Barbara Feijen

Creative Agentforce Deployments: Belgium Ministry of Defense Human Resources Recruitment. The Creative Task. Agentforce is a cutting-edge solution designed to automate candidate-position matching and help job seekers navigate complex employment eligibility requirements. We wanted a system to be designed and implemented for the Belgium Ministry of Defense. The AI Challenge. The system had a few major bottlenecks: Unstructured CV Processing: Effectively extracting desired qualifications from diverse formats. Matching Candidates: Matching potential candidates with positions in military, civilian, and internship roles. Eligibility Rules: The intricate rules that apply to Belgian citizenship, residency, educational credentials, and criminal background check applicants. Guiding Candidates: Directing candidates through a multi-step application process. Creating Candidate Scores: Set up a scoring system for HR's review. MY VIEW On Intelligent Remixing Versus Original Creativity Agentforce is a prime case of ‘intelligent remixing’ as opposed to true creativity in my view. Such implementation indicates that AI processes more pattern recognition and recombination and less original creative thought. Here’s the evidence: What AI Really Does Well: Pattern Matching — CV parsing does not get out of bounds of the established frameworks of candidates' experiences. “3 years IT support,” for example, or “3 ans support informatique,” are keywords the AI notices and starts recognizing by analyzing patterns it has found in thousands of CVs it has been trained on. “The analysis of eligibility is not innovative – it is not about creativity.” Instead, it takes a methodical view of predetermined rules. Although the AI can conduct conversation, it always follows deterministic logic. Chat-Oriented Counsel; Agentforce constructs contexts from customer-service dialogues which makes the candidate interaction seem natural. To the extent that it inquires, for instance, “Do you currently live in Belgium?” once citizenship is detected, it takes learned conversational cues instead of coming up with new ways of communicating. Limitations of AI. In our implementation work, we found edge cases which the AI was not able to solve innovatively: Example 1: Security clearance issues of dual citizenship cases with Belgian and non-EU candidates. The AI wouldn’t be able to solve this in an independent way so it would need to explicitly program some of those cases. Example 2: A candidate with significant international military experience (e.g., NATO operations) received a low score as the phrase "foreign military service" was mistakenly perceived as irrelevant towards a Belgian post and could not be seen as relevant here. Scoring was manually adapted. Example 3: Many candidates who come from unusual backgrounds (career changers, adult learners) experience challenges as the AI fails to spot transferable skills that are not included in the more conventional route. The Value Proposition: Efficiency + Creative Insight. For Candidates: The platform is supremely capable of stripping friction out of the application process, directly matching applicants with guides. But it fails to offer career-inspiring insights or even personalized ones. It won’t suggest unconventional roles based on unusual skill combinations, but just patterns already existing. For HR: It captures data well, but does not identify the “diamond in the rough” candidates who fit well in other environments. At the end of the day and in fact, it's up to the human HR professionals to do that creative judgement. Conclusion: As a Tool, Not a Substitute for Human Insight. Agentforce exposes what can be an extraordinary capacity for AI to turn knowledge into action: writing documents and applying rules and enhancing communication on an incredible scale. But the architects of this system should be humans: Creative Logic Development: Coming up with the matching logic was a creative challenge. Anticipating Edge Cases: Putting the AI to the test in a way that it couldn't solve in isolation. Defining ‘Good Fit’: Set standards outside of just pattern match. As Agentforce does the recruiting but you need to conceptualize the architectural stuff, and that has to be what the Ministry needs, where the conversation goes, the scoring logic, the exception handling. Through intelligent remixing, Agentforce puts our creative vision into action. Understanding this distinction becomes instrumental for AI Solution Architects that will need to design AI systems that combine creative insights with human intervention to solve complex scenarios which involve contextual understanding and ethical dilemmas at the level of learning from past behavior.

Technology and processes evolve over time and eventually become outdated. Replacing a technology comes from the decision of when that tool no longer brings value to the businesse and customer experience. At this moment, I have two business cases in my project which are to replace a Rule-Based AI for Customer Support and the company’s website ChatBot. We detected that many customers start the chatbot and give up, either because they did not get the expected support (they want a human agent or a more intelligent virtual assistant) or because the chatbot really did not deliver the solution and the customer had to contact support by email. The value that these tools once delivered to the business is no longer there when there was an increase in the number of cases being created in the system by customers. The KPI of cases resolved by the chatbot has decreased, feedback is negative, and the “next best action” tool has become outdated and not always the “next best action” corresponded to the real scenario, even if most of the criteria pointed to the suggested action. To replace this Rule-Based AI, we are implementing an AI Agent that will understand the context, capture customer sentiment, and assist the support agent in resolving the case. For the website, we will implement an AI Chat so that customers have the experience of talking to a human, and this AI Agent has the autonomy to solve more complex problems, with escalation to a human support agent in cases of greater complexity. The plan is to keep the existing tools running in production until the development of the AIs is ready in the UAT environment for rigorous testing and feedback for adjustments and improvements. After approval, the AIs will be implemented in production and will have continuous monitoring through KPIs and sentiment data to maintain quality and compliance.

I’ve always believed that both business and people are in constant evolution, everything moves in cycles, and things can change overnight. That’s why continuous monitoring and feedback are crucial after any solution is implemented. In one of the projects where I had the pleasure of leading the functional stream, we integrated CRM, Power BI, and AI to deliver real-time insights to our sales representatives. We built a dashboard to monitor the accuracy of those insights, user adoption, and direct feedback from the sales team within the CRM. This required strong business alignment to compare AI performance with key KPIs (for which we defined specific thresholds) and to track data drift indicating shifts in customer consumption behaviour. Whenever anomalies were detected, we initiated a retraining or recalibration cycle to realign the AI model with business objectives.