AI Expense Analysis and Transaction Categorization for Mobile Apps

Implementing AI Expense Analysis and Transaction Categorization in a Mobile App

Manual transaction categorization — users do it the first week, then abandon it. Rule-based automation ("if memo contains 'LENTA' → 'Groceries'") works for major retailers but fails on "OOO PERSPEKTIVA" or "IP Ivanov A.V.". An ML categorizer with an LLM on top delivers a different quality level. We implement turnkey solutions — from data collection to deployment in App Store and Google Play. Our 5+ years of mobile development experience (40+ projects) lets us embed AI modules without sacrificing device performance.

How the Hybrid Approach Works

ML classifier (TF-IDF + LightGBM or distilBERT). Trained on historical transactions with labels. Inference < 10 ms, works offline, cost — zero after training. Accuracy on top-100 merchants: 95%+, on long-tail (small businesses): 60–70%.

LLM for unrecognized transactions. Transactions with low classifier confidence (< 0.7) are sent to an LLM. GPT-4o-mini, temperature=0, single prompt with categories and examples — response in 300–500 ms, accuracy on unusual names: 80–90%.

# Server-side categorization pipeline
async def categorize_transaction(transaction: Transaction) -> CategoryResult:
    # 1. Fast classifier
    ml_result = classifier.predict(transaction.description)

    if ml_result.confidence >= 0.75:
        return CategoryResult(
            category=ml_result.category,
            confidence=ml_result.confidence,
            method="ml_classifier"
        )

    # 2. LLM for uncertain predictions
    llm_category = await llm_categorize(
        description=transaction.description,
        amount=transaction.amount,
        merchant=transaction.merchant_name
    )
    return CategoryResult(
        category=llm_category,
        confidence=0.85,  # LLM more confident in complex cases
        method="llm_fallback"
    )

Hybrid approach: 85–90% of transactions handled by fast classifier (free), 10–15% by LLM. At 1,000 transactions per day per user, LLM query cost is negligible.

Comparison of Categorization Approaches

Hybrid wins overall: high accuracy across the spectrum at minimal cost.

Merchant Data Enrichment

Bank statement names are dirty data. "MAGNIT COSMETIC 0001" and "МАГНИТ КОСМЕТИК" are the same merchant. Normalization via merchant databases (Clearbit, Plaid Enrich, or custom mapping) significantly boosts classifier accuracy.

An additional signal is the MCC code (Merchant Category Code) that banks transmit with each transaction. MCC 5411 — grocery stores, MCC 5812 — restaurants. Using MCC as a classifier feature yields +5–10% accuracy.

AI Analysis of Spending Patterns

Categorization is step one. AI analysis on top of categorized data — that turns an app from a tracker into an advisor.

// iOS — Swift: LLM request for monthly expense analysis
func generateExpenseInsights(transactions: [CategorizedTransaction]) async -> [Insight] {
    let summary = transactions.groupBy(\.category)
        .mapValues { txs in (count: txs.count, total: txs.map(\.amount).reduce(0, +)) }
        .map { "\($0.key): \($0.value.total) RUB (\($0.value.count) transactions)" }
        .joined(separator: "\n")

    let prompt = """
    Analyze the user's monthly expenses and give 2-3 specific observations.
    Not generic advice — concrete patterns from the data.
    Expenses by category:\n\(summary)
    """

    let response = await llmClient.complete(prompt, maxTokens: 300, temperature: 0.4)
    return parseInsights(response)
}

The LLM sees: "Delivery food spending increased significantly compared to last month" and generates a concrete observation, not a generic "watch your food expenses".

Why Choose AI Categorization?

Rules become stale, and users don't want to spend time on manual entry. AI categorization with personalization boosts app retention by 20–30%. We guarantee classification accuracy of at least 90% on complete data after two weeks of training. Contact us for a consultation — we'll assess your project and propose the optimal architecture.

Training on User Corrections

Users correct misclassified categories — that's gold for retraining. Each correction is a new labeled example. After accumulating enough corrections (50–100 per user), we can fine-tune a personalized model or add user-specific rules:

// Android — saving user correction
fun saveUserCorrection(transactionId: String, correctedCategory: Category) {
    val correction = UserCorrection(
        transactionDescription = getTransaction(transactionId).description,
        merchantId = getTransaction(transactionId).merchantId,
        correctedCategory = correctedCategory,
        timestamp = System.currentTimeMillis()
    )
    localDatabase.saveCorrection(correction)
    // Sync to server for retraining
    syncService.scheduleCorrectionUpload(correction)
}

What's Included

• AI module architecture and integration with existing app
• ML classifier development (LightGBM or BERT) with training pipeline
• LLM wrapper for handling complex transactions
• User correction collection and retraining mechanism
• Integration with App Store and Google Play (via Firebase App Distribution)
• Code documentation, maintenance and retraining instructions
• One month of technical support post-release

Timeline Estimates

Rule + MCC classifier: 3–5 days. ML classifier with LLM fallback: 1–2 weeks. Full system with pattern analysis, insights, and correction learning: 2–4 weeks.

Order AI categorization development for your mobile app. Contact us — we'll prepare a commercial proposal tailored to your data and requirements.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.