Trust by Design — Launch Demir Asistan Without Breaking Brand Trust

A four-round, advanced brand-management simulation set inside Demir Bank A.Ş., a mid-size Istanbul retail bank (4.2M customers, 310 branches, net banking revenue ₺38.5 billion) whose decade-long franchise rests on a published Brand Trust Index of 74/100 — the highest in its peer set against a sector median of 61.

On 9 June 2026 the COO mandates the six-week launch of 'Demir Asistan', a generative-AI assistant inside the mobile app, ahead of the Q3 earnings call. Finance has banked ₺310 million in annual savings on a 55% AI-containment target; every 10 points of lost containment costs ≈ ₺56 million.

But the bank's own pre-test mirrors the published research (Lefkeli, Karataş & Gürhan-Canli, IJRM 2024): when customers learn they are sharing with AI rather than a human, brand trust drops sharply — driven by an inference that their data reaches a far larger audience and a resulting sense of exploitation — and the drop roughly doubles among the privacy-concerned 38% who hold 56% of deposits.

A naïve 'AI-first, covert, full-data' design cuts the Index 9 points (74→65). KVKK Law No. 6698 demands a lawful basis and explicit consent; a consumer-affairs journalist has filed an information request, and a covert launch breaks as a reputation story during the earnings call.

Playing the Brand Manager on a fixed ₺4.0 million build budget, you (1) diagnose WHY trust falls — separating audience-size inference and the sense of exploitation from generic 'AI is creepy'; (2) configure the trust architecture across disclosure, confidentiality assurance, anthropomorphism and data-use scope with a defensible KVKK consent flow; (3) set a segmented human-fallback rule under a viral 'tone-deaf bot' incident, balancing containment savings against trust; and (4) defend the numbers to the COO and a Board member and commit a monitoring + recovery plan if trust slips below the 70 floor.

The math rewards the three evidence-based mitigations (confidentiality assurance, anthropomorphism, data minimisation) and segmented fallback, and punishes the five classic errors — covert launch, cosmetic avatar, over-correction that misses savings, empty confidentiality promises that breach KVKK, and a one-size fallback that routes fraud and hardship to the bot.

Final KPIs track Brand Trust Index, AI-Containment %, Perceived-Exploitation score, and Annual Savings against the ₺310M target.

An advanced simulation for participants used to working with the main frameworks and trade-offs of the subject, designed for executive education and experienced professionals.