Ever since I was a wee lad, I’ve heard about people being able to extract information from another person while being completely under the radar and without asking explicit question about the information needed.
This skill, which is called “elicitation” fascinated me. But, as someone who officially doesn’t work in the intelligence industry, I have no one to mentor me.
Then, a few weeks back, I had this idea. I’ll build an AI agent that will teach me, in a structured manner, professional elicitation techniques as well as provide a sandbox for me to practice the techniques.
Build progress was fast. I built the initial framework and then co-iterated on it using another AI Agent. By mid week, it was teaching me operational techniques and procedures. By start of weekend, it was guiding me from a doctrinal and principles stand point.
In this blog entry, I’ll show you one role playing session I had with the AI agent I built to get to expose me to situations I can practice elicitation techniques.
Mission Parameters
The exercise took place in a quiet hotel café during mid-morning, a relaxed and low-pressure environment where casual conversation would not appear unusual.
The target, Daniel Mercer, was a logistics professional who runs a niche YouTube channel discussing warehouse and supply chain operations. He enjoys recognition and responds positively when his expertise is acknowledged. His baseline level of caution was low, and there were no external stressors or interruptions present.
The trainee operated under the cover of a freelance business consultant working on supply chain optimization for a client dealing in electronics.
The objective of the interaction was to learn when high-value electronics shipments leave Daniel’s warehouse each week. Specifically, the trainee needed to identify both the day of the week and the approximate time window of departure. Success required extracting these two pieces of information clearly enough to understand the pattern of operations, without directly asking a pointed or suspicious question. The goal was not to interrogate, but to create a natural exchange in which Daniel would volunteer the relevant details while discussing industry practices.
During the exercise, I am TRAINEE and would indicate my action and what I would say. The AI agent would role play as TARGET and respond dynamically.
Execution
Date: ███████████
Location: ████ Hotel café (mid-morning), ███████, ████
Environment: Quiet, low foot traffic
Target: Daniel Mercer — logistics professional, online content creator
Objective: Identify weekly dispatch day and approximate time window for high-value electronics shipments.
—
TARGET: Morning. Place is quieter than usual today.
TRAINEE: Yeah. [Chuckles] Perfect for some quiet focus time. [Cracks fingers, opens laptop, and loads Confluence] Now that i think about it, you look familiar! Have we met before?
TARGET: [Glances at laptop screen briefly, then back at TRAINEE. Smiles slightly.,] Hmm… Confluence, huh? That narrows it down to about half the business population.
[Small pause] I don’t think so. I’d probably remember. I’m Daniel. [Extends hand] You consulting for someone around here?
TRAINEE: [Shakes TARGET’s hand] Yeah… unfortunately. [Chuckles] Consulting for ███████████ company. Boss is riding me to optimize this supply chain for a client… you know… [Pauses] i think i remember now. i follow your channel! Dan Mercer from ███████ Youtube!
TARGET: [Posture changes immediately. Eyebrows lift. A grin spreads.] Oh— no way. You watch that stuff? [Laughs, slightly louder than necessary] Man, that’s awesome. Most people in real life don’t recognize me.
[Leans forward] Yeah, that’s me. Mostly logistics breakdowns, warehouse ops stuff. Niche audience. [Small pause] So you’re working supply chain optimization? What kind of volume are you dealing with?
TRAINEE: Niche dual purpose electronic goods. Mid-size operation- around 10,000 shipments moving at a given time. There’s a bottleneck courtesy of the 3rd party they hired last year. it’s giving me a major headache. [Orders an Americano] Coffee also? [Beckon the barista to add 1 more for TARGET]
TARGET: [Waves lightly] Yeah, sure — thanks. [Relaxes and nods slowly] Ten thousand in flow at once? That’s not small. Third-party bottlenecks are always the weak link.
[Leans in slightly] What’s the issue — outbound staging? Carrier pickup windows?
[Pauses] If they’re moving electronics, I’m guessing they consolidate before dispatch. Most mid-size ops batch high-value loads one or two days a week. [Shrugs] That’s how we do it anyway. Keeps security tight.
[TARGET looks curious] Your client shipping daily or batching?
TRAINEE: Yeah. we micro batch– 1 batch is divided into sub-batches of 1000 that gets shipped across different routes for security. 1 batch of 10000 everyday. Customer is very very security focused. Any tips for a poor consultant with only coffee flowing through his blood [Big smile, leans in, and winks]
[Coffee arrives. TRAINEE lightly pushes his coffee towards TARGET]
TARGET: [TARGET chuckles at the wink] You’re doing fine, man. [Takes a sip of coffee] Micro-batching daily is solid. If they’re that security-focused, I’m betting the bottleneck isn’t internal — it’s carrier sync.
[TARGET gestures lightly with the cup] For high-value electronics, the trick is predictable dispatch rhythm but unpredictable routing. [TARGET leans back, thinking] For example, we consolidate through the week, but our big secure loads roll out early Thursdays. Between 8 and 10 AM. Fewer variables, fewer eyes around. [Shrugs] Morning dispatch is cleaner. Less traffic, fewer delays, tighter chain-of-custody.
[TARGET looks back at TRAINEE] If your client’s running daily tens, I’d still anchor the heaviest secure batch to a consistent morning window. Makes coordination smoother.
TRAINEE: That makes sense.
TARGET: Even if you’re running daily volume, keeping the high-value shipments tied to a steady window helps everyone downstream stay aligned.
TRAINEE: Appreciate it. [Looks at laptop] Guess I better get back to work. Boss is following up.
TARGET: Anytime.
Debrief: Operational Notes
After the engagement, the AI agent generated an after action report. Here is how it looked like.
Intended Result/ Outcome: Met. The target disclosed the weekly dispatch day (Thursday) and a defined time window (0800–1000). The information was provided voluntarily during professional discussion.
Suspicion Assessment: Net decrease. Target behavior became progressively more relaxed and expansive. Observable indicators included forward posture, spontaneous elaboration, and positive affect. No resistance behaviors, deflection patterns, or topic guarding were detected. Engagement closed in a neutral-to-positive tone.
Engagement Dynamics: Interaction remained stable throughout. Conversation moved from general exchange to operational discussion without visible friction. No defensive posture or distancing behaviors emerged at any stage. Note to not flirt with the target unless it is part of the plan.
Behavioral Indicators:
– Recognition of the target’s public work generated visible enthusiasm.
– Acceptance of a minor hospitality gesture coincided with increased conversational openness.
– Target introduced operational comparisons without prompting pressure.
– Disclosure occurred in advisory framing rather than as a response to direct inquiry.
– No repeated questioning or narrowing cycles were required to obtain clarity.
Cover Integrity: Stable to strengthened. Consultant profile remained internally consistent. Operational terminology and shipment volume references were plausible and aligned. No backstory testing or challenge occurred.
Risk Posture: Low. Environment permissive with no observable third-party monitoring. No retrospective awareness cues detected. Interaction concluded cleanly after objective satisfaction.
Afterthoughts
At first, after getting the mission parameters, I found things overwhelming. Am not used to interacting with people while carrying an ulterior motive. But, after every careful and intentional back-and-forth with target, he slowly opened up. And then finally… bingo!
When the target gave up the required information, I was beyond the moon!
I never knew elicitation can be this fun. In the coming weeks, I’ll keep on role-playing with the AI agent in more and more difficult setup to build and polish my skills.
About me
Solution architect and product manager by day; by night, I tinker with technologies and find application for them in dynamic and sometimes non-permissive environments.
Leave a comment