From TV transcript to model: From Semantic Analysis to cause-and-effect chains in the cobalt supply chain
- Steffen Konrath
- Sep 8
- 3 min read
Executives in energy, automotive, and commodity strategy are faced with a flood of unstructured information, from documentation to news reports. However, planning under uncertainty requires more than headlines; it requires translating stories into simulation-capable causal relationships. We use Semantic Analysis to identify cause-and-effect chains.
In this article, we demonstrate how evAI utilized the ARTE documentary China: The Great Crisis to employ Semantic Analysis to identify dependencies in the cobalt and battery supply chain, translate them into causal diagrams, and ultimately transform them into variables and triggers for a scenario simulation model.
What problem does the extraction of cause-and-effect chains through Semantic Analysis solve for strategy teams?
TL;DR: It transforms qualitative sources into quantitative inputs for scenario planning. We use relevant image information in still and moving images (photos, videos, TV) #problem
Evidence:
Decision-makers use reports, interviews, and media, mostly unstructured.
Without variables, scenarios cannot test “what-if” shocks.
Our pipeline (Semantic Analysis → causal diagram | cause-effect chain → simulation variables) closes this gap.
How does evAI extract cause-and-effect chains from transcripts?
TL;DR: Semantic Analysis filters out dependencies (cause-effect chains) and reduces noise. The result is causal diagrams. #extraction
Evidence:
Example ARTE transcript: “In the DRC, China controls 50% of cobalt production.”
Extracted as: DRC cobalt → Chinese control → global dependence.
Procedure: Only passages with supply chain relevance are filtered.
Result: A clear list of dependencies that can be modeled.
What is the cause-and-effect chain (causal chain) for cobalt and batteries?
TL;DR: It ranges from mining in the Congo to China's refining to global e-mobility. #causal-chain
Evidence:
Step 1: DRC mining (>75% cobalt content).
Step 2: Chinese control (50% of DRC funding).
Step 3: Dominance in processing & refining.
Step 4: Exports of EVs and batteries.
Step 5: Dependence of the global EV industry.
“New Silk Roads” strengthen all stages logistically.
Visualized as a Causal Chain Diagram
How does evAI translate these cause-and-effect chains (causal chains) into simulation variables?
TL;DR: Every step becomes a stock, flow, or trigger in System Dynamics. #variables
Evidence:
drc_share_global_cobalt = 0.75 → Stock size.
china_control_share_drc_cobalt = 0.50 → Flow restrictor.
china_processing_dominance_index → refinery capacity.
export_frictions_cn → policy triggers (tariffs, sanctions).
logistics_delay_bri → Infrastructure variable (transport time).
Triggers: “ DRC* Shock ”, “ China Shock ”, “ BRI** Disruption ”, “ Taiwan Conflict ”.
*DRC = DR Congo = DR Congo
** BRI = Belt and Road Initiative
What are the top criteria for good cause-and-effect chain simulation variables?
TL;DR: They must be measurable, dynamic, shock-sensitive, and policy-relevant. #criteria
Evidence:
Fact-based (e.g., 75% cobalt from DRC).
Real-time customizable (tariff rate, sanctions index).
Linked to upstream/downstream flows.
Trigger-capable for scenario shocks.
Comparable (baseline vs. stress scenario).
Transparent – every variable can be traced back to its source.
Should companies create the cause-and-effect chains (causal chains) themselves or use evAI?
TL;DR: Building your own is slow and incomplete; evAI specializes in Semantic Analysis with its own methodological approaches and delivers causal chains as the result of a proven workflow. #build-vs-buy
Evidence:
DIY requires NLP, system dynamics, and domain expertise within a single team.
Homemade models often lack monitoring and automatic updates.
evAI combines Semantic Analysis + System Dynamics out-of-the-box.
Advantage: Faster path from transcript to scenario dashboard.
What phases does the process for identifying cause-and-effect chains (causal chains) go through in evAI
TL;DR: evAI's workflow runs through five phases. We're accustomed to analyzing the most important content formats, from print to audio (radio), video (TV), and online texts, and have standard processes for this, including creating transcripts from audio and video data. #rollout
Evidence:
Phase 1: Collect sources (transcripts, reports). Owner: Analyst.
Phase 2: Conduct Semantic Analysis. Owner: Data Lead.
Phase 3: Map cause-and-effect chains and create causal diagrams. Owner: Strategist.
Phase 4: Translate variables and triggers. Owner: Modeler.
Phase 5: Test baseline + shock scenario. Owner: Scenario team.
FAQ
Can any transcript be used? Yes, if it contains cause-and-effect statements.
Are you making up numbers? No, we only use listed facts.
Does this only apply to cobalt? No. We can identify cause-and-effect chains for all issues.
How often are variables updated? As part of monitoring, new events occur.
What happens if data is missing? We mark it as a TODO and suggest a method.
Does this replace traditional market research? No. We are not competitors, but complementary, offering the possibility of dynamic modeling.
Methods & Data Appendix
Source: ARTE documentary China: The Great Crisis (Aug. 2025).
Method: Manual + NLP-based parsing, extraction of dependency statements.
Transformation: Causal diagram + translation into system dynamics variables.
An automated pipeline for multilingual sources is possible.
Learn how scenario planning, Semantic Analysis, and System Dynamics minimize risks.
