Session10_LSCM.pdf

# Introduction to AI in Supply Chain Management This section provides an overview of Artificial Intelligence (AI) and its applications within supply chain management, introduced by LP Kerkhove, founder and CTO of Crunch Analytics. ### 1.1 Speaker and company introduction * LP Kerkhove is the Founder and CTO of Crunch Analytics, a company established in 2016 [2](#page=2). * Crunch Analytics specializes in data-driven innovation, particularly for fashion retailers [2](#page=2). * Kerkhove's academic background includes an MS in Business Engineering and a PhD in Operations Research, and he holds a professorship at Ghent University [2](#page=2). * He is also the author of "Data-driven Retailing: A Non-technical Practitioners' Guide" [2](#page=2). * Crunch Analytics is described as a data and AI partner that delivers actionable results, with offices in Ghent and Rotterdam and a team of 25 experienced data professionals [3](#page=3). ### 1.2 Crunch Analytics' solution approach Crunch Analytics' solutions frequently integrate multiple technologies to address business challenges. These capabilities typically encompass [4](#page=4): * **Data platform:** Creating or enhancing data architectures for stable and cost-effective solutions [4](#page=4). * **Optimization:** Determining the best compromises at a portfolio level and adhering to operational constraints [4](#page=4). * **Predictive models:** Forecasting demand and analyzing its reaction to factors like price changes [4](#page=4). * **User interface:** Integrating developed tools into existing business processes [4](#page=4). ### 1.3 Understanding Artificial Intelligence (AI) AI refers to a collection of algorithmic tools designed to predict the future, understand relationships, and reduce uncertainty [5](#page=5) [6](#page=6). #### 1.3.1 Core AI capabilities and applications in supply chain AI can be broadly categorized by its core functionalities, which translate into powerful applications within supply chain management: * **Prediction:** * **Purpose:** To forecast future events or trends, thereby reducing uncertainty [6](#page=6). * **Example Tool:** Neural networks are often employed for predictive tasks [6](#page=6). * **Optimization:** * **Purpose:** To make better, more optimal decisions in complex situations with numerous variables [6](#page=6). * **Example Tools:** Technologies like CPLEX, Gurobi, Linear programming, and heuristics are used for optimization [6](#page=6). * **Simulation:** * **Purpose:** To answer "what-if" questions and understand the dynamic interactions between different parts of a system [6](#page=6). * **Example Tools:** Digital twins and Monte Carlo simulations are common tools for this purpose [6](#page=6). * **Generation:** * **Purpose:** To structure information and create new content such as documents or images by following existing patterns [6](#page=6). * **Example Tools:** Large Language Models like ChatGPT and tools like CoPilot fall under generative AI [6](#page=6). #### 1.3.2 Supporting capabilities for AI implementation Successful AI implementation in supply chain management also relies on several supporting capabilities: * **Data engineering:** Selecting appropriate tools and building functional data structures [6](#page=6). * **Software engineering:** Developing interfaces for interaction with systems [6](#page=6). * **Business acumen:** Effectively integrating AI solutions into business processes [6](#page=6). --- # AI for Forecasting and Demand Prediction AI significantly enhances forecasting accuracy and reduces uncertainty in demand predictions by addressing noise and historical biases in sales data, ultimately aiming to prevent future lost sales [10](#page=10) [15](#page=15) [17](#page=17). ### 2.1 The role of AI in reducing forecasting uncertainty Improving forecasting accuracy through AI is considered a major advantage in supply chain management, often referred to as the "closest thing to a free lunch". By reducing uncertainty, AI helps in optimizing inventory levels and mitigating the need for excessive safety stock. Methods to reduce the need for safety stock include shortening lead times, reducing lead time variability, increasing supply chain visibility, and improving replenishment frequency. AI contributes to these improvements by providing more accurate demand predictions [10](#page=10) [8](#page=8) [9](#page=9). ### 2.2 Examples of AI in forecasting #### 2.2.1 Weather patterns and demand forecasting AI can leverage external factors, such as weather patterns, to improve demand predictions, especially for perishable goods [11](#page=11). > **Example:** A company selling perishable products uses a simple 5-day rolling average for demand prediction. They observe a strong correlation between temperature and sales. By implementing an AI forecast that considers weather data, they can significantly reduce total daily costs associated with overstocking and understocking compared to a traditional rolling average method. For instance, a rolling average might result in daily costs of 32,421 dollars, while an AI forecast could reduce this to 16,274 dollars. Further improvements can be seen with more sophisticated AI approaches, potentially reducing costs to 10,762 dollars per day [11](#page=11) [13](#page=13) [14](#page=14). #### 2.2.2 AI for structuring data and complex models AI serves multiple primary uses in Supply Chain Management (SCM), including reducing noise in data, forecasting demand, structuring data, automating processes, and aiding understanding through complex models [15](#page=15) [16](#page=16). ### 2.3 Advanced techniques in AI-driven demand prediction Effective AI-driven demand prediction involves several key components: #### 2.3.1 Correcting historical bias for lost sales Observed demand data often under-represents true demand because it doesn't account for periods of stockouts where sales were lost. Working solely with observed demand favors models that underestimate future demand, leading to recurring lost sales [18](#page=18). > **Tip:** AI can reconstruct an estimate of the "counterfactual" demand – what demand would have been if sufficient inventory had been available. This is crucial for preventing future lost sales that are often less visible than excess inventory [18](#page=18). One method to address this is using Bayesian time series models. If the sales of one stock-keeping unit (SKU) are correlated with another, AI can estimate the lost sales of the out-of-stock SKU based on the sales of the correlated SKU. For example, if SKU2 sales drop to zero due to a stockout, but are correlated with SKU1 sales, AI can estimate the true demand for SKU2 [18](#page=18). #### 2.3.2 Stochastic forecasting Stochastic forecasting provides a probabilistic view of future demand, offering a clearer understanding of the value at risk and potential fluctuations. This approach moves beyond single-point estimates to provide a range of possible outcomes, often expressed as confidence intervals (e.g., 95% CI) [19](#page=19). > **Tip:** Stochastic forecasts are essential for identifying risks that might be masked in an average-case scenario. They allow for better planning around potential deviations [19](#page=19). These forecasts can also be used to project metrics like On-Time In-Full (OTIF) performance, showing the expected percentage and its confidence bounds alongside target values [19](#page=19). #### 2.3.3 Hierarchical aggregations and multi-model combinations Demand prediction can be performed at various levels of aggregation (e.g., by product, region, or channel), and AI can manage these hierarchical structures effectively. Combining multiple forecasting models (multi-model combinations) often yields more robust and accurate predictions than relying on a single model [17](#page=17). #### 2.3.4 Application-specific performance optimization AI allows for optimization tailored to specific applications and business goals. This includes incorporating factors like price-driven models and considering service level targets [17](#page=17). #### 2.3.5 Stock-ruptures and lost sales AI directly addresses the prediction and mitigation of stock-ruptures and their associated lost sales [17](#page=17). #### 2.3.6 Working with probabilities AI-driven forecasting that outputs probabilities enables better downstream decision-making in inventory management, logistics, and sales planning [17](#page=17). ### 2.4 The limitations of aggregated performance measures Aggregated performance measures can obscure underlying issues within a supply chain. For example, a global service level might meet targets at 99%, but this can hide significant performance deviations for individual customers. If Customer D's service level drops to 96% while others are at 100%, the overall average may appear acceptable, masking potential customer dissatisfaction and lost future business [20](#page=20). ### 2.5 Beyond traditional time series It's important to recognize that effective forecasting involves more than just traditional time series analysis. AI can incorporate a wider array of variables and complex relationships to create more comprehensive and accurate predictions [21](#page=21). --- # Causal models and product performance analysis Causal models are employed to understand the underlying relationships between product features and their performance dimensions, enabling predictions for new products and explanations for existing ones [25](#page=25). ### 3.1 Quantifying product performance Product performance is assessed using multi-dimensional measures that capture various aspects of a product's value. These dimensions often include [23](#page=23): * **Sales volume**: The total quantity of a product sold, indicating its operational scale [23](#page=23). * **Margin**: The profitability of a product, directly related to logistical and production costs [23](#page=23). * **Uniqueness**: A measure of how similar a product is to others in the assortment and the likelihood of customer substitution [23](#page=23). * **Segment importance**: The disproportionate significance of a product to specific key customer segments or strategically important regions [23](#page=23). * **Unique customers**: The number of distinct customers purchasing a product and their importance to the business [23](#page=23). * **Sales trend / Product lifecycle**: Assessed using time series analysis to identify upward, constant, or downward sales trajectories [23](#page=23). * **Sustainability**: The degree to which a product contributes positively to environmental or social goals [23](#page=23). * **Competitive advantage**: The extent to which a product is essential for competing with other suppliers, acting as a key staple or diversification tool [23](#page=23). ### 3.2 Mapping existing and past products A multi-dimensional performance space can be constructed to visualize and analyze existing and past products. This involves mapping products across different performance dimensions, such as uniqueness and gross margin [24](#page=24). > **Tip:** Products with the lowest scores across multiple dimensions are often candidates for elimination [24](#page=24). Pairwise charts can be used to explore the relationships between products and performance dimensions, providing explainability for recommendations, such as SKU elimination. While manual analysis is difficult for multi-dimensional decision-making, these visualizations aid the process [24](#page=24). ### 3.3 Causal models for explaining relationships The core idea of causal models in this context is to determine if product performance can be predicted based on its features. These models aim to uncover "hidden relations" between product attributes and their performance [22](#page=22) [25](#page=25). By analyzing product attributes, causal models can: * Estimate the performance of a new product before its launch [25](#page=25). * Provide insights into why an existing product might be underperforming in a specific dimension [25](#page=25). * Predict the likely performance of new products entering the market [25](#page=25). A common approach involves using a waterfall diagram to summarize the effects of complex underlying models in an easily understandable format. This diagram can illustrate the positive and negative contributions of specific product attributes to a performance dimension, such as expected sales volume [26](#page=26). For example, a model might predict expected sales volume based on attributes like "Fairtrade," "Contains gluten," "Shelf life < 2mth," and "Env. Score A". The model can quantify the impact of each attribute, showing, for instance, a positive contribution from "Broccoli" and a negative impact from "Contains gluten". This analysis can either confirm existing hypotheses about performance drivers or challenge them [26](#page=26). ### 3.4 Combining perspectives for a comprehensive solution Often, a complete solution integrates multiple analytical perspectives. This can involve combining demand forecasts with estimates of price-response elasticity and historical sales data for product clusters [27](#page=27) [28](#page=28). For instance, a demand forecast can be generated by considering: * The product cluster's observable trend [28](#page=28). * The product-specific historical sales data [28](#page=28). * An estimate of the price-response, which indicates how demand changes with price variations [28](#page=28). Models can also predict outcomes at different price points (e.g., "Demand at 0%", "Demand at 30%", "Demand at 50%"). Similarly, cumulative objectives, such as margin, can be projected under different scenarios [28](#page=28). > **Example:** A sigmoid function might be used to model the relative increase in demand as a product's price changes from a markdown to full price [28](#page=28). Ultimately, a final selection of products or strategies is made by accounting for available inventory and broader business constraints. This multi-faceted approach ensures that decisions are informed by causal understanding, performance metrics, and practical business realities [28](#page=28). --- # AI for Automation and Structuring Data in SCM AI plays a crucial role in automating repetitive tasks and structuring data within Supply Chain Management (SCM), significantly enhancing efficiency and decision-making. This involves various forms of automation and the intelligent processing of sales information and product assortments [29](#page=29) [33](#page=33) [39](#page=39). ### 4.1 Forms of automation Automation in SCM can manifest in several ways, ranging from simple task execution to complex decision-making and the operation of autonomous agents [34](#page=34). #### 4.1.1 Narrow automation This involves the use of "bots" to automate tasks typically performed by human agents on a computer. These tasks are often repetitive and data-intensive [34](#page=34). #### 4.1.2 Broad automation This form of automation utilizes technology to automate entire business processes and workflows. It can encompass decision-making and leverage advanced algorithms such as AI and Machine Learning [34](#page=34). #### 4.1.3 Extended automation This represents the highest level, where autonomous agents take on roles similar to human employees. These systems aim to perform complex reasoning and maintain stateful information [34](#page=34). ### 4.2 Structuring sales information A significant application of AI in SCM is structuring sales information, which is often communicated informally by sales teams who may not be statisticians [35](#page=35) [36](#page=36). > **Example:** Salespeople commonly communicate updates like "The customer contracted for high volume, but their actual pull is lagging. Please slash the forecast to match their current run-rate, or we’ll be sitting on stock". Another example is: "Customer X just confirmed a 250T contract on Item Y. We need to start loading site Z next month. It’ll be full trucks spread out over the next 90 days" [36](#page=36). AI, particularly through Large Language Models (LLMs), can parse these natural language communications and structure the information for planning systems. This enables automation of repetitive work for planners, such as updating forecasts or requesting approvals [37](#page=37). The process typically involves: * A salesperson communicating an update. * An LLM structuring this information into a format understandable by the system. * The system processing the structured information to update forecasts or trigger approval requests [37](#page=37). * Planners then approve or reject these automated updates, ultimately leading to an ERP update [37](#page=37). ### 4.3 Benchmarking product assortment AI is also utilized to benchmark a company's product assortment against that of its competitors [30](#page=30) [31](#page=31) [32](#page=32). This process involves several steps: * **Scraped product:** Gathering product data from competitor websites or other sources. * **Mapping:** Using logic and potentially an LLM to map these scraped products to internal product databases or a dedicated product database [32](#page=32). * **Category matcher/extender:** Identifying how competitor products fit into existing categories or extending the understanding of new categories [32](#page=32). * **Exact match lookup:** Attempting to find direct matches between scraped products and internal product catalog [32](#page=32). * **Category database:** Leveraging a database of categories for comparison [32](#page=32). This capability allows businesses to understand their competitive positioning in terms of product offerings [31](#page=31). --- # Risks and considerations of AI implementation Implementing AI is not without its risks and necessitates careful consideration of several factors to ensure effective and ethical integration. These risks span from overestimating human understanding of AI to the potential impact on cognitive processes and task enjoyment [41](#page=41) [42](#page=42) [48](#page=48). ### 5.1 Overestimating understanding A significant risk lies in overestimating people's understanding of how AI operates and its capabilities. This can lead to misaligned expectations and improper use of AI tools [42](#page=42). ### 5.2 Automation errors and management Automation, while offering efficiency, introduces the risk of errors occurring at a much greater velocity than in non-automated processes. Managing these risks requires robust strategies [43](#page=43): * **Guardrails:** Limiting the permissions granted to AI systems to prevent unintended actions [43](#page=43). * **Automated sanity checks:** Implementing systems that monitor AI output and trigger a "kill switch" if the process goes out of control [43](#page=43). * **Human supervision:** Retaining a human in the final step of a process to provide approval and oversight [43](#page=43). ### 5.3 Tacit knowledge and conflicting objectives Tacit knowledge, the often unarticulated "how" and "why" behind human actions, presents a challenge for AI implementation. This can lead to [44](#page=44): * **Conflicting objectives:** When prompting different actors or systems, the underlying purpose may not be clearly defined or understood, resulting in disparate goals [44](#page=44). * **Divergent approaches:** Different individuals or systems may employ vastly different methods for performing the same task, making it difficult to determine the "correct" approach [44](#page=44). #### 5.3.1 Example of tacit knowledge in forecasting Surveys of sales teams using existing processes highlight how tacit knowledge differences manifest. For instance, when asked what a "forecast" number represents, salespeople provided varied interpretations: * **Maximum Potential:** The amount achievable if all favorable conditions are met [45](#page=45). * **Most Likely:** A realistic, middle-ground estimate [45](#page=45). * **Conservative Estimate:** An amount the individual is highly confident of selling, for safety [45](#page=45). Similarly, when asked about agreement structures for sales volume, interpretations differed significantly regarding formal contracts versus informal arrangements. These discrepancies underscore the difficulty of automating processes that rely on nuanced, unstated assumptions [45](#page=45) [46](#page=46). ### 5.4 Impact on human cognitive processes and enjoyment The interaction model of many AI tools, particularly large language models (LLMs), can disrupt cognitive states conducive to deep work and enjoyment [48](#page=48). * **Flow state:** AI interaction can break up the "flow state," characterized by complete concentration, altered time perception, a balance between challenge and capability, and overall enjoyment [48](#page=48). * **Multitasking and reduced output:** The wait-and-response nature of AI interactions can encourage multitasking and potentially lessen overall output and enjoyment [48](#page=48). ### 5.5 The Chinese Room Argument and understanding The philosophical "Chinese Room" thought experiment, proposed by John Searle, questions whether a system can truly "understand" without conscious awareness [49](#page=49). * **Searle's perspective:** Argues that a person following rules to translate Chinese without understanding the language demonstrates that the system itself does not understand semantics, only syntax [49](#page=49). * **Dennett's perspective:** Counter-argues that from a holistic view, there's no fundamental difference between the Chinese Room system and the processes within a biological brain [49](#page=49). The relevance to AI implementation is whether the overuse of LLMs risks users becoming like the person in the box – performing tasks successfully but without genuine comprehension of what they are doing. This raises the question of whether organizations should be concerned about this potential disconnect [50](#page=50). > **Tip:** Be mindful of how you phrase prompts to AI, ensuring they capture the underlying intent rather than just surface-level instructions, especially when dealing with tacit knowledge. > > **Tip:** When automating processes, explicitly define different types of estimates or objectives, as demonstrated by the sales forecasting example, to mitigate ambiguity arising from tacit knowledge differences. > > **Tip:** Consider the design of AI tools and workflows to minimize disruption to deep work and flow states, rather than inadvertently promoting multitasking. --- ## Common mistakes to avoid - Review all topics thoroughly before exams - Pay attention to formulas and key definitions - Practice with examples provided in each section - Don't memorize without understanding the underlying concepts

| Term | Definition | |------|------------| | AI | Artificial Intelligence, a field of computer science focused on creating systems that can perform tasks typically requiring human intelligence, such as learning, problem-solving, and decision-making. | | Supply Chain Management (SCM) | The oversight of materials, information, and finances as they move in a process from supplier to manufacturer to wholesaler to retailer to consumer. It involves managing all activities in procurement and purchasing, conversion, and all logistics management activities. | | Data platform | A centralized system for collecting, storing, processing, and analyzing data, enabling efficient data-driven decision-making and the development of AI solutions. | | Predictive models | Algorithms designed to forecast future events or outcomes based on historical data, identifying patterns and correlations to make informed predictions. | | Optimization | A mathematical process used to find the best possible solution to a problem under a given set of constraints, aiming to maximize or minimize a specific objective function. | | Simulation | A technique used to model real-world processes or systems over time, allowing for the exploration of "what-if" scenarios and understanding the dynamic interactions between various components. | | Generation | In the context of AI, this refers to the creation of new content, such as text, images, or data, based on learned patterns and existing information. | | Neural networks | A type of machine learning model inspired by the structure and function of the human brain, capable of learning complex patterns from data. | | Data scientist | A professional who uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from structured and unstructured data. | | Linear programming | A mathematical method for determining a way to achieve the best outcome in a mathematical model whose requirements are represented by linear relationships. | | Heuristics | Problem-solving methods that employ a practical approach, often used when finding an optimal solution is impossible or impractical. | | Digital twin | A virtual representation of a physical asset, process, or system, used for monitoring, analysis, and simulation to improve performance and predict issues. | | Monte Carlo | A broad class of computational algorithms that rely on repeated random sampling to obtain numerical results, often used for simulating complex systems. | | Chat GPT | A large language model developed by OpenAI, capable of generating human-like text in response to prompts. | | CoPilot | Refers to AI-powered coding assistants designed to help developers write code more efficiently by suggesting code snippets and completing lines of code. | | Data engineering | The process of designing, building, and maintaining systems and infrastructure for collecting, storing, processing, and analyzing data at scale. | | Software engineering | The application of systematic, disciplined, quantifiable approach to the development, operation, and maintenance of software. | | Business acumen | The ability to understand and effectively manage business operations, including strategic thinking, financial literacy, and market awareness. | | Safety stock | Extra inventory held to mitigate the risk of stockouts caused by uncertainties in supply and demand, such as fluctuations in lead times or demand variability. | | Lead time | The total time elapsed between the initiation and completion of a process; in supply chain, this is often the time from placing an order to receiving it. | | Supply chain visibility | The ability to track and monitor the movement of goods and information across the entire supply chain in real-time. | | Replenishment frequency | The rate at which inventory is reordered or restocked to maintain desired stock levels. | | Assortment rationalization | The process of analyzing and optimizing the range of products offered to customers, often involving the removal of slow-moving or unprofitable items. | | Risk pooling | A strategy where demand or supply risks are shared across multiple entities or locations to reduce overall variability and uncertainty. | | Postponement | A supply chain strategy where final product customization is delayed until the last possible moment, often in response to actual customer orders. | | Service levels | The probability of meeting customer demand from stock on hand, often expressed as a percentage. | | Inventory policies | Rules and guidelines that determine how much inventory to hold, when to reorder, and how to manage stock levels to balance costs and service. | | Forecasting | The process of predicting future events, typically based on historical data and analysis of trends. | | Rolling average | A statistical method for analyzing data points by creating a series of averages of different subsets of the full data set, smoothing out short-term fluctuations. | | Perishable products | Goods that have a limited shelf life and will spoil or become unusable if not sold or consumed within a specific period. | | Scrapping cost | The cost incurred when products must be discarded due to spoilage, damage, or obsolescence. | | Margin | The difference between the selling price of a product and its cost, representing the profit. | | Weather patterns | Long-term shifts in temperatures and other weather conditions, which can significantly impact demand for certain products. | | Counterfactual | An analysis that considers what would have happened under different circumstances, such as if a particular event had not occurred or a different decision had been made. | | Bayesian time series model | A statistical model that incorporates prior beliefs into the analysis of time-dependent data, allowing for probabilistic modeling of trends and seasonality. | | Stochastic forecasting | A forecasting method that produces a range of possible outcomes with associated probabilities, providing a clearer understanding of potential risks and uncertainties. | | Confidence Interval (CI) | A range of values that is likely to contain the true value of an unknown population parameter, providing a measure of uncertainty around an estimate. | | OTIF (On-Time, In-Full) | A key performance indicator in logistics and supply chain management that measures the percentage of orders delivered to customers on the agreed-upon date and with the complete quantity ordered. | | Stockouts | A situation where demand for a product exceeds the available inventory, resulting in lost sales and potential customer dissatisfaction. | | Aggregated performance measures | Metrics that combine data from multiple sources or entities to provide a high-level overview of performance, potentially masking underlying issues. | | Causal models | Models that aim to identify and quantify the cause-and-effect relationships between variables, explaining why certain outcomes occur. | | Product performance | Metrics used to evaluate how well a product is selling, its profitability, and its overall impact on the business. | | Sales volume | The total quantity of a product sold over a specific period. | | Gross margin | The profit a company makes after deducting the costs associated with making and selling its products, or the costs associated with providing its services. | | Uniqueness | A measure of how distinct a product is within an assortment, indicating the likelihood of customer substitution. | | Segment importance | The significance of a product to specific customer segments or strategic regions. | | Sales trend | The general direction in which sales are moving over time (e.g., upward, downward, or stable). | | Product lifecycle | The stages a product goes through from its introduction to the market to its eventual withdrawal, typically including introduction, growth, maturity, and decline. | | Sustainability | The practice of developing and implementing business strategies that minimize negative environmental and social impacts while maximizing economic benefits. | | Competitive advantage | A factor or combination of factors that allows a company to produce goods or services better or more cheaply than its rivals, resulting in an advantage over its competitors. | | Pareto frontier | In multi-dimensional analysis, it represents the set of optimal solutions where no objective can be improved without sacrificing another. | | Pairwise charts | Visualizations that display the relationship between two variables at a time, useful for exploring multidimensional data. | | SKU elimination | The process of removing Stock Keeping Units (SKUs) from a product assortment, typically based on performance metrics. | | Waterfall diagram | A type of chart that shows how an initial value is affected by a series of intermediate positive or negative values, leading to a final value. | | Price-response | The relationship between the price of a product and the demand for that product. | | Product cluster | A group of similar products that share common characteristics or exhibit similar demand patterns. | | Sigmoid | A mathematical function that produces an S-shaped curve, often used to model relationships that saturate or have a threshold effect, like price elasticity. | | Benchmarking | The process of comparing one's business processes and performance metrics to industry bests or best practices from other companies. | | Product assortment | The full set of products offered for sale by a retailer or manufacturer. | | LLM | Large Language Model, an AI model trained on a massive amount of text data to understand and generate human-like text. | | Category matcher | A tool or algorithm that identifies and groups products into predefined categories. | | Category extender | A function that expands or refines existing product categories, often by suggesting new related categories. | | Automation | The use of technology to perform tasks with minimal human intervention. | | Narrow automation | Automation focused on specific, repetitive tasks, often involving "bots" acting as human agents on computers. | | Broad automation | Automation of business processes and workflows, potentially including decision-making and advanced algorithms. | | Extended automation | Automation involving autonomous agents that perform complex reasoning and maintain state, acting more like human employees. | | Tacit knowledge | Knowledge that is difficult to articulate, transfer, or codify, often gained through experience and intuition. | | Flow state | A mental state in which a person performing an activity is fully immersed in a feeling of energized focus, full involvement, and enjoyment in the process of the activity. | | Chinese room argument | A philosophical thought experiment proposed by John Searle to challenge the idea that a machine can truly understand a language simply by manipulating symbols according to rules. | | Semantics | The meaning of words and sentences. | | Syntax | The rules governing the structure of sentences in a language. | | GenAI | Generative Artificial Intelligence, AI that can create new content, such as text, images, or code. | | ERP | Enterprise Resource Planning, a type of software system that organizations use to manage day-to-day business activities such as accounting, procurement, project management, risk management and compliance, and supply chain operations. | | FC | Forecast. | | NOK | No-OK, indicating a failure or rejection. | | OK | Indicates approval or success. | | N/A | Not Applicable, or Not Available. |