AI Mobile App Development in Riyadh, Saudi Arabia

Digital Banking / Fintech / Mobile-First

A Riyadh-based digital bank (SAMA-licensed — 450,000 customers, mobile-only bank targeting Saudi millennials and Gen Z, 180 employees) needed to differentiate through AI-powered mobile features. (1) Engagement: average session length 2.1 minutes — customers opening the app for balance check and leaving. Competitor: traditional Saudi banks adding mobile features — the digital bank needing to justify its mobile-only model through superior experience. Target: increasing session length to 5+ minutes through engaging AI features that provide financial value beyond basic banking. (2) Arabic voice: Saudi customers wanting to interact with their bank by voice — but existing voice assistants (Siri, Google) having limited Saudi banking vocabulary. Request: "كم رصيدي" (what's my balance), "حول ٥٠٠ ريال لأبوي" (transfer 500 riyals to my dad), "كم صرفت هالشهر على المطاعم" (how much did I spend on restaurants this month). Current: voice not available — all interactions through traditional UI. (3) Financial intelligence: Saudi millennials wanting spending insights — not just transaction lists but understanding of their financial behaviour. Current: basic transaction list with merchant names (often in English even for local merchants — "POS JARIR BOOKSTORE RIYADH" not meaningful to many users). Need: automatic categorisation, Arabic merchant name recognition, spending trends, and actionable savings recommendations. Cultural: understanding Saudi spending patterns — Ramadan spending peaks, Eid gift giving, and Saudi National Day celebrations. (4) Document processing: customers needing to submit documents — IBAN certificates, salary certificates, and government documents. Current: manual upload with back-office processing — 24-48 hour turnaround. Need: instant document processing — scan, extract data, and process within the app. Arabic documents: Saudi government documents in Arabic — AI needing to read Arabic text, extract relevant fields, and validate. (5) Fraud: increasing fraud targeting Saudi mobile banking — social engineering, SIM swap, and unauthorised transactions. Current: rule-based fraud detection catching 65% of fraud — missing sophisticated attacks. Need: AI-powered real-time fraud detection — analysing transaction patterns, device behaviour, and user behaviour to detect anomalies.

We built AI features integrated into the mobile banking app over 8 weeks. (1) Arabic voice banking: conversational finance. Voice AI: Saudi Arabic voice assistant within the banking app. ASR: Arabic speech recognition tuned for Saudi accents — Najdi, Hejazi, and mixed Saudi-English banking vocabulary. NLU: understanding banking intents in Saudi Arabic — balance enquiry, transfer, spending query, bill payment, and card management. Conversation: multi-turn dialogue — "حول ٥٠٠ ريال" → "لمين؟" (to whom?) → "لمحمد" → showing matching contacts → "محمد الحربي" → confirmation → transfer executed. Natural: voice assistant using conversational Saudi Arabic — not formal MSA. Responding like a helpful Saudi friend, not a robot. Security: voice biometric verification — user's voiceprint confirming identity for transactions. Two-factor: voice biometric + device trust for high-value transfers. On-device: voice processing using on-device ASR for privacy — audio converted to text on the phone, only text sent to the server. No voice recordings stored. Capabilities: balance check, transfers (between accounts, to beneficiaries, bill payments), spending queries ("how much did I spend on food last month?"), card management ("freeze my card"), and support ("I have a problem with my last transaction"). (2) Smart spending: AI financial intelligence. Categorisation: automatic transaction categorisation using NLP — merchant name analysis, MCC codes, and transaction patterns. Arabic merchants: recognising Saudi merchant names (حلويات سعد الدين = Saadeddin Pastry = Food & Dining) — mapping Arabic merchant names to categories. Saudi categories: culturally relevant categories — "مطاعم" (restaurants), "سوبرماركت" (supermarket), "ملابس" (clothing), "ترفيه" (entertainment), "حكومي" (government), and "صدقة/زكاة" (charity/zakat). Insights: weekly spending digest — "This week you spent SAR 1,240 on dining — 45% more than your 4-week average." Arabic: insights delivered in conversational Saudi Arabic with optional English. Trends: monthly spending trends with visual charts — showing spending patterns, category breakdown, and comparison to previous months. Ramadan: Ramadan-specific spending insights — tracking Ramadan spending patterns, iftar expenses, and zakat planning. Recommendations: actionable savings suggestions — "You have 3 active subscriptions totalling SAR 89/month. You haven't used [streaming service] in 6 weeks — consider cancelling to save SAR 40/month." Salary prediction: predicting end-of-month balance based on spending patterns — "At your current spending rate, you'll have SAR 2,340 remaining at month end" — helping users budget. (3) Document AI: instant document processing. Arabic OCR: on-device Arabic document scanning — camera captures document → AI extracts text and data. Documents: IBAN certificate (extracting bank name, IBAN, account holder — Arabic text), salary certificate (extracting employer, salary amount, date — mix of Arabic and numerals), national ID (extracting name, ID number, and expiry from Saudi national ID), and government letters (extracting reference numbers, dates, and content from government correspondence). On-device: document scanning and text extraction running entirely on-device using Core ML — document images never leave the phone. Processing: extracted data automatically populating relevant forms — customer scanning salary certificate → salary field auto-filled → employment verification initiated. Speed: from 24-48 hours (manual upload + back-office processing) to instant (AI extraction + automated processing). Accuracy: Arabic OCR achieving 96% character accuracy on typed Arabic documents, 91% on government stamps and seals. (4) Behavioural fraud detection: AI security. Model: real-time behavioural analysis — monitoring how the user interacts with the app (typing speed, scroll patterns, navigation paths) alongside transaction patterns. Device: device fingerprinting — identifying the user's normal device characteristics and detecting when the device profile changes (potential SIM swap or device takeover). Transaction: real-time transaction scoring — every transaction analysed for fraud indicators (unusual amount, new recipient, unusual time, unusual location). Biometric: behavioural biometric verification — the way the user holds the phone, types, and navigates being a continuous identity verification. Arabic social engineering: detecting social engineering indicators — unusual transaction patterns following phone calls (potential vishing), rapid successive transfers (potential coerced transfers), and transactions to known mule accounts. Response: risk-based authentication — low risk (proceed), medium risk (step-up authentication — face ID or voice verification), high risk (block and notify). Real-time: fraud scoring in under 100ms — user never experiences delay from fraud checking. (5) Personalised experience: AI-driven UX. Home screen: AI-curated home screen — showing the most relevant information and actions for each user at each time. Morning: showing daily balance and commute-related payments. Lunch: showing nearby restaurant spending. Evening: showing subscription reminders and daily summary. Ramadan: during Ramadan — iftar expense tracking, zakat calculator, and charity quick-transfer. Predictive: predicting what the user wants to do — if the user transfers to the same person every month, showing a quick-transfer button as the date approaches. Recommendation: financial product recommendations based on behaviour — savings account suggestion when large balance sits idle, investment suggestion when regular surplus exists.

Results over 8 months. Voice banking: 38% of customers using voice banking monthly. Voice sessions: 125,000 voice banking sessions/month (from zero). Arabic accuracy: Saudi Arabic speech recognition at 93% accuracy (from initial 88% — improving with Saudi usage data). Top voice commands: balance check (35%), transfers (28%), spending queries (22%), and card management (15%). Voice satisfaction: 4.4/5 user rating for voice banking feature. Smart spending: 72% of active users engaging with spending insights weekly. Session length: from 2.1 minutes to 6.8 minutes average (224% increase) — users spending more time exploring financial insights. Category accuracy: transaction categorisation at 94% accuracy — Arabic merchant names correctly identified. Savings: users engaging with recommendations saving average SAR 340/month more than non-engaging users. Ramadan: Ramadan spending insights used by 85% of active users during Ramadan — highest engagement feature. Document AI: 45,000 documents scanned monthly — reducing back-office document processing by 78%. Processing time: from 24-48 hours to instant for supported document types. Arabic OCR accuracy: 96% on typed documents — customers rarely needing to correct extracted data. IBAN certificates: most-scanned document (55%) — instant IBAN verification enabling faster onboarding. Fraud: fraud detection rate from 65% to 91% (40% improvement). False positives: from 3.2% to 1.1% — fewer legitimate transactions blocked. Fraud loss: SAR 2.8M annual fraud loss reduction. Customer impact: zero customers experiencing undetected fraud on AI-monitored transactions. Social engineering: 47 social engineering attempts detected and blocked (new capability — previously undetectable). App metrics: DAU (Daily Active Users) from 89,000 to 156,000 (75% increase). MAU from 310,000 to 420,000 (35% increase). App Store rating from 4.1 to 4.7. New customer acquisition: 85,000 new customers citing AI features in attribution — "I joined because of the voice banking" and "spending insights were the reason I switched." Revenue: increased engagement driving cross-sell — investment product adoption up 45%, savings account opening up 62%. Customer lifetime value: customers using AI features having 2.3x higher CLV than non-users. Financial: AI feature development SAR 1.2M (8-week project). Annual operation: SAR 380,000 (compute, model hosting, and maintenance). Revenue impact: SAR 8.5M from increased engagement, cross-sell, and reduced fraud. ROI: 7.1x first-year return.