Top 10 AI Prompts and Use Cases and in the Financial Services Industry in Turkey

Türkiye's financial services sector is racing to turn AI from promise into practice: banks deploy chatbots, robo‑advisors and fraud detection that - according to a 2025 legal review - now scan roughly 40 million transactions a day and flag about 500 potential fraud cases, with one leading bank reporting a 98.7% fall in fraud losses over seven years Turkey AI 2025/26 legal trends report on AI in financial services.

That scale brings big gains - better customer experience, cost savings and sharper risk management - but also a shifting rulebook: BRSA and KVKK oversight, new data‑localization requirements and the transfer of DTO duties to the Cybersecurity Authority in 2025 mean governance, transparency and model validation are front‑and‑center.

Generative AI's operational lift in banking is mirrored globally EY analysis of generative AI in banking, and practical upskilling - like Nucamp's Nucamp AI Essentials for Work course syllabus - is becoming a strategic priority for Turkish FIs that want safe, compliant scale.

Metric	Value
Fraud loss reduction (leading bank)	98.7% over 7 years
Transactions analysed	~40 million / day
Potential fraud cases flagged	~500 / day
Data requirement	Customer data processed by AI must be stored in Türkiye

“We are using it for two reasons, to save on call centres, but also we want to give customers a better experience.” - Burak Arik, Maxitech / İşbank

The selection process filtered dozens of ideas down to ten prompts by insisting on three practical tests: clear ROI metrics, regulatory and operational fit for Türkiye, and fast path to scale.

That approach mirrors industry guidance - BCG urges banks to move beyond pilots and

anchor AI strategy in business strategy

to capture real returns (BCG 2025 report: AI Reckoning for Banks) - and reflects the hard wake‑up in the field where a recent analysis tracked 173 AI use cases across 50 banks yet found 70% of initiatives with nothing tangible to show (Analysis of 173 banking AI use cases showing low ROI).

Each candidate prompt was scored on measurable KPIs (cost‑benefit, payback, NPV/IRR and benchmarking), a lens drawn from practical ROI playbooks used by vendors and consultants (Practical ROI playbooks for AI in financial services), and then stress‑tested for fraud, customer‑service and compliance impact - areas where IBM and others show the biggest operational savings.

The result is a short, battle‑ready list: prompts that turn pilot slide decks into a single, auditable P&L line and can survive Türkiye's governance and data‑localization demands.

Design a company‑health analysis prompt for Akbank that extracts and continuously scores the bank's most telling KPIs: H1 net income TRY 24.852B, ROE 20.1%, revenue TRY 96.828B, fee‑income growth +60% YoY (fee‑to‑OpEx 100% quarterly), NPL ratio 3.4% with total provisions ≈ TRY 58B and a coverage ratio of 32.7%, total capital ratio 17.4% (CET1 12.6%), swap‑adjusted NIM ≈ 2% in Q2 versus margin guidance of 3–3.5%, TL loan growth 13% YTD (guidance >30% full year) and a loan‑to‑deposit ratio of 82% - all details sourced from Akbank's investor snapshot (Akbank investor relations snapshot (H1 financials)).

The prompt should flag divergences (for example, a NIM persistently below guidance or declining coverage against Stage‑2/3 balances) and produce concise, audit‑ready summaries for risk, funding and profitability scenarios; a useful touch is to pair this with transaction‑level anomaly detection and AI governance checks so outputs meet Türkiye's evolving validation rules and feed downstream use cases like AI‑driven credit decisioning or fraud alerts (AI-powered fraud detection in Turkish banks (financial services AI case study)).

One vivid signal: Akbank's fee income currently covers its operating costs - an instant red/green profitability light for modelled stress tests.

Metric	Value
H1 Net Income	TRY 24.852B
Revenue	TRY 96.828B
ROE	20.1%
Fee Income Growth	+60% YoY (fee/OpEx 100%)
NPL Ratio	3.4%
Total Provisions	≈ TRY 58B
Coverage Ratio (Stage 2+3)	32.7%
Total Capital Ratio	17.4%
Swap‑adjusted NIM (Q2)	~2%
TL Loan Growth (YTD)	13% (guidance >30% full year)

Turn Garanti BBVA's customer‑centric tooling into a single, auditable prompt that builds personalized retail and wealth plans: start with the My Status diagnosis (monthly saving capacity, months of cushion, debt health), ingest transaction rules from Smart Transactions (Pay Bill, Spend & Save, Regular Gold Saver) and private‑banking risk profiles, then ask the model to propose step‑by‑step savings, automated rules and rebalancing nudges that map to on‑app actions and human adviser handoffs - so a single prompt produces an action plan, next‑best offers and compliance flags that feed UGI's assistant and branch advisors.

Tie this to Garanti BBVA's existing roadmap approach and generative‑AI enhancements (UGI handled 61.7 million interactions in 2023), so prompts can keep context across chats and scale from retail nudges to bespoke wealth strategies; see Garanti BBVA's Financial Health tools for the behavioral building blocks and BBVA's financial coach write‑up for a proven app‑based coaching pattern Garanti BBVA Financial Health tools and BBVA financial coach app write-up.

Metric	Value
UGI smart assistant interactions (2023)	61.7 million
Total customers (2023)	25 million
Mobile banking customers (2023)	15 million

“Regardless of income, proper advice and follow‑up regarding money management is necessary.” - Işıl Akdemir Evlioğlu, Garanti BBVA

For Türkiye İş Bankası, a concise robo‑advisor memo prompt should turn portfolio policy into audit‑ready actions: instruct the model to rebalance on a periodic cadence (quarterly) or when allocations breach range‑bound thresholds, produce the exact trades to restore targets, and log tax, cost and cash‑management decisions for compliance reviews.

The prompt can borrow practical rules - eg, a 10% target with a 20% deviation band (rebalance when an asset falls below 8% or rises above 12%) - so the system avoids excessive churn while keeping the portfolio on track (a pattern highlighted in PureFinancial rebalancing best practices for portfolio management and Investopedia robo‑advisor portfolio checklist and guidance).

Include exception handling (withdrawals, distributions, tax‑loss harvesting opportunities) and require an auditable rationale for each trade so human overseers can validate model behavior under Türkiye's governance expectations; pairing this with broader AI controls also links back to local efforts to scale safe automation in banking (see Nucamp AI Essentials for Work case study: AI in Turkish financial services).

One vivid rule: treat a single 4% drift past a band as the trigger for one decisive sweep trade rather than a string of tiny orders - clear, explainable, and easier to validate.

For Yapı Kredi, a practical fraud‑detection rule creation prompt should do three things: spin up and manage the bank's 1,000+ machine‑learning‑backed rules, stitch in precalculated enriched scores and device‑level signals (including Trojan detection), and export auditable decisions that play nicely with KVKK‑governed data flows - so rules can scale across cards, DDA, open banking and crypto without exploding false positives.

The prompt must expect real‑time responses at millisecond speed (YKB's FICO Falcon implementation averages ~4 ms), prioritise blocking (about 98% of attempts are stopped before loss), and embed escalation lanes and explainability so a human reviewer can validate why 500 flagged cases a day were raised.

Tie rule creation to Yapı Kredi's trojan‑monitoring layer (the IHS‑backed project that protects internet/mobile channels) and to in‑house platforms like SIDE/SIDE‑C so detection models retrain quickly as attack patterns shift - one memorable test: insist the prompt auto‑generate a “drift summary” when daily volumes (40M txns) or the false‑positive ratio move beyond historical bounds.

For reference, see Yapı Kredi's award‑winning Falcon deployment and their trojan‑detection work for practical inputs and constraints.

Metric	Value
Fraud loss reduction (since Falcon)	98.7% over 7 years
Transactions processed	~40 million / day
Potential fraud cases flagged	~500 / day
Average response time	~4 milliseconds
False positive ratio (current)	~10:1 (improved from 30:1)
Card fraud basis points	<0.6
Blocked before damage	~98%
System success rate	99.98%

“The global economic slowdown and Turkiye's recent earthquake significantly increased transaction volumes and fraud risks.” - Halis Köseoğlu, fraud prevention director at Yapı Kredi

A production-ready prompt for QNB Finansbank should generate an auditable regulatory compliance checklist and the exact reporting artefacts regulators expect: a MASAK-aligned KYC/CIP flow, risk‑based CDD & EDD rules, a list of obliged‑party controls, SAR templates and the electronic payloads to submit to MASAK, plus executive dashboards for the board and the named Chief Compliance Officer.

Make the prompt parameterise Turkish thresholds (wire vs. high‑value vs. crypto), support remote e‑KYC per BDDK rules, and auto‑populate UBO checks (≥25%) and document retention rules (keep KYC files for 8 years) so every SAR and sanctions screen is traceable to Law No.

5549 and MASAK guidance - see practical KYC rules in Türkiye for detail (SanctionScanner: KYC requirements in Türkiye) and the 2025 crypto/threshold updates (2025 Turkey crypto KYC threshold updates (Akkas Law)).

Include routine independent testing, regex‑safe watchlist matching, and one vivid guardrail: auto‑raise a compliance incident when linked transfers push cumulative flows over the configured wire threshold in 24 hours, then produce the exact SAR packet and audit trail for review (AML compliance checklist for financial institutions (Feedzai)).

Item	Typical Turkish requirement / example
KYC record retention	Minimum 8 years
Beneficial ownership	Disclose ≥25% UBO
SAR destination	MASAK
Crypto KYC trigger (example)	₺15,000+ (2025 update)
Non‑face‑to‑face rules	e‑KYC allowed under BDDK/regulated conditions

A Findeks‑centred “credit decision explanation” prompt should stitch together real‑time Findeks feeds (scores, Risk and Cheque reports, the 36‑month cheque histories and QR‑cheque signals) into a single, auditable narrative that underwrites every accept/decline or pricing choice - think: a one‑page decision memo that says which bureau attributes, recent loan behaviour or cheque alerts moved the score and by how much, and then outputs the exact documents and API calls used for the decision (Findeks automation credit risk management with real-time feeds).

Pairing that with an XAI explanation layer (global and local feature attributions, counterfactuals and a human‑readable justification) closes the “why” gap regulators and customers care about (Explainable AI framework for interpretable credit scoring (SSRN paper)), while built‑in checks for bureau vs custom scores and override tracking map directly to fair‑lending review procedures and audit trails (FDIC guidance on fair lending implications of credit scoring systems).

One memorable test for production: if the prompt can turn a 36‑month cheque‑index swing into a two‑line customer rationale that a compliance officer and a loan applicant both understand, it has done its job - faster decisions, clearer appeals, and fewer surprise overrides.

Embedding an Intelligent Document Processing (IDP) pipeline on top of Türkiye's MERNIS/KPS backbone turns a paper‑heavy KYC bottleneck into a secure, auditable API flow: because KPS already lets 2,500 public and private users query up‑to‑date civil‑status and address records on demand, an IDP front end can ingest scanned IDs and forms, extract fields with layout‑aware OCR+LLM models and then validate them in real time against the official MERNIS feed - shaving hours from onboarding while keeping legal proof of source (Turkish Identity Information Sharing System (KPS) official documentation).

The engineering playbook is straightforward: phased integration, quality preprocessing and robust routing so the IDP hands only vetted extracts to downstream credit or fraud systems, with expected extraction accuracy in leading IDP setups at 95–100% and rapid ROI for high volumes (Intelligent Document Processing (IDP) guide (2025)).

Compliance is non‑negotiable - bilateral access agreements, careful handling of Republic of Türkiye identity numbers and the PDPA‑focused guidelines published in 2024 must shape data minimisation, retention and access logs - do this and the result is not just faster decisions but an auditable chain that replaces a mammoth filing room with a single, defensible API call (2024 guidelines on processing Turkish ID numbers).

Item	Value / Note
KPS / MERNIS users	~2,500 public & private entities
Peak enquiries (historical)	Expected >4 billion enquiries (2009 projection)
IDP extraction accuracy (leading)	95–100%
Legal framework	Civil Registration Services Law No. 5490; 2024 PDPA guidelines on Turkish ID numbers
Access requirement	Bilateral access agreement + legitimate interest statement

Design the Akbank Proactive Assistant prompt to do more than answer FAQs - make it a context‑aware, multilingual front line that reduces routine traffic and routes complex cases to humans with an auditable trail.

Start by training the script on Turkish patterns (loan types and interest questions account for roughly 16% of VA queries, loan calculations 14%, password resets 10%) and include channel logic for the platforms Turkish customers already use most: web and mobile (each 73% placement), with WhatsApp as a growing escalation path; see the CBOT report on chatbot adoption in Türkiye for these usage patterns (CBOT report on chatbot adoption in the Turkish banking sector).

Embed Tovie AI's best practices - a branded, competent persona, natural conversation, clear escalation rules and continuous learning - so the prompt outputs an explicit escalation packet (customer context, transcripts, risk flags, required documents) whenever the script detects fraud, regulatory triggers or a high‑value request (Tovie AI banking chatbot implementation best practices).

Add multilingual fallbacks per Vistatec guidance so customers can switch languages smoothly and keep compliance notes in Turkish for audit trails (Vistatec guide to multilingual fintech support).

One vivid test: the assistant should turn the 16% of loan/interest queries into an instant, two‑step path - answer plus next action (apply, schedule a call, or escalate) - so branch staff see only the cases that truly need human judgement.

Statistic	Value
Digital banking customers experiencing VAs	39%
Placement on websites	73%
Placement in mobile apps	73%
Banks offering VA on both web & app	45%
Common VA queries - loan types	16%
Institutions including voice tech	36%

A Borsa İstanbul market‑scan & investment‑insights prompt should act like a high‑frequency analyst for Turkish markets: ingest CBRT feeds, gross‑reserve movements, policy‑rate decisions, CDS spreads, net portfolio flows and headline inflation, then surface concise trade signals, portfolio tilts and regulator‑ready risk flags.

For Türkiye that means watching the narrow but decisive indicators - a $22.4bn fall in gross international reserves since early March and a mid‑April policy rate reset to ~46% are the kind of “single‑metric” triggers that should flip a model from neutral to defensive - while CDS spikes and a 13% mid‑March slide in equity markets map to tactical underweights or liquidity buffers.

The prompt should output (1) a one‑line market call (risk on/off) with provenance (which data moved and by how much), (2) a ranked list of 3–5 actionable ideas (eg, switch duration, hedge FX exposure, trim cyclical positions), and (3) audit trails for compliance - all tied to live CBRT releases and macro commentary so portfolio teams can explain moves to boards and clients (see the CBRT Monetary Policy Committee summary and the Capital Economics briefing on FX reserves for the exact signals to monitor).

A clever touch: include a “reserve‑room” flag that raises urgency when net reserves or swap‑adjusted buffers cross a preconfigured threshold, turning abstract macro risk into a single, memorable capital‑preservation decision for traders and wealth managers.

Metric	Recent value / note
Gross international reserves	Decreased by USD 22.4bn to USD 147.5bn (since Mar 7 → Apr 11)
Policy rate	Raised to ~46% (MPC decision, 17 Apr 2025)
Consumer inflation (CPI)	≈ 38.1% (March 2025)
5‑yr CDS premium	≈ 343 bps (mid‑April 2025)
Equity market move	~13% decline (mid‑March shock)
12‑month current account deficit	USD 12.8bn (Feb 2025)

Design a BDDK‑focused AI risk & ethics assessment prompt that reads like a regulator's checklist and a compliance playbook: require an AI inventory, a risk‑tier classification tied to Türkiye's emerging bill and the EU AI Act “Brussels effect,” automated bias and explainability tests, and an auditable human‑override decision for any high‑impact outcome; anchor outputs to local constraints such as the banking board's authority over remote ID rules and the hard data‑localization requirement that customer data processed by AI be stored in Türkiye so vendor contracts include on‑shore processing clauses (Turkey AI 2025–26 financial services legal trends).

The prompt should auto‑generate regulator‑ready packets (risk assessment, provenance of training data, model cards, test logs) and flag prohibited practices outlined by external frameworks - use the national tracker to map penalties, registration needs and enforcement expectations (AI Watch global regulatory tracker - Turkey).

A vivid acceptance test: when the prompt can produce a one‑page remediation plan plus a native‑language audit trail in under five minutes, it has moved AI governance from theory to boardroom practice.

Item	Note
AI‑specific regs (BDDK/BRSA)	None yet; existing frameworks applied adaptively
Authority example	BRSA board can set procedures for AI‑based remote ID actions
Data requirement	Customer data processed by AI must be stored in Türkiye

Next steps for Turkish financial institutions are practical and urgent: scale interpretable, niche fraud models that cut false positives and customer friction, harden governance and validation so boards can sign off on AI decisions, and invest in workforce skills to close the expertise gap quickly.

The data are clear - targeted models can boost detection accuracy (Stratyfy's work with FIS reported measurable gains) while industry research shows most firms are already seeing benefits but cite cost and skills as barriers (Stratyfy and FIS case study on reducing false-positive card fraud alerts, FIS and Oxford Economics Harmony Gap research on AI and fraud detection, Feedzai report: 90% of financial institutions use AI for fraud and financial crime).

Practically that means prioritising pilots that report simple ROI (fraud loss reduction, false‑positive rate, time‑to‑investigation), requiring explainability and weekly model refresh for fast‑moving scams, and pairing vendor tech with in‑house controls and skilled teams; upskilling can be actioned through short, work‑focused programs such as Nucamp's AI Essentials for Work to make prompt‑writing and model oversight routine (Nucamp AI Essentials for Work syllabus).

Do this well and the payoff is concrete: fewer blocked cards, faster decisions and a defensible audit trail when regulators ask “why.”

Metric	Value / Finding
FIS / Oxford finding	78% report AI improved fraud detection
Firms scaling AI	56% scaling or fully implementing
Feedzai survey	90% of FIs use AI for fraud/financial crime
Stratyfy + FIS result	Model improved fraud-detection accuracy (~51% in tests)

“The new model has helped banks [prioritize their reactions to fraud alerts] so that one less consumer is impacted for every potential fraud ...”