Benedicta Olorungbade

PRACTICE: Daily Automated OEE Reports (Overall Equipment Effectiveness) PORTRAYED Efficiency: In this scenario, my organization automated the daily Overall Equipment Effectiveness (OEE) report. The automated generation of the daily Overall Equipment Effectiveness (OEE) reports was seen as a lean best practice. These reports were derived from data gotten directly from machines and were displayed and used as real-time performance metrics showing availability, performance, and quality across the various production lines. Stakeholders thought it was an example of the industry 4.0-driven efficiency which is known to be fast, data-driven, and standardized. Why It’s Actually Ineffective: Though this practice looked effective from a business excellence point of view, it actually masked significant inefficiencies like: Over dependent on surface level metrics: In depth analysis could not be gotten with surface level data. Even though OEE is a valuable KPI, it should not be used to drive primary performance decisions. For example, lines with high OEE scores may sometimes experience frequent stoppages or unforeseen quality issues that wouldn’t be flagged clearly in the summary metric. Instead of investigating underlying problems, operators would only be concerned about “protecting the numbers” Lack of Contextual Interpretation: Because reports were auto generated and shared, they lacked frontline insights and narrative context. This in turn gave room for doubts regarding the data provided in meetings because the integrity of the data was not trusted. Illusion of Continuous Improvement: The automation portrayed a false sense of control and progress because the data could not be depended on. Without accurate data for RCAs, stakeholders could not use the report to drive real improvement. The organization tracked problems but could not manage them, this act violates a core principle of business excellence. Business Excellence Lens: In a framework like lean six sigma, excellence requires not just the collection of data, but the usage of data for learning and innovation. The OEE reports that were automated lacked feedback loops and cross-functional engagements with other teams like shopfloor teams, thereby limiting organizational learning. Conclusion: What seems efficient will not always be effective. In this scenario, the automation of reporting replaced human insight with speed, optimizing for data delivery, but did not give data value. Business Excellence calls us to go deeper: using metrics as tools for engagement, improvement, and aligned decision-making, and not just dashboards for compliance’s sake.

The avoidance of harmful LLM hallucinations requires extensive investments when developing any medical LLMs before real-world deployment. Scenario: AI Hallucination During the National Agency for Food and Drug Administration and Control (NAFDAC)-Triggered Drug Recall Context A major pharmaceutical - consumer goods company in Lagos, Nigeria uses an LLM-powered AI chatbot to handle customer inquiries across its website and social media platforms. During a real product recall of a popular anti-malarial drug, triggered by NAFDAC’s discovery of unregistered batches in circulation, concerned customers begin flooding the chatbot with queries like: “Is batch DX41-002 part of the NAFDAC recall?” The chatbot which is powered by a general-purpose LLM, tries to give response based on its training and internal logic, but it lacks direct access to the official NAFDAC data. Because there is no accurate real-time information, the LLM confidently generates a hallucinated response saying: “No, batch DX41-002 is not affected by the recall and is safe to use.” But realistically, AX45-003 has been recalled and identified by NAFDAC as unregistered batches. Impact of the Hallucination Mistreatment: Consumers may ingest the drug thinking it is safe to use, and since not recommended by NAFDAC, may be unsafe for use and this may have an adverse effect of the safety of the patient. Compliance Issue: Misleading safety statements given by the Chatbot could trigger penalties by NAFDAC. Legal Liability: The company responsible for the manufacture of the drug could become susceptible to lawsuits and consumer protection action. Damage of Reputation: The company can incur bad reputation if the false information gets widely spread on social media platforms. Reason for the hallucination Root Cause Description Lack of data grounding The LLM has no access to live NAFDAC data or an internal batch registry Ambiguous user input Batch codes follow different formats and naming conventions which could lead to misinterpretation Sampling configuration The use of high top-p or temperature values give room for more speculative completions Training Uncertainty The LLM overuses reassuring language when uncertain Strategies to mitigate hallucinations 1. Prompt Engineering and Intent Restriction Give clear and specific prompts so that the chatbot can be guided towards desired outputs and intent restrictions. Disclaimers should also be included. E.g.: “Only respond if the batch code exists in the verified NAFDAC recall database. If missing or unsure, refer the user to an official support. Do not guess.” “Do note that this response is based on the data currently available. Please, kindly verify this information with NAFDAC or contact your healthcare service provider.” 2. Flow Logic and Escalation Paths Design the chatbot’s behaviour with layered safeguards and fallback logic e.g., If the user mentions “batch,” “recall,” “NAFDAC”, trigger a recall-check intent. If no match or confidence level is low, escalate to human agent or provide a static contact link. 3. System Architecture Enhancements By using Retrieval-Augmented Generation (RAG), connect the LLM to a live database of affected batches sourced from NAFDAC or internal compliance systems. Also, instead of fetching responses from scratch, the model can retrieve the relevant facts first, then summarize them. The below model configuration can be adopted. Restrict randomness by using lower top_p (e.g., 0.1–0.2) feature. Use temperature = 0 for deterministic outputs in safety-critical queries. Make use of only template-based responses and generate answers only as configured. 4. Monitoring, Feedback, and Recovery Process · By detecting potential misleading or overconfident statements, automatically flag risky responses. · Provision of user feedback controls by allowing users to report incorrect or unhelpful information. · Keep audit logs for all responses by maintaining traceability for legal and regulatory reviews. Conclusion In pharmaceutical manufacturing domains especially in heavy markets in Nigeria where compliance is a major subject, LLM hallucination is a great ordeal. Mitigating these risks means combining LLM safeguards with a high-level system architecture, data integration and human fallback channels.

Scenario: FMCG – Customer Care Team Responding to Consumer Complaints about Product Safety Compliance Scenario Relevance In the fast-moving consumer goods (FMCG) sector, especially in high-stakes areas like food, beverages, or personal care, there is a high level of expectation when it comes to compliance with the safety of products, labels, and advertising standards. Customer service teams are usually required to quickly respond to complaints about potential allergic reactions, contamination, or misleading labelling. Example of a risk: A customer sends in a message about experiencing an adverse reaction to a snack item that was sold by a company. A team member who is trying to manage the situation then drafts a response which reads: "Though our products are 100% safe, we’re confident this was just an isolated incident." Potential of Compliance Violation: “100% safe” could be seen as false assurance, because if it was 100% safe, there shouldn’t have been any issue occurrence with the item. This can be seen as a legally indefensible claim. “Isolated incident” can be misread as admission of liability. Meaning that one might begin to think there had been a prior knowledge or precedence. Making assumptions before an investigation is complete can be suggestive of verified language. Ways in which the AI can Identify and Flag Risks. Using a prompt + flow-based system, the AI Agent checks the draft and flags the patterns identified with compliance pitfalls Phrases like “100% safe,” “isolated incident,” “guarantee,” “harmless,” or any unqualified safety claims are seen as Risk Triggers. It is aware of the context being used by understanding the domain (e.g., product complaints vs. marketing claims) in order to avoid over-flagging harmless content. Intent Analysis is adopted, i.e., it identifies the difference between empathy and liability, e.g., “We’re sorry you experienced this”. Feedback Mechanism The AI assistant is integrated into the team’s communication platform (e.g., CRM or email client) and provides suggestion-based feedback: Inline Highlights: Risky phrases are underlined softly, like spellchecks. Rewrites suggestions like: “We take your concern seriously and are actively investigating. Our products meet stringent safety standards, but we welcome any information that can help us ensure quality.” Mini-Guidance Popups: Quick tooltips link to internal compliance playbooks for self-learning. Why This Works in FMCG This works in the FMCG space because it is fast Paced, it carries a high-risk context which combines speed and sensitivity where even small phrasing errors can result in legal fallout. It also empowers frontline staff, in the sense that it helps non-experts’ flag and correct issues without the need for unnecessary escalation. Summary Compliance Risk Prevented: Legal liabilities, regulatory breaches, erosion of the trust of the brand. AI’s Approach: Recognition of contextual pattern, lightweight coaching, and inline suggestions. Outcome: Messages that are empathetic, accurate, and compliant without slowing the team down.

Scenario: Detecting Cross-Functional Patterns in Escalations and Observations Departments like HR, Finance, Operations, and Tech Support in many organizations usually work independently, making it difficult to identify recurring patterns that may exist across functions. A Prompt + Flow-Based AI Solution can act as a link to uncover these hidden patterns, allowing leaders to address systemic issues that might go unnoticed. Ideal Solution: 1. Data Integration: The first step is to aggregate and normalize the various input sources like escalations, issues, or observations that are reported by HR, Finance, Operations, and Tech Support. Different formats for the inputs can include descriptions, numerical values, timestamps, categories, or severity ratings. The Inputs required include descriptions of issues, problems, or incidents, tags or labels associated with each report (e.g., “system downtime,” “employee dissatisfaction,” “payment delay”), date and time data for trend analysis, ratings or indicators that show the urgency of an issue. The AI solution will use natural language processing (NLP) and classification methods to organize and transform this unstructured data into a standardized format. 2. Pattern Recognition via Flow-Based AI: The AI can process these datasets and look for commonalities and sequences ranging across various departments. Using the flow-based approach, where information from each department is mapped against one another, the AI can discover patterns like Inter-departmental dependencies, escalation patterns, and temporal correlations. 3. AI Workflow: The AI can use NLP models to extract entities (e.g., "payment delay," "employee turnover," "server downtime") and map their relationships across departments. Based on timestamps, the AI can also identify connections across reports that may not be obvious. Once patterns are identified, the AI can trigger a set of insights that suggest an action plan. 4. Visualization & Reporting: After careful analysis and identification of the patterns, the AI can present the insights through a clear and user-friendly dashboard for departmental heads and leaders in teams: The insights generated would be given based on High-Level Summary, Root Cause Insight, Priority Action Recommendations: E.g., Finance should streamline approval processes during peak operational periods, and Predictive Alerting: Value and Usefulness of Insights: Proactive Issue Resolution: The AI’s ability to identify recurring cross-departmental patterns enables leaders to take preventive actions before issues escalate. Enhanced Functional Collaboration: By uncovering connections between different departments, the AI fosters better communication and collaboration. Departments can see how their actions impact others, breaking down silos and encouraging a more cohesive, unified approach to problem-solving and process improvement. Data-Driven Decision-Making: With the AI providing clear, actionable insights based on actual data, leaders can make informed decisions that are rooted in evidence Increased Operational Efficiency: Addressing systemic issues identified through AI-driven analysis can streamline operations across departments. Quick Issue Resolution: With patterns and dependencies identified early, departments can quickly respond to problems that have a broader organizational impact. Improved Resource Allocation: The AI can prioritize issues based on their impact across multiple departments. This allows leadership to allocate resources more effectively. 7. Anticipating Future Challenges The AI’s ability to analyze historical trends and identify correlations over time enables it to provide forecasts and early warnings about potential future issues. In conclusion, the insights provided by the AI solution allows leaders to address issues earlier, improve collaboration, make informed decisions, and streamline operations, all of which contribute to a more efficient, productive, and strategically aligned organization.

Scenario: Real-Time, Hyper-Localized Retail Campaign Activation Overview: One of the high-level advantages of the application of prompt + flow-based AI in the FMCG sector, is the automation of the creation of micro-campaigns specifically for individual retail stores. These campaigns are set in motion by hyper-local data such as weather conditions, in-store foot traffic, inventory levels, or nearby competitor promotions. Instead of deploying national-level promotions, this system uses real-time inputs to generate specifically tailored to suit in-store messaging, signage, staff scripts, and dynamic digital shelf content, all by using prompt orchestration and decision flow logic. Application of Prompt + Flow Design. 1. Flow Node Trigger Detection Activation of the flow is done by using incoming data from localized sources, each serving as a trigger that aligns with the specific campaign opportunity. Examples are POS inventory data e.g., surplus milk approaching expiry, competitor activity (e.g., online detection of nearby promotional offers) 2. Campaign Logic Decision Tree (Flow Layer) This uses a rules-based flow to evaluate: Which SKU category is affected? What is the promotion urgency? What assets are available? Which customer personas frequent the store? Example: "If stock > threshold and product = cold beverage and weather = hot → trigger hydration-themed promotion flow". 3. Prompt Layered – AI-Powered Content Generation. For every campaign scenario, prompts that are structured are sent to the AI to generate specific marketing assets across multiple formats, examples are highlighted below. (a) Shelf Talkers / In-Store Signage: E.g. “Write a catchy, persuasive shelf message, maximum of 20 words, promoting mango electrolyte drinks to gym-goers during a hot day in Lagos.” (b) Staff Sales Scripts: E.g. “Create a 10-second friendly pitch for store staff to recommend strawberry yogurt to mothers with children, focusing on health benefits.” All the contents generated must undergo quality-check through a human-in-the-loop review process before being deployed either in-store or digitally. Reasons why this approach is effective (Originality + High Impact) Originality Scalability Conversion Impact Operational Efficiency Conclusion: This scenario integrates intelligent, real-time data flows with prompt-based generative AI to transform every retail outlet into a responsive, adaptive marketing engine. A rare but game-changing innovation in the FMCG landscape.

Challenge: Demand Forecasting for New Product Launch in the FMCG Industry Problem Description In the FMCG industry, forecasting demand for new product launches can be quite challenging. Unlike established products, there may not be historical data for the proposed new product. Accurate forecasting is a must and if missed, may lead to excess inventory and inevitably cause wastage Traditional demand forecasting methods, which rely on time dependent data, struggle with the "cold start" problem. Because there is no history of past sales, machine learning models which require large, labelled datasets may not meet up to standard requirements and this may need to market loss. Proposed AI Approach: Flow + Prompt-based Design (Leveraging Generalist LLMs) Why This Approach Works Best 1. Handling Unstructured Contextual Data: A flow + prompt-based LLM solution can incorporate different unstructured information e.g., product reviews from similar items, similar products from competitors, feedback from retailers & distributors. A general purpose LLM can use this information as a basis to provide solution. 2. Guiding the AI using step by step prompts: The AI is giving step by step prompts in a structured format to enable analyse the problem. Using trends, it generates demand estimates which makes it’s results reliable and very easy to adjust whenever there is an inflow of new information. 3. Easy to Set Up and Use Quickly: No training of any new model is required for this, therefore making it easy to build faster. It can fit into already existing tools that the business already uses and allow for quick improvements, even without high technical requirement. 4. Originality: This gives room for a fresh and unique approach, because most current forecasting models in FMCG still rely on structured, supervised learning. The use of LLMs in interpreting marketing narratives, perceptions of the consumer and the demand history gives the business a new and unique way for demand prediction.

My Dream Paediatric Oncology AI Agent – Jenny_2030 Future Scenario: A children’s cancer centre. Jenny now supports paediatric oncology teams, young patients (ages 4–17), and their families throughout the chemotherapy process, prioritizing safety, empathy, education, and emotional resilience. Enhanced Capabilities of Jenny for Paediatric Chemotherapy: 1. Communication Jenny uses an interaction style that is based on a child’s age and emotional maturity considering different age ranges. 2. Family-Centered Care Coordination Jenny keeps parents informed with real-time updates on vitals, treatment milestones, and side-effect risks and reminder alerts for hospital visitations and drugs usage. 3. Emotional Monitoring Consideration Jenny analyses facial expressions such as tone, movement, and sleep pattern and also alerts paediatric psychologists and care teams when it detects any irregularities in the child’s emotions. 4. Safe Chemotherapy Customization Jenny continually references global paediatric oncology databases to ensure evidence-based, child-specific protocols are followed. Risk to Guard Against: Emotional Overattachment Children may develop strong attachments to Jenny as a comforting presence, which could lead to distress if Jenny is perceived as “withholding” bad news. Safeguards: Jenny is designed with psychological advisors to balance truth and empathy. She also supports difficult conversations alongside clinicians (humans), it does not replace humans.

Scenario: AI in Safety Monitoring vs. Employee Trust Ethical Dilemma: A large construction company implements AI-powered surveillance systems on-site to improve safety compliance. These systems use computer vision to detect when workers are not wearing PPE (Personal Protective Equipment), are in hazardous zones without supervision, or are showing signs of fatigue or distraction. To reduce incident rates and meet client expectations for zero-tolerance safety compliance, the AI begins automatically flagging employees for disciplinary actions and sometimes even recommending immediate suspension or removal from the site based on real-time behavior patterns. While this seems efficient and protective on paper, it causes serious friction: • Some flags are false positives: The AI misinterprets harmless behavior as violations (e.g., a worker adjusting gear briefly or gesturing). • Employees begin to feel constantly surveilled and distrusted, leading to lower morale and resistance to safety protocols. • There’s also concern the AI may disproportionately flag certain workers due to biased data (e.g., posture or movement differences related to age, body type, or disability). The ethical tension lies between ensuring client-mandated safety and liability protection, and respecting worker dignity, fairness, and privacy. Resolution Approach: 1. AI as an Observer, Not an Enforcer: Configure the AI system to flag potential violations for human review, not to autonomously penalize employees. A safety officer should validate AI observations before any action is taken. 2. Error Tolerance + Worker Feedback Loop: Build in a margin for human behavior e.g., don’t trigger alerts for brief, low-risk actions. Allow workers to review flagged incidents and provide context before decisions are made. This promotes fairness and engagement rather than fear. 3. Bias Testing and Dataset Audits: Routinely audit the AI’s training data and output to check for patterns of bias. For example, are certain body types or demographics flagged more often? Introduce regular retraining of models with diverse data to avoid systemic discrimination. 4. Transparent Ethics Charter and Training: Communicate clearly how the AI system works, what it monitors, what it doesn’t, and how decisions are reviewed. Educate both workers and managers so that AI is seen as a tool for protection, not punishment. AI Decision Boundaries in Construction The AI can do: • Detect PPE compliance and environmental hazards using real-time data. • Alert site managers to repeated safety breaches or high-risk patterns. • Recommend training refreshers for individuals or teams based on trends. • Support compliance reporting and documentation for clients or regulators. The AI should not do: • Make final decisions about employee discipline or dismissal. • Access biometric or emotional data (e.g., facial expressions) without consent. • Monitor private, off-task behavior unrelated to safety (e.g., social interaction). • Operate without regular bias testing or human accountability. Conclusion: In construction where safety is life-critical, AI can be a game-changer. But over-reliance on automation without empathy can backfire, damaging trust and culture. The right approach is to make AI a guardian, not a judge, one that empowers safety teams, protects workers, and ensures fair, transparent decisions.

What If AI Agents Worked as a Team? Use Case: AI-Driven Procure-to-Pay Process in SAP Scenario Overview: A top notch FMCG company, NutriChufa Group uses an AI-powered SAP Procure to Pay process system, where different AI agents, each with a distinct specialization, collaborate to manage the end-to-end procurement lifecycle of the company. Roles of the AI Agents and Handoff. 1. Agent A – RequestBot (Procurement Request Assistant) • Role: Handles employee procurement requests. • Capabilities: i. Converses with end-users to capture purchase needs. ii. Classifies the request (e.g., materials vs services). iii. Checks budgets and preferred suppliers in SAP. • Handoff: Sends structured purchase requisition (PR) data to the next agent. 2. Agent B – VendorCheckAI (Supplier Verification, Negotiator & Onboarding) • Role: Verifies vendors and checks compliance and risk. • Capabilities: i. Flags high-risk suppliers, according to the company’s audit requirements. ii. Checks for duplicates, if the vendor already exists in the system. iii. Selects the best fitting vendor from the requests for quotation (RFQ) documents submitted over emails/chats with suppliers. iv. Initiates onboarding if it is a completely new vendor and passes audit check. • Handoff: Updates vendor master data and passes verified PR to BuyerBot. 3. Agent C – BuyerBot (Purchase Order Creator) • Role: Creates POs and communicates with suppliers. • Capabilities: i. Converts PRs into POs. ii. Logs decisions with audit trails in SAP. • Handoff: Sends final PO to DocuVerifyBot and notifies suppliers. 4. Agent D – DocuVerifyBot (Document Compliance) • Role: Validates invoices and delivery notes. • Capabilities: i. Uses AI to read delivery documents and check if they match the purchase order and invoice ii. Detects anomalies (e.g., quantity mismatch, price variance, duplicate invoices). • Handoff: Approves matched invoices and sends them to FinanceBot. 5. Agent E – FinanceBot (Payment & Record Update) • Role: Handles payment execution and SAP ledger updates. • Capabilities: i. Triggers payment runs based on invoice approval. ii. Reconciles accounts. iii. Generates reports for audit and compliance. Challenges in Coordination + Conflict Resolution Mechanisms 1. Challenge: Timing Conflicts or Deadlocks • Example: BuyerBot can’t create a PO because VendorCheckAI hasn't completed its compliance check. • Solution: Introduce a central coordinator that ensures tasks run in the correct order and handles delays by triggering backups or involving a human when needed. 2. Challenge: Conflicting Data or Decisions • Example: DocuVerifyBot detects invoice inconsistency that RequestBot missed (e.g., price variance). • Solution: Set up a shared system where agents record their actions and reasons. If there's a conflict, they can review the logs and either resolve it together or pass it to a human for review. 3. Challenge: Difficulty in understanding why some AI decisions are made. • Solution: Use explainable AI tools that show clear reasons for each action, with details shown directly in SAP (e.g., “Price changed due to contract clause with supplier”).

I am using a scenario in healthcare where AI agents are used for reminding outpatients of their next appointment with a doctor. The reminder is based on the severity of their health condition. The problem, AI agent did not remind some patients with severe conditions about their next appointment. ASSIGNMENT OF RESPONSIBILITY - The main responsibility would be assigned to the designer of the AI agent. This is because there is a possibility that the model the AI agent used to recognize and update severity is not up to date. Newer, undefined health challenges may fall under the category of un-severe. - The next line of responsibility will fall on the human that needs to check the category the AI agent places reminder. If for some reason, the human neglects their responsibility, there is a high chance for the AI agent to assign wrong categories. Ultimately, the AI agent cannot be blamed for the mistakes they made.

For AI to earn trust in my team, the following should be considered. • Creation of awareness amongst team members regarding the capabilities of AI, explaining the "black box" idea so that team members are not completely oblivious of what it means. • Sharing past records of successful activity deliveries performed by the AI. It helps the team gain trust, when there is a track record of success. • Encouraging collaborations amongst the team with AI. • Team members should be trained in how to use AI, as this in turn boosts their confidence. This should also be championed by the lead, who can drive the importance and advantages of the use of AI for effective tasks delivery.

• Contextualize the situation and opt for the best fit: AI can check the context and pick the best option that applies to the scenario. • Explore alternatives by tracing historical data to pick the most suitable option: By running an analysis on past records, AI can pick the best alternative outcome. • Leverage stakeholder priority to determine what is the best option: AI can prioritize stakeholder interest and use that to give the best suitable outcome that aligns with human values and interests.