commercial analysis does not replace site-specific engineering or legal assessment.
Discuss your digital transformation investment with TİCİS
The legal framework in Türkiye: personal data and workplace safety
The Turkish data protection authority’s generative-AI guide addresses personal data processing across the system lifecycle. Assess purpose, processing conditions, information duties, retention and access for each use case. The guide provides guidance alongside the law.
KVKK — Üretken Yapay Zekâ Rehberi
Sending personal data to a foreign cloud service also requires a cross-border transfer assessment. Article 9 mechanisms involving adequacy, appropriate safeguards and exceptional transfers have distinct conditions. A service subscription alone does not establish a lawful transfer basis.
KVKK — Cross-border data transfers
The Ministry of Labour explains employers’ duty to conduct or commission risk assessment under Law No. 6331. Assess new hazards, equipment changes and worker exposure when introducing automation.
Turkish Ministry of Labour and Social Security — Risk assessment
Transaction-specific legal assessment
Automated recruitment and performance scores, employee privacy, biometric data, supplier liability, intellectual property and allocation of production losses require legal assessment. This section provides general information based on official sources; it does not establish legal compliance or exemption from liability for a specific business.